mirror of
https://github.com/openSUSE/osc.git
synced 2024-11-10 06:46:15 +01:00
17a5242110
This reverts commit b190f27d00
.
Do not use "pathname" for this. Its sole purpose is to print out the
"correct" pathname to the user.
The code broke the Project's class update mechanism
6176 lines
227 KiB
Python
6176 lines
227 KiB
Python
# Copyright (C) 2006 Novell Inc. All rights reserved.
|
|
# This program is free software; it may be used, copied, modified
|
|
# and distributed under the terms of the GNU General Public Licence,
|
|
# either version 2, or version 3 (at your option).
|
|
|
|
__version__ = '0.132git'
|
|
|
|
# __store_version__ is to be incremented when the format of the working copy
|
|
# "store" changes in an incompatible way. Please add any needed migration
|
|
# functionality to check_store_version().
|
|
__store_version__ = '1.0'
|
|
|
|
import locale
|
|
import os
|
|
import os.path
|
|
import sys
|
|
import urllib2
|
|
from urllib import pathname2url, quote_plus, urlencode, unquote
|
|
from urlparse import urlsplit, urlunsplit
|
|
from cStringIO import StringIO
|
|
import shutil
|
|
import oscerr
|
|
import conf
|
|
import subprocess
|
|
import re
|
|
import socket
|
|
try:
|
|
from xml.etree import cElementTree as ET
|
|
except ImportError:
|
|
import cElementTree as ET
|
|
|
|
|
|
|
|
DISTURL_RE = re.compile(r"^(?P<bs>.*)://(?P<apiurl>.*?)/(?P<project>.*?)/(?P<repository>.*?)/(?P<revision>.*)-(?P<source>.*)$")
|
|
BUILDLOGURL_RE = re.compile(r"^(?P<apiurl>https?://.*?)/build/(?P<project>.*?)/(?P<repository>.*?)/(?P<arch>.*?)/(?P<package>.*?)/_log$")
|
|
BUFSIZE = 1024*1024
|
|
store = '.osc'
|
|
|
|
new_project_templ = """\
|
|
<project name="%(name)s">
|
|
|
|
<title></title> <!-- Short title of NewProject -->
|
|
<description></description>
|
|
<!-- This is for a longer description of the purpose of the project -->
|
|
|
|
<person role="maintainer" userid="%(user)s" />
|
|
<person role="bugowner" userid="%(user)s" />
|
|
<!-- remove this block to publish your packages on the mirrors -->
|
|
<publish>
|
|
<disable />
|
|
</publish>
|
|
<build>
|
|
<enable />
|
|
</build>
|
|
<debuginfo>
|
|
<disable />
|
|
</debuginfo>
|
|
|
|
<!-- remove this comment to enable one or more build targets
|
|
|
|
<repository name="openSUSE_Factory">
|
|
<path project="openSUSE:Factory" repository="standard" />
|
|
<arch>x86_64</arch>
|
|
<arch>i586</arch>
|
|
</repository>
|
|
<repository name="openSUSE_11.2">
|
|
<path project="openSUSE:11.2" repository="standard"/>
|
|
<arch>x86_64</arch>
|
|
<arch>i586</arch>
|
|
</repository>
|
|
<repository name="openSUSE_11.1">
|
|
<path project="openSUSE:11.1" repository="standard"/>
|
|
<arch>x86_64</arch>
|
|
<arch>i586</arch>
|
|
</repository>
|
|
<repository name="Fedora_12">
|
|
<path project="Fedora:12" repository="standard" />
|
|
<arch>x86_64</arch>
|
|
<arch>i586</arch>
|
|
</repository>
|
|
<repository name="SLE_11">
|
|
<path project="SUSE:SLE-11" repository="standard" />
|
|
<arch>x86_64</arch>
|
|
<arch>i586</arch>
|
|
</repository>
|
|
-->
|
|
|
|
</project>
|
|
"""
|
|
|
|
new_package_templ = """\
|
|
<package name="%(name)s">
|
|
|
|
<title></title> <!-- Title of package -->
|
|
|
|
<description></description> <!-- for long description -->
|
|
|
|
<!-- following roles are inherited from the parent project
|
|
<person role="maintainer" userid="%(user)s"/>
|
|
<person role="bugowner" userid="%(user)s"/>
|
|
-->
|
|
<!--
|
|
<url>PUT_UPSTREAM_URL_HERE</url>
|
|
-->
|
|
|
|
<!--
|
|
use one of the examples below to disable building of this package
|
|
on a certain architecture, in a certain repository,
|
|
or a combination thereof:
|
|
|
|
<disable arch="x86_64"/>
|
|
<disable repository="SUSE_SLE-10"/>
|
|
<disable repository="SUSE_SLE-10" arch="x86_64"/>
|
|
|
|
Possible sections where you can use the tags above:
|
|
<build>
|
|
</build>
|
|
<debuginfo>
|
|
</debuginfo>
|
|
<publish>
|
|
</publish>
|
|
<useforbuild>
|
|
</useforbuild>
|
|
|
|
Please have a look at:
|
|
http://en.opensuse.org/Restricted_formats
|
|
Packages containing formats listed there are NOT allowed to
|
|
be packaged in the openSUSE Buildservice and will be deleted!
|
|
|
|
-->
|
|
|
|
</package>
|
|
"""
|
|
|
|
new_attribute_templ = """\
|
|
<attributes>
|
|
<attribute namespace="" name="">
|
|
<value><value>
|
|
</attribute>
|
|
</attributes>
|
|
"""
|
|
|
|
new_user_template = """\
|
|
<person>
|
|
<login>%(user)s</login>
|
|
<email>PUT_EMAIL_ADDRESS_HERE</email>
|
|
<realname>PUT_REAL_NAME_HERE</realname>
|
|
<watchlist>
|
|
<project name="home:%(user)s"/>
|
|
</watchlist>
|
|
</person>
|
|
"""
|
|
|
|
info_templ = """\
|
|
Project name: %s
|
|
Package name: %s
|
|
Path: %s
|
|
API URL: %s
|
|
Source URL: %s
|
|
srcmd5: %s
|
|
Revision: %s
|
|
Link info: %s
|
|
"""
|
|
|
|
new_pattern_template = """\
|
|
<!-- See https://gitorious.org/opensuse/libzypp/blobs/master/zypp/parser/yum/schema/patterns.rng -->
|
|
|
|
<pattern>
|
|
</pattern>
|
|
"""
|
|
|
|
buildstatus_symbols = {'succeeded': '.',
|
|
'disabled': ' ',
|
|
'expansion error': 'U', # obsolete with OBS 2.0
|
|
'unresolvable': 'U',
|
|
'failed': 'F',
|
|
'broken': 'B',
|
|
'blocked': 'b',
|
|
'building': '%',
|
|
'finished': 'f',
|
|
'scheduled': 's',
|
|
'excluded': 'x',
|
|
'dispatching': 'd',
|
|
'signing': 'S',
|
|
}
|
|
|
|
|
|
# os.path.samefile is available only under Unix
|
|
def os_path_samefile(path1, path2):
|
|
try:
|
|
return os.path.samefile(path1, path2)
|
|
except:
|
|
return os.path.realpath(path1) == os.path.realpath(path2)
|
|
|
|
class File:
|
|
"""represent a file, including its metadata"""
|
|
def __init__(self, name, md5, size, mtime, skipped=False):
|
|
self.name = name
|
|
self.md5 = md5
|
|
self.size = size
|
|
self.mtime = mtime
|
|
self.skipped = skipped
|
|
def __repr__(self):
|
|
return self.name
|
|
def __str__(self):
|
|
return self.name
|
|
|
|
|
|
class Serviceinfo:
|
|
"""Source service content
|
|
"""
|
|
def __init__(self):
|
|
"""creates an empty serviceinfo instance"""
|
|
self.services = None
|
|
self.project = None
|
|
self.package = None
|
|
|
|
def read(self, serviceinfo_node, append=False):
|
|
"""read in the source services <services> element passed as
|
|
elementtree node.
|
|
"""
|
|
if serviceinfo_node == None:
|
|
return
|
|
if not append or self.services == None:
|
|
self.services = []
|
|
services = serviceinfo_node.findall('service')
|
|
|
|
for service in services:
|
|
name = service.get('name')
|
|
mode = service.get('mode', None)
|
|
data = { 'name' : name, 'mode' : '' }
|
|
if mode:
|
|
data['mode'] = mode
|
|
try:
|
|
for param in service.findall('param'):
|
|
option = param.get('name', None)
|
|
value = ""
|
|
if param.text:
|
|
value = param.text
|
|
name += " --" + option + " '" + value + "'"
|
|
data['command'] = name
|
|
self.services.append(data)
|
|
except:
|
|
msg = 'invalid service format:\n%s' % ET.tostring(serviceinfo_node)
|
|
raise oscerr.APIError(msg)
|
|
|
|
def getProjectGlobalServices(self, apiurl, project, package):
|
|
# get all project wide services in one file, we don't store it yet
|
|
u = makeurl(apiurl, ['source', project, package], query='cmd=getprojectservices')
|
|
try:
|
|
f = http_POST(u)
|
|
root = ET.parse(f).getroot()
|
|
self.read(root, True)
|
|
self.project = project
|
|
self.package = package
|
|
except urllib2.HTTPError, e:
|
|
if e.code != 403 and e.code != 400:
|
|
raise e
|
|
|
|
def addVerifyFile(self, serviceinfo_node, filename):
|
|
import hashlib
|
|
|
|
f = open(filename, 'r')
|
|
digest = hashlib.sha256(f.read()).hexdigest()
|
|
f.close()
|
|
|
|
r = serviceinfo_node
|
|
s = ET.Element( "service", name="verify_file" )
|
|
ET.SubElement(s, "param", name="file").text = filename
|
|
ET.SubElement(s, "param", name="verifier").text = "sha256"
|
|
ET.SubElement(s, "param", name="checksum").text = digest
|
|
|
|
r.append( s )
|
|
return r
|
|
|
|
|
|
def addDownloadUrl(self, serviceinfo_node, url_string):
|
|
from urlparse import urlparse
|
|
url = urlparse( url_string )
|
|
protocol = url.scheme
|
|
host = url.netloc
|
|
path = url.path
|
|
|
|
r = serviceinfo_node
|
|
s = ET.Element( "service", name="download_url" )
|
|
ET.SubElement(s, "param", name="protocol").text = protocol
|
|
ET.SubElement(s, "param", name="host").text = host
|
|
ET.SubElement(s, "param", name="path").text = path
|
|
|
|
r.append( s )
|
|
return r
|
|
|
|
def addGitUrl(self, serviceinfo_node, url_string):
|
|
r = serviceinfo_node
|
|
s = ET.Element( "service", name="tar_scm" )
|
|
ET.SubElement(s, "param", name="url").text = url_string
|
|
ET.SubElement(s, "param", name="scm").text = "git"
|
|
r.append( s )
|
|
return r
|
|
|
|
def addRecompressTar(self, serviceinfo_node):
|
|
r = serviceinfo_node
|
|
s = ET.Element( "service", name="recompress" )
|
|
ET.SubElement(s, "param", name="file").text = "*.tar"
|
|
ET.SubElement(s, "param", name="compression").text = "bz2"
|
|
r.append( s )
|
|
return r
|
|
|
|
def execute(self, dir, callmode = None, singleservice = None, verbose = None):
|
|
import tempfile
|
|
|
|
# cleanup existing generated files
|
|
for filename in os.listdir(dir):
|
|
if filename.startswith('_service:') or filename.startswith('_service_'):
|
|
os.unlink(os.path.join(dir, filename))
|
|
|
|
allservices = self.services or []
|
|
if singleservice and not singleservice in allservices:
|
|
# set array to the manual specified singleservice, if it is not part of _service file
|
|
data = { 'name' : singleservice, 'command' : singleservice, 'mode' : '' }
|
|
allservices = [data]
|
|
|
|
# set environment when using OBS 2.3 or later
|
|
if self.project != None:
|
|
os.putenv("OBS_SERVICE_PROJECT", self.project)
|
|
os.putenv("OBS_SERVICE_PACKAGE", self.package)
|
|
|
|
# recreate files
|
|
ret = 0
|
|
for service in allservices:
|
|
if singleservice and service['name'] != singleservice:
|
|
continue
|
|
if service['mode'] == "disabled" and callmode != "disabled":
|
|
continue
|
|
if service['mode'] != "disabled" and callmode == "disabled":
|
|
continue
|
|
if service['mode'] != "trylocal" and service['mode'] != "localonly" and callmode == "trylocal":
|
|
continue
|
|
call = service['command']
|
|
temp_dir = tempfile.mkdtemp()
|
|
name = call.split(None, 1)[0]
|
|
if not os.path.exists("/usr/lib/obs/service/"+name):
|
|
raise oscerr.PackageNotInstalled("obs-service-"+name)
|
|
c = "/usr/lib/obs/service/" + call + " --outdir " + temp_dir
|
|
if conf.config['verbose'] > 1 or verbose:
|
|
print "Run source service:", c
|
|
r = subprocess.call(c, shell=True)
|
|
if r != 0:
|
|
print "Aborting: service call failed: " + c
|
|
# FIXME: addDownloadUrlService calls si.execute after
|
|
# updating _services.
|
|
ret = r
|
|
|
|
if service['mode'] == "disabled" or service['mode'] == "trylocal" or service['mode'] == "localonly" or callmode == "local" or callmode == "trylocal":
|
|
for filename in os.listdir(temp_dir):
|
|
shutil.move( os.path.join(temp_dir, filename), os.path.join(dir, filename) )
|
|
else:
|
|
for filename in os.listdir(temp_dir):
|
|
shutil.move( os.path.join(temp_dir, filename), os.path.join(dir, "_service:"+name+":"+filename) )
|
|
os.rmdir(temp_dir)
|
|
|
|
return ret
|
|
|
|
class Linkinfo:
|
|
"""linkinfo metadata (which is part of the xml representing a directory
|
|
"""
|
|
def __init__(self):
|
|
"""creates an empty linkinfo instance"""
|
|
self.project = None
|
|
self.package = None
|
|
self.xsrcmd5 = None
|
|
self.lsrcmd5 = None
|
|
self.srcmd5 = None
|
|
self.error = None
|
|
self.rev = None
|
|
self.baserev = None
|
|
|
|
def read(self, linkinfo_node):
|
|
"""read in the linkinfo metadata from the <linkinfo> element passed as
|
|
elementtree node.
|
|
If the passed element is None, the method does nothing.
|
|
"""
|
|
if linkinfo_node == None:
|
|
return
|
|
self.project = linkinfo_node.get('project')
|
|
self.package = linkinfo_node.get('package')
|
|
self.xsrcmd5 = linkinfo_node.get('xsrcmd5')
|
|
self.lsrcmd5 = linkinfo_node.get('lsrcmd5')
|
|
self.srcmd5 = linkinfo_node.get('srcmd5')
|
|
self.error = linkinfo_node.get('error')
|
|
self.rev = linkinfo_node.get('rev')
|
|
self.baserev = linkinfo_node.get('baserev')
|
|
|
|
def islink(self):
|
|
"""returns True if the linkinfo is not empty, otherwise False"""
|
|
if self.xsrcmd5 or self.lsrcmd5:
|
|
return True
|
|
return False
|
|
|
|
def isexpanded(self):
|
|
"""returns True if the package is an expanded link"""
|
|
if self.lsrcmd5 and not self.xsrcmd5:
|
|
return True
|
|
return False
|
|
|
|
def haserror(self):
|
|
"""returns True if the link is in error state (could not be applied)"""
|
|
if self.error:
|
|
return True
|
|
return False
|
|
|
|
def __str__(self):
|
|
"""return an informatory string representation"""
|
|
if self.islink() and not self.isexpanded():
|
|
return 'project %s, package %s, xsrcmd5 %s, rev %s' \
|
|
% (self.project, self.package, self.xsrcmd5, self.rev)
|
|
elif self.islink() and self.isexpanded():
|
|
if self.haserror():
|
|
return 'broken link to project %s, package %s, srcmd5 %s, lsrcmd5 %s: %s' \
|
|
% (self.project, self.package, self.srcmd5, self.lsrcmd5, self.error)
|
|
else:
|
|
return 'expanded link to project %s, package %s, srcmd5 %s, lsrcmd5 %s' \
|
|
% (self.project, self.package, self.srcmd5, self.lsrcmd5)
|
|
else:
|
|
return 'None'
|
|
|
|
|
|
# http://effbot.org/zone/element-lib.htm#prettyprint
|
|
def xmlindent(elem, level=0):
|
|
i = "\n" + level*" "
|
|
if len(elem):
|
|
if not elem.text or not elem.text.strip():
|
|
elem.text = i + " "
|
|
for e in elem:
|
|
xmlindent(e, level+1)
|
|
if not e.tail or not e.tail.strip():
|
|
e.tail = i + " "
|
|
if not e.tail or not e.tail.strip():
|
|
e.tail = i
|
|
else:
|
|
if level and (not elem.tail or not elem.tail.strip()):
|
|
elem.tail = i
|
|
|
|
class Project:
|
|
"""represent a project directory, holding packages"""
|
|
REQ_STOREFILES = ('_project', '_apiurl')
|
|
if conf.config['do_package_tracking']:
|
|
REQ_STOREFILES += ('_packages',)
|
|
def __init__(self, dir, getPackageList=True, progress_obj=None, wc_check=True):
|
|
import fnmatch
|
|
self.dir = dir
|
|
self.absdir = os.path.abspath(dir)
|
|
self.progress_obj = progress_obj
|
|
|
|
self.name = store_read_project(self.dir)
|
|
self.apiurl = store_read_apiurl(self.dir, defaulturl=not wc_check)
|
|
|
|
dirty_files = []
|
|
if wc_check:
|
|
dirty_files = self.wc_check()
|
|
if dirty_files:
|
|
msg = 'Your working copy \'%s\' is in an inconsistent state.\n' \
|
|
'Please run \'osc repairwc %s\' and check the state\n' \
|
|
'of the working copy afterwards (via \'osc status %s\')' % (self.dir, self.dir, self.dir)
|
|
raise oscerr.WorkingCopyInconsistent(self.name, None, dirty_files, msg)
|
|
|
|
if getPackageList:
|
|
self.pacs_available = meta_get_packagelist(self.apiurl, self.name)
|
|
else:
|
|
self.pacs_available = []
|
|
|
|
if conf.config['do_package_tracking']:
|
|
self.pac_root = self.read_packages().getroot()
|
|
self.pacs_have = [ pac.get('name') for pac in self.pac_root.findall('package') ]
|
|
self.pacs_excluded = [ i for i in os.listdir(self.dir)
|
|
for j in conf.config['exclude_glob']
|
|
if fnmatch.fnmatch(i, j) ]
|
|
self.pacs_unvers = [ i for i in os.listdir(self.dir) if i not in self.pacs_have and i not in self.pacs_excluded ]
|
|
# store all broken packages (e.g. packages which where removed by a non-osc cmd)
|
|
# in the self.pacs_broken list
|
|
self.pacs_broken = []
|
|
for p in self.pacs_have:
|
|
if not os.path.isdir(os.path.join(self.absdir, p)):
|
|
# all states will be replaced with the '!'-state
|
|
# (except it is already marked as deleted ('D'-state))
|
|
self.pacs_broken.append(p)
|
|
else:
|
|
self.pacs_have = [ i for i in os.listdir(self.dir) if i in self.pacs_available ]
|
|
|
|
self.pacs_missing = [ i for i in self.pacs_available if i not in self.pacs_have ]
|
|
|
|
def wc_check(self):
|
|
global store
|
|
dirty_files = []
|
|
for fname in Project.REQ_STOREFILES:
|
|
if not os.path.exists(os.path.join(self.absdir, store, fname)):
|
|
dirty_files.append(fname)
|
|
return dirty_files
|
|
|
|
def wc_repair(self, apiurl=None):
|
|
global store
|
|
if not os.path.exists(os.path.join(self.dir, store, '_apiurl')) or apiurl:
|
|
if apiurl is None:
|
|
msg = 'cannot repair wc: the \'_apiurl\' file is missing but ' \
|
|
'no \'apiurl\' was passed to wc_repair'
|
|
# hmm should we raise oscerr.WrongArgs?
|
|
raise oscerr.WorkingCopyInconsistent(self.prjname, self.name, [], msg)
|
|
# sanity check
|
|
conf.parse_apisrv_url(None, apiurl)
|
|
store_write_apiurl(self.dir, apiurl)
|
|
self.apiurl = store_read_apiurl(self.dir, defaulturl=False)
|
|
|
|
def checkout_missing_pacs(self, expand_link=False):
|
|
for pac in self.pacs_missing:
|
|
|
|
if conf.config['do_package_tracking'] and pac in self.pacs_unvers:
|
|
# pac is not under version control but a local file/dir exists
|
|
msg = 'can\'t add package \'%s\': Object already exists' % pac
|
|
raise oscerr.PackageExists(self.name, pac, msg)
|
|
else:
|
|
print 'checking out new package %s' % pac
|
|
checkout_package(self.apiurl, self.name, pac, \
|
|
pathname=getTransActPath(os.path.join(self.dir, pac)), \
|
|
prj_obj=self, prj_dir=self.dir, expand_link=expand_link, progress_obj=self.progress_obj)
|
|
|
|
def status(self, pac):
|
|
exists = os.path.exists(os.path.join(self.absdir, pac))
|
|
st = self.get_state(pac)
|
|
if st is None and exists:
|
|
return '?'
|
|
elif st is None:
|
|
raise oscerr.OscIOError(None, 'osc: \'%s\' is not under version control' % pac)
|
|
elif st in ('A', ' ') and not exists:
|
|
return '!'
|
|
elif st == 'D' and not exists:
|
|
return 'D'
|
|
else:
|
|
return st
|
|
|
|
def get_status(self, *exclude_states):
|
|
res = []
|
|
for pac in self.pacs_have:
|
|
st = self.status(pac)
|
|
if not st in exclude_states:
|
|
res.append((st, pac))
|
|
if not '?' in exclude_states:
|
|
res.extend([('?', pac) for pac in self.pacs_unvers])
|
|
return res
|
|
|
|
def get_pacobj(self, pac, *pac_args, **pac_kwargs):
|
|
try:
|
|
st = self.status(pac)
|
|
if st in ('?', '!') or st == 'D' and not os.path.exists(os.path.join(self.dir, pac)):
|
|
return None
|
|
return Package(os.path.join(self.dir, pac), *pac_args, **pac_kwargs)
|
|
except oscerr.OscIOError:
|
|
return None
|
|
|
|
def set_state(self, pac, state):
|
|
node = self.get_package_node(pac)
|
|
if node == None:
|
|
self.new_package_entry(pac, state)
|
|
else:
|
|
node.set('state', state)
|
|
|
|
def get_package_node(self, pac):
|
|
for node in self.pac_root.findall('package'):
|
|
if pac == node.get('name'):
|
|
return node
|
|
return None
|
|
|
|
def del_package_node(self, pac):
|
|
for node in self.pac_root.findall('package'):
|
|
if pac == node.get('name'):
|
|
self.pac_root.remove(node)
|
|
|
|
def get_state(self, pac):
|
|
node = self.get_package_node(pac)
|
|
if node != None:
|
|
return node.get('state')
|
|
else:
|
|
return None
|
|
|
|
def new_package_entry(self, name, state):
|
|
ET.SubElement(self.pac_root, 'package', name=name, state=state)
|
|
|
|
def read_packages(self):
|
|
global store
|
|
|
|
packages_file = os.path.join(self.absdir, store, '_packages')
|
|
if os.path.isfile(packages_file) and os.path.getsize(packages_file):
|
|
return ET.parse(packages_file)
|
|
else:
|
|
# scan project for existing packages and migrate them
|
|
cur_pacs = []
|
|
for data in os.listdir(self.dir):
|
|
pac_dir = os.path.join(self.absdir, data)
|
|
# we cannot use self.pacs_available because we cannot guarantee that the package list
|
|
# was fetched from the server
|
|
if data in meta_get_packagelist(self.apiurl, self.name) and is_package_dir(pac_dir) \
|
|
and Package(pac_dir).name == data:
|
|
cur_pacs.append(ET.Element('package', name=data, state=' '))
|
|
store_write_initial_packages(self.absdir, self.name, cur_pacs)
|
|
return ET.parse(os.path.join(self.absdir, store, '_packages'))
|
|
|
|
def write_packages(self):
|
|
xmlindent(self.pac_root)
|
|
store_write_string(self.absdir, '_packages', ET.tostring(self.pac_root))
|
|
|
|
def addPackage(self, pac):
|
|
import fnmatch
|
|
for i in conf.config['exclude_glob']:
|
|
if fnmatch.fnmatch(pac, i):
|
|
msg = 'invalid package name: \'%s\' (see \'exclude_glob\' config option)' % pac
|
|
raise oscerr.OscIOError(None, msg)
|
|
state = self.get_state(pac)
|
|
if state == None or state == 'D':
|
|
self.new_package_entry(pac, 'A')
|
|
self.write_packages()
|
|
# sometimes the new pac doesn't exist in the list because
|
|
# it would take too much time to update all data structs regularly
|
|
if pac in self.pacs_unvers:
|
|
self.pacs_unvers.remove(pac)
|
|
else:
|
|
raise oscerr.PackageExists(self.name, pac, 'package \'%s\' is already under version control' % pac)
|
|
|
|
def delPackage(self, pac, force = False):
|
|
state = self.get_state(pac.name)
|
|
can_delete = True
|
|
if state == ' ' or state == 'D':
|
|
del_files = []
|
|
for filename in pac.filenamelist + pac.filenamelist_unvers:
|
|
filestate = pac.status(filename)
|
|
if filestate == 'M' or filestate == 'C' or \
|
|
filestate == 'A' or filestate == '?':
|
|
can_delete = False
|
|
else:
|
|
del_files.append(filename)
|
|
if can_delete or force:
|
|
for filename in del_files:
|
|
pac.delete_localfile(filename)
|
|
if pac.status(filename) != '?':
|
|
# this is not really necessary
|
|
pac.put_on_deletelist(filename)
|
|
print statfrmt('D', getTransActPath(os.path.join(pac.dir, filename)))
|
|
print statfrmt('D', getTransActPath(os.path.join(pac.dir, os.pardir, pac.name)))
|
|
pac.write_deletelist()
|
|
self.set_state(pac.name, 'D')
|
|
self.write_packages()
|
|
else:
|
|
print 'package \'%s\' has local modifications (see osc st for details)' % pac.name
|
|
elif state == 'A':
|
|
if force:
|
|
delete_dir(pac.absdir)
|
|
self.del_package_node(pac.name)
|
|
self.write_packages()
|
|
print statfrmt('D', pac.name)
|
|
else:
|
|
print 'package \'%s\' has local modifications (see osc st for details)' % pac.name
|
|
elif state == None:
|
|
print 'package is not under version control'
|
|
else:
|
|
print 'unsupported state'
|
|
|
|
def update(self, pacs = (), expand_link=False, unexpand_link=False, service_files=False):
|
|
if len(pacs):
|
|
for pac in pacs:
|
|
Package(os.path.join(self.dir, pac), progress_obj=self.progress_obj).update()
|
|
else:
|
|
# we need to make sure that the _packages file will be written (even if an exception
|
|
# occurs)
|
|
try:
|
|
# update complete project
|
|
# packages which no longer exists upstream
|
|
upstream_del = [ pac for pac in self.pacs_have if not pac in self.pacs_available and self.get_state(pac) != 'A']
|
|
|
|
for pac in upstream_del:
|
|
if self.status(pac) != '!' or pac in self.pacs_broken:
|
|
p = Package(os.path.join(self.dir, pac))
|
|
self.delPackage(p, force = True)
|
|
delete_storedir(p.storedir)
|
|
try:
|
|
os.rmdir(pac)
|
|
except:
|
|
pass
|
|
self.pac_root.remove(self.get_package_node(p.name))
|
|
self.pacs_have.remove(pac)
|
|
|
|
for pac in self.pacs_have:
|
|
state = self.get_state(pac)
|
|
if pac in self.pacs_broken:
|
|
if self.get_state(pac) != 'A':
|
|
checkout_package(self.apiurl, self.name, pac,
|
|
pathname=getTransActPath(os.path.join(self.dir, pac)), prj_obj=self, \
|
|
prj_dir=self.dir, expand_link=not unexpand_link, progress_obj=self.progress_obj)
|
|
elif state == ' ':
|
|
# do a simple update
|
|
p = Package(os.path.join(self.dir, pac), progress_obj=self.progress_obj)
|
|
rev = None
|
|
if expand_link and p.islink() and not p.isexpanded():
|
|
if p.haslinkerror():
|
|
try:
|
|
rev = show_upstream_xsrcmd5(p.apiurl, p.prjname, p.name, revision=p.rev)
|
|
except:
|
|
rev = show_upstream_xsrcmd5(p.apiurl, p.prjname, p.name, revision=p.rev, linkrev="base")
|
|
p.mark_frozen()
|
|
else:
|
|
rev = p.linkinfo.xsrcmd5
|
|
print 'Expanding to rev', rev
|
|
elif unexpand_link and p.islink() and p.isexpanded():
|
|
rev = p.linkinfo.lsrcmd5
|
|
print 'Unexpanding to rev', rev
|
|
elif p.islink() and p.isexpanded():
|
|
rev = p.latest_rev()
|
|
print 'Updating %s' % p.name
|
|
p.update(rev, service_files)
|
|
if unexpand_link:
|
|
p.unmark_frozen()
|
|
elif state == 'D':
|
|
# TODO: Package::update has to fixed to behave like svn does
|
|
if pac in self.pacs_broken:
|
|
checkout_package(self.apiurl, self.name, pac,
|
|
pathname=getTransActPath(os.path.join(self.dir, pac)), prj_obj=self, \
|
|
prj_dir=self.dir, expand_link=expand_link, progress_obj=self.progress_obj)
|
|
else:
|
|
Package(os.path.join(self.dir, pac), progress_obj=self.progress_obj).update()
|
|
elif state == 'A' and pac in self.pacs_available:
|
|
# file/dir called pac already exists and is under version control
|
|
msg = 'can\'t add package \'%s\': Object already exists' % pac
|
|
raise oscerr.PackageExists(self.name, pac, msg)
|
|
elif state == 'A':
|
|
# do nothing
|
|
pass
|
|
else:
|
|
print 'unexpected state.. package \'%s\'' % pac
|
|
|
|
self.checkout_missing_pacs(expand_link=not unexpand_link)
|
|
finally:
|
|
self.write_packages()
|
|
|
|
def commit(self, pacs = (), msg = '', files = {}, verbose = False, skip_local_service_run = False):
|
|
if len(pacs):
|
|
try:
|
|
for pac in pacs:
|
|
todo = []
|
|
if files.has_key(pac):
|
|
todo = files[pac]
|
|
state = self.get_state(pac)
|
|
if state == 'A':
|
|
self.commitNewPackage(pac, msg, todo, verbose=verbose, skip_local_service_run=skip_local_service_run)
|
|
elif state == 'D':
|
|
self.commitDelPackage(pac)
|
|
elif state == ' ':
|
|
# display the correct dir when sending the changes
|
|
if os_path_samefile(os.path.join(self.dir, pac), os.getcwd()):
|
|
p = Package('.')
|
|
else:
|
|
p = Package(os.path.join(self.dir, pac))
|
|
p.todo = todo
|
|
p.commit(msg, verbose=verbose, skip_local_service_run=skip_local_service_run)
|
|
elif pac in self.pacs_unvers and not is_package_dir(os.path.join(self.dir, pac)):
|
|
print 'osc: \'%s\' is not under version control' % pac
|
|
elif pac in self.pacs_broken:
|
|
print 'osc: \'%s\' package not found' % pac
|
|
elif state == None:
|
|
self.commitExtPackage(pac, msg, todo, verbose=verbose)
|
|
finally:
|
|
self.write_packages()
|
|
else:
|
|
# if we have packages marked as '!' we cannot commit
|
|
for pac in self.pacs_broken:
|
|
if self.get_state(pac) != 'D':
|
|
msg = 'commit failed: package \'%s\' is missing' % pac
|
|
raise oscerr.PackageMissing(self.name, pac, msg)
|
|
try:
|
|
for pac in self.pacs_have:
|
|
state = self.get_state(pac)
|
|
if state == ' ':
|
|
# do a simple commit
|
|
Package(os.path.join(self.dir, pac)).commit(msg, verbose=verbose, skip_local_service_run=skip_local_service_run)
|
|
elif state == 'D':
|
|
self.commitDelPackage(pac)
|
|
elif state == 'A':
|
|
self.commitNewPackage(pac, msg, verbose=verbose, skip_local_service_run=skip_local_service_run)
|
|
finally:
|
|
self.write_packages()
|
|
|
|
def commitNewPackage(self, pac, msg = '', files = [], verbose = False, skip_local_service_run = False):
|
|
"""creates and commits a new package if it does not exist on the server"""
|
|
if pac in self.pacs_available:
|
|
print 'package \'%s\' already exists' % pac
|
|
else:
|
|
user = conf.get_apiurl_usr(self.apiurl)
|
|
edit_meta(metatype='pkg',
|
|
path_args=(quote_plus(self.name), quote_plus(pac)),
|
|
template_args=({
|
|
'name': pac,
|
|
'user': user}),
|
|
apiurl=self.apiurl)
|
|
# display the correct dir when sending the changes
|
|
olddir = os.getcwd()
|
|
if os_path_samefile(os.path.join(self.dir, pac), os.curdir):
|
|
os.chdir(os.pardir)
|
|
p = Package(pac)
|
|
else:
|
|
p = Package(os.path.join(self.dir, pac))
|
|
p.todo = files
|
|
print statfrmt('Sending', os.path.normpath(p.dir))
|
|
p.commit(msg=msg, verbose=verbose, skip_local_service_run=skip_local_service_run)
|
|
self.set_state(pac, ' ')
|
|
os.chdir(olddir)
|
|
|
|
def commitDelPackage(self, pac):
|
|
"""deletes a package on the server and in the working copy"""
|
|
try:
|
|
# display the correct dir when sending the changes
|
|
if os_path_samefile(os.path.join(self.dir, pac), os.curdir):
|
|
pac_dir = pac
|
|
else:
|
|
pac_dir = os.path.join(self.dir, pac)
|
|
p = Package(os.path.join(self.dir, pac))
|
|
#print statfrmt('Deleting', os.path.normpath(os.path.join(p.dir, os.pardir, pac)))
|
|
delete_storedir(p.storedir)
|
|
try:
|
|
os.rmdir(p.dir)
|
|
except:
|
|
pass
|
|
except OSError:
|
|
pac_dir = os.path.join(self.dir, pac)
|
|
#print statfrmt('Deleting', getTransActPath(os.path.join(self.dir, pac)))
|
|
print statfrmt('Deleting', getTransActPath(pac_dir))
|
|
delete_package(self.apiurl, self.name, pac)
|
|
self.del_package_node(pac)
|
|
|
|
def commitExtPackage(self, pac, msg, files = [], verbose=False):
|
|
"""commits a package from an external project"""
|
|
if os_path_samefile(os.path.join(self.dir, pac), os.getcwd()):
|
|
pac_path = '.'
|
|
else:
|
|
pac_path = os.path.join(self.dir, pac)
|
|
|
|
project = store_read_project(pac_path)
|
|
package = store_read_package(pac_path)
|
|
apiurl = store_read_apiurl(pac_path, defaulturl=False)
|
|
if not meta_exists(metatype='pkg',
|
|
path_args=(quote_plus(project), quote_plus(package)),
|
|
template_args=None, create_new=False, apiurl=apiurl):
|
|
user = conf.get_apiurl_usr(self.apiurl)
|
|
edit_meta(metatype='pkg',
|
|
path_args=(quote_plus(project), quote_plus(package)),
|
|
template_args=({'name': pac, 'user': user}), apiurl=apiurl)
|
|
p = Package(pac_path)
|
|
p.todo = files
|
|
p.commit(msg=msg, verbose=verbose)
|
|
|
|
def __str__(self):
|
|
r = []
|
|
r.append('*****************************************************')
|
|
r.append('Project %s (dir=%s, absdir=%s)' % (self.name, self.dir, self.absdir))
|
|
r.append('have pacs:\n%s' % ', '.join(self.pacs_have))
|
|
r.append('missing pacs:\n%s' % ', '.join(self.pacs_missing))
|
|
r.append('*****************************************************')
|
|
return '\n'.join(r)
|
|
|
|
@staticmethod
|
|
def init_project(apiurl, dir, project, package_tracking=True, getPackageList=True, progress_obj=None, wc_check=True):
|
|
global store
|
|
|
|
if not os.path.exists(dir):
|
|
# use makedirs (checkout_no_colon config option might be enabled)
|
|
os.makedirs(dir)
|
|
elif not os.path.isdir(dir):
|
|
raise oscerr.OscIOError(None, 'error: \'%s\' is no directory' % dir)
|
|
if os.path.exists(os.path.join(dir, store)):
|
|
raise oscerr.OscIOError(None, 'error: \'%s\' is already an initialized osc working copy' % dir)
|
|
else:
|
|
os.mkdir(os.path.join(dir, store))
|
|
|
|
store_write_project(dir, project)
|
|
store_write_apiurl(dir, apiurl)
|
|
if package_tracking:
|
|
store_write_initial_packages(dir, project, [])
|
|
return Project(dir, getPackageList, progress_obj, wc_check)
|
|
|
|
|
|
class Package:
|
|
"""represent a package (its directory) and read/keep/write its metadata"""
|
|
|
|
# should _meta be a required file?
|
|
REQ_STOREFILES = ('_project', '_package', '_apiurl', '_files', '_osclib_version')
|
|
OPT_STOREFILES = ('_to_be_added', '_to_be_deleted', '_in_conflict', '_in_update',
|
|
'_in_commit', '_meta', '_meta_mode', '_frozenlink', '_pulled', '_linkrepair',
|
|
'_size_limit', '_commit_msg')
|
|
|
|
def __init__(self, workingdir, progress_obj=None, size_limit=None, wc_check=True):
|
|
global store
|
|
|
|
self.dir = workingdir
|
|
self.absdir = os.path.abspath(self.dir)
|
|
self.storedir = os.path.join(self.absdir, store)
|
|
self.progress_obj = progress_obj
|
|
self.size_limit = size_limit
|
|
if size_limit and size_limit == 0:
|
|
self.size_limit = None
|
|
|
|
check_store_version(self.dir)
|
|
|
|
self.prjname = store_read_project(self.dir)
|
|
self.name = store_read_package(self.dir)
|
|
self.apiurl = store_read_apiurl(self.dir, defaulturl=not wc_check)
|
|
|
|
self.update_datastructs()
|
|
dirty_files = []
|
|
if wc_check:
|
|
dirty_files = self.wc_check()
|
|
if dirty_files:
|
|
msg = 'Your working copy \'%s\' is in an inconsistent state.\n' \
|
|
'Please run \'osc repairwc %s\' (Note this might _remove_\n' \
|
|
'files from the .osc/ dir). Please check the state\n' \
|
|
'of the working copy afterwards (via \'osc status %s\')' % (self.dir, self.dir, self.dir)
|
|
raise oscerr.WorkingCopyInconsistent(self.prjname, self.name, dirty_files, msg)
|
|
|
|
self.todo = []
|
|
|
|
def wc_check(self):
|
|
dirty_files = []
|
|
for fname in self.filenamelist:
|
|
if not os.path.exists(os.path.join(self.storedir, fname)) and not fname in self.skipped:
|
|
dirty_files.append(fname)
|
|
for fname in Package.REQ_STOREFILES:
|
|
if not os.path.isfile(os.path.join(self.storedir, fname)):
|
|
dirty_files.append(fname)
|
|
for fname in os.listdir(self.storedir):
|
|
if fname in Package.REQ_STOREFILES or fname in Package.OPT_STOREFILES or \
|
|
fname.startswith('_build'):
|
|
continue
|
|
elif fname in self.filenamelist and fname in self.skipped:
|
|
dirty_files.append(fname)
|
|
elif not fname in self.filenamelist:
|
|
dirty_files.append(fname)
|
|
for fname in self.to_be_deleted[:]:
|
|
if not fname in self.filenamelist:
|
|
dirty_files.append(fname)
|
|
for fname in self.in_conflict[:]:
|
|
if not fname in self.filenamelist:
|
|
dirty_files.append(fname)
|
|
return dirty_files
|
|
|
|
def wc_repair(self, apiurl=None):
|
|
if not os.path.exists(os.path.join(self.storedir, '_apiurl')) or apiurl:
|
|
if apiurl is None:
|
|
msg = 'cannot repair wc: the \'_apiurl\' file is missing but ' \
|
|
'no \'apiurl\' was passed to wc_repair'
|
|
# hmm should we raise oscerr.WrongArgs?
|
|
raise oscerr.WorkingCopyInconsistent(self.prjname, self.name, [], msg)
|
|
# sanity check
|
|
conf.parse_apisrv_url(None, apiurl)
|
|
store_write_apiurl(self.dir, apiurl)
|
|
self.apiurl = store_read_apiurl(self.dir, defaulturl=False)
|
|
# all files which are present in the filelist have to exist in the storedir
|
|
for f in self.filelist:
|
|
# XXX: should we also check the md5?
|
|
if not os.path.exists(os.path.join(self.storedir, f.name)) and not f.name in self.skipped:
|
|
# if get_source_file fails we're screwed up...
|
|
get_source_file(self.apiurl, self.prjname, self.name, f.name,
|
|
targetfilename=os.path.join(self.storedir, f.name), revision=self.rev,
|
|
mtime=f.mtime)
|
|
for fname in os.listdir(self.storedir):
|
|
if fname in Package.REQ_STOREFILES or fname in Package.OPT_STOREFILES or \
|
|
fname.startswith('_build'):
|
|
continue
|
|
elif not fname in self.filenamelist or fname in self.skipped:
|
|
# this file does not belong to the storedir so remove it
|
|
os.unlink(os.path.join(self.storedir, fname))
|
|
for fname in self.to_be_deleted[:]:
|
|
if not fname in self.filenamelist:
|
|
self.to_be_deleted.remove(fname)
|
|
self.write_deletelist()
|
|
for fname in self.in_conflict[:]:
|
|
if not fname in self.filenamelist:
|
|
self.in_conflict.remove(fname)
|
|
self.write_conflictlist()
|
|
|
|
def info(self):
|
|
source_url = makeurl(self.apiurl, ['source', self.prjname, self.name])
|
|
r = info_templ % (self.prjname, self.name, self.absdir, self.apiurl, source_url, self.srcmd5, self.rev, self.linkinfo)
|
|
return r
|
|
|
|
def addfile(self, n):
|
|
if not os.path.exists(os.path.join(self.absdir, n)):
|
|
raise oscerr.OscIOError(None, 'error: file \'%s\' does not exist' % n)
|
|
if n in self.to_be_deleted:
|
|
self.to_be_deleted.remove(n)
|
|
# self.delete_storefile(n)
|
|
self.write_deletelist()
|
|
elif n in self.filenamelist or n in self.to_be_added:
|
|
raise oscerr.PackageFileConflict(self.prjname, self.name, n, 'osc: warning: \'%s\' is already under version control' % n)
|
|
# shutil.copyfile(os.path.join(self.dir, n), os.path.join(self.storedir, n))
|
|
if self.dir != '.':
|
|
pathname = os.path.join(self.dir, n)
|
|
else:
|
|
pathname = n
|
|
self.to_be_added.append(n)
|
|
self.write_addlist()
|
|
print statfrmt('A', pathname)
|
|
|
|
def delete_file(self, n, force=False):
|
|
"""deletes a file if possible and marks the file as deleted"""
|
|
state = '?'
|
|
try:
|
|
state = self.status(n)
|
|
except IOError, ioe:
|
|
if not force:
|
|
raise ioe
|
|
if state in ['?', 'A', 'M', 'R', 'C'] and not force:
|
|
return (False, state)
|
|
# special handling for skipped files: if file exists, simply delete it
|
|
if state == 'S':
|
|
exists = os.path.exists(os.path.join(self.dir, n))
|
|
self.delete_localfile(n)
|
|
return (exists, 'S')
|
|
|
|
self.delete_localfile(n)
|
|
was_added = n in self.to_be_added
|
|
if state in ('A', 'R') or state == '!' and was_added:
|
|
self.to_be_added.remove(n)
|
|
self.write_addlist()
|
|
elif state == 'C':
|
|
# don't remove "merge files" (*.r, *.mine...)
|
|
# that's why we don't use clear_from_conflictlist
|
|
self.in_conflict.remove(n)
|
|
self.write_conflictlist()
|
|
if not state in ('A', '?') and not (state == '!' and was_added):
|
|
self.put_on_deletelist(n)
|
|
self.write_deletelist()
|
|
return (True, state)
|
|
|
|
def delete_storefile(self, n):
|
|
try: os.unlink(os.path.join(self.storedir, n))
|
|
except: pass
|
|
|
|
def delete_localfile(self, n):
|
|
try: os.unlink(os.path.join(self.dir, n))
|
|
except: pass
|
|
|
|
def put_on_deletelist(self, n):
|
|
if n not in self.to_be_deleted:
|
|
self.to_be_deleted.append(n)
|
|
|
|
def put_on_conflictlist(self, n):
|
|
if n not in self.in_conflict:
|
|
self.in_conflict.append(n)
|
|
|
|
def put_on_addlist(self, n):
|
|
if n not in self.to_be_added:
|
|
self.to_be_added.append(n)
|
|
|
|
def clear_from_conflictlist(self, n):
|
|
"""delete an entry from the file, and remove the file if it would be empty"""
|
|
if n in self.in_conflict:
|
|
|
|
filename = os.path.join(self.dir, n)
|
|
storefilename = os.path.join(self.storedir, n)
|
|
myfilename = os.path.join(self.dir, n + '.mine')
|
|
if self.islinkrepair() or self.ispulled():
|
|
upfilename = os.path.join(self.dir, n + '.new')
|
|
else:
|
|
upfilename = os.path.join(self.dir, n + '.r' + self.rev)
|
|
|
|
try:
|
|
os.unlink(myfilename)
|
|
# the working copy may be updated, so the .r* ending may be obsolete...
|
|
# then we don't care
|
|
os.unlink(upfilename)
|
|
if self.islinkrepair() or self.ispulled():
|
|
os.unlink(os.path.join(self.dir, n + '.old'))
|
|
except:
|
|
pass
|
|
|
|
self.in_conflict.remove(n)
|
|
|
|
self.write_conflictlist()
|
|
|
|
# XXX: this isn't used at all
|
|
def write_meta_mode(self):
|
|
# XXX: the "elif" is somehow a contradiction (with current and the old implementation
|
|
# it's not possible to "leave" the metamode again) (except if you modify pac.meta
|
|
# which is really ugly:) )
|
|
if self.meta:
|
|
store_write_string(self.absdir, '_meta_mode', '')
|
|
elif self.ismetamode():
|
|
os.unlink(os.path.join(self.storedir, '_meta_mode'))
|
|
|
|
def write_sizelimit(self):
|
|
if self.size_limit and self.size_limit <= 0:
|
|
try:
|
|
os.unlink(os.path.join(self.storedir, '_size_limit'))
|
|
except:
|
|
pass
|
|
else:
|
|
store_write_string(self.absdir, '_size_limit', str(self.size_limit) + '\n')
|
|
|
|
def write_addlist(self):
|
|
self.__write_storelist('_to_be_added', self.to_be_added)
|
|
|
|
def write_deletelist(self):
|
|
self.__write_storelist('_to_be_deleted', self.to_be_deleted)
|
|
|
|
def delete_source_file(self, n):
|
|
"""delete local a source file"""
|
|
self.delete_localfile(n)
|
|
self.delete_storefile(n)
|
|
|
|
def delete_remote_source_file(self, n):
|
|
"""delete a remote source file (e.g. from the server)"""
|
|
query = 'rev=upload'
|
|
u = makeurl(self.apiurl, ['source', self.prjname, self.name, pathname2url(n)], query=query)
|
|
http_DELETE(u)
|
|
|
|
def put_source_file(self, n, copy_only=False):
|
|
cdir = os.path.join(self.storedir, '_in_commit')
|
|
try:
|
|
if not os.path.isdir(cdir):
|
|
os.mkdir(cdir)
|
|
query = 'rev=repository'
|
|
tmpfile = os.path.join(cdir, n)
|
|
shutil.copyfile(os.path.join(self.dir, n), tmpfile)
|
|
# escaping '+' in the URL path (note: not in the URL query string) is
|
|
# only a workaround for ruby on rails, which swallows it otherwise
|
|
if not copy_only:
|
|
u = makeurl(self.apiurl, ['source', self.prjname, self.name, pathname2url(n)], query=query)
|
|
http_PUT(u, file = os.path.join(self.dir, n))
|
|
os.rename(tmpfile, os.path.join(self.storedir, n))
|
|
finally:
|
|
if os.path.isdir(cdir):
|
|
shutil.rmtree(cdir)
|
|
if n in self.to_be_added:
|
|
self.to_be_added.remove(n)
|
|
|
|
def __generate_commitlist(self, todo_send):
|
|
root = ET.Element('directory')
|
|
keys = todo_send.keys()
|
|
keys.sort()
|
|
for i in keys:
|
|
ET.SubElement(root, 'entry', name=i, md5=todo_send[i])
|
|
return root
|
|
|
|
def __send_commitlog(self, msg, local_filelist):
|
|
"""send the commitlog and the local filelist to the server"""
|
|
query = {'cmd' : 'commitfilelist',
|
|
'user' : conf.get_apiurl_usr(self.apiurl),
|
|
'comment': msg}
|
|
if self.islink() and self.isexpanded():
|
|
query['keeplink'] = '1'
|
|
if conf.config['linkcontrol'] or self.isfrozen():
|
|
query['linkrev'] = self.linkinfo.srcmd5
|
|
if self.ispulled():
|
|
query['repairlink'] = '1'
|
|
query['linkrev'] = self.get_pulled_srcmd5()
|
|
if self.islinkrepair():
|
|
query['repairlink'] = '1'
|
|
u = makeurl(self.apiurl, ['source', self.prjname, self.name], query=query)
|
|
f = http_POST(u, data=ET.tostring(local_filelist))
|
|
root = ET.parse(f).getroot()
|
|
return root
|
|
|
|
def __get_todo_send(self, server_filelist):
|
|
"""parse todo from a previous __send_commitlog call"""
|
|
error = server_filelist.get('error')
|
|
if error is None:
|
|
return []
|
|
elif error != 'missing':
|
|
raise oscerr.PackageInternalError(self.prjname, self.name,
|
|
'__get_todo_send: unexpected \'error\' attr: \'%s\'' % error)
|
|
todo = []
|
|
for n in server_filelist.findall('entry'):
|
|
name = n.get('name')
|
|
if name is None:
|
|
raise oscerr.APIError('missing \'name\' attribute:\n%s\n' % ET.tostring(server_filelist))
|
|
todo.append(n.get('name'))
|
|
return todo
|
|
|
|
def commit(self, msg='', verbose=False, skip_local_service_run=False):
|
|
# commit only if the upstream revision is the same as the working copy's
|
|
upstream_rev = self.latest_rev()
|
|
if self.rev != upstream_rev:
|
|
raise oscerr.WorkingCopyOutdated((self.absdir, self.rev, upstream_rev))
|
|
|
|
if not skip_local_service_run:
|
|
r = self.run_source_services(mode="trylocal", verbose=verbose)
|
|
if r is not 0:
|
|
raise oscerr.ServiceRuntimeError(r)
|
|
|
|
if not self.todo:
|
|
self.todo = [i for i in self.to_be_added if not i in self.filenamelist] + self.filenamelist
|
|
|
|
pathn = getTransActPath(self.dir)
|
|
|
|
todo_send = {}
|
|
todo_delete = []
|
|
real_send = []
|
|
for filename in self.filenamelist + [i for i in self.to_be_added if not i in self.filenamelist]:
|
|
if filename.startswith('_service:') or filename.startswith('_service_'):
|
|
continue
|
|
st = self.status(filename)
|
|
if st == 'C':
|
|
print 'Please resolve all conflicts before committing using "osc resolved FILE"!'
|
|
return 1
|
|
elif filename in self.todo:
|
|
if st in ('A', 'R', 'M'):
|
|
todo_send[filename] = dgst(os.path.join(self.absdir, filename))
|
|
real_send.append(filename)
|
|
print statfrmt('Sending', os.path.join(pathn, filename))
|
|
elif st in (' ', '!', 'S'):
|
|
if st == '!' and filename in self.to_be_added:
|
|
print 'file \'%s\' is marked as \'A\' but does not exist' % filename
|
|
return 1
|
|
f = self.findfilebyname(filename)
|
|
if f is None:
|
|
raise oscerr.PackageInternalError(self.prjname, self.name,
|
|
'error: file \'%s\' with state \'%s\' is not known by meta' \
|
|
% (filename, st))
|
|
todo_send[filename] = f.md5
|
|
elif st == 'D':
|
|
todo_delete.append(filename)
|
|
print statfrmt('Deleting', os.path.join(pathn, filename))
|
|
elif st in ('R', 'M', 'D', ' ', '!', 'S'):
|
|
# ignore missing new file (it's not part of the current commit)
|
|
if st == '!' and filename in self.to_be_added:
|
|
continue
|
|
f = self.findfilebyname(filename)
|
|
if f is None:
|
|
raise oscerr.PackageInternalError(self.prjname, self.name,
|
|
'error: file \'%s\' with state \'%s\' is not known by meta' \
|
|
% (filename, st))
|
|
todo_send[filename] = f.md5
|
|
|
|
if not real_send and not todo_delete and not self.islinkrepair() and not self.ispulled():
|
|
print 'nothing to do for package %s' % self.name
|
|
return 1
|
|
|
|
print 'Transmitting file data ',
|
|
filelist = self.__generate_commitlist(todo_send)
|
|
sfilelist = self.__send_commitlog(msg, filelist)
|
|
send = self.__get_todo_send(sfilelist)
|
|
real_send = [i for i in real_send if not i in send]
|
|
# abort after 3 tries
|
|
tries = 3
|
|
while len(send) and tries:
|
|
for filename in send[:]:
|
|
sys.stdout.write('.')
|
|
sys.stdout.flush()
|
|
self.put_source_file(filename)
|
|
send.remove(filename)
|
|
tries -= 1
|
|
sfilelist = self.__send_commitlog(msg, filelist)
|
|
send = self.__get_todo_send(sfilelist)
|
|
if len(send):
|
|
raise oscerr.PackageInternalError(self.prjname, self.name,
|
|
'server does not accept filelist:\n%s\nmissing:\n%s\n' \
|
|
% (ET.tostring(filelist), ET.tostring(sfilelist)))
|
|
# these files already exist on the server
|
|
# just copy them into the storedir
|
|
for filename in real_send:
|
|
self.put_source_file(filename, copy_only=True)
|
|
|
|
self.rev = sfilelist.get('rev')
|
|
print
|
|
print 'Committed revision %s.' % self.rev
|
|
|
|
if self.ispulled():
|
|
os.unlink(os.path.join(self.storedir, '_pulled'))
|
|
if self.islinkrepair():
|
|
os.unlink(os.path.join(self.storedir, '_linkrepair'))
|
|
self.linkrepair = False
|
|
# XXX: mark package as invalid?
|
|
print 'The source link has been repaired. This directory can now be removed.'
|
|
|
|
if self.islink() and self.isexpanded():
|
|
li = Linkinfo()
|
|
li.read(sfilelist.find('linkinfo'))
|
|
if li.xsrcmd5 is None:
|
|
raise oscerr.APIError('linkinfo has no xsrcmd5 attr:\n%s\n' % ET.tostring(sfilelist))
|
|
sfilelist = ET.fromstring(self.get_files_meta(revision=li.xsrcmd5))
|
|
for i in sfilelist.findall('entry'):
|
|
if i.get('name') in self.skipped:
|
|
i.set('skipped', 'true')
|
|
store_write_string(self.absdir, '_files', ET.tostring(sfilelist) + '\n')
|
|
for filename in todo_delete:
|
|
self.to_be_deleted.remove(filename)
|
|
self.delete_storefile(filename)
|
|
self.write_deletelist()
|
|
self.write_addlist()
|
|
self.update_datastructs()
|
|
|
|
print_request_list(self.apiurl, self.prjname, self.name)
|
|
|
|
# FIXME: add testcases for this codepath
|
|
sinfo = sfilelist.find('serviceinfo')
|
|
if sinfo is not None:
|
|
print 'Waiting for server side source service run'
|
|
u = makeurl(self.apiurl, ['source', self.prjname, self.name])
|
|
while sinfo is not None and sinfo.get('code') == 'running':
|
|
sys.stdout.write('.')
|
|
sys.stdout.flush()
|
|
# does it make sense to add some delay?
|
|
sfilelist = ET.fromstring(http_GET(u).read())
|
|
# if sinfo is None another commit might have occured in the "meantime"
|
|
sinfo = sfilelist.find('serviceinfo')
|
|
print ''
|
|
rev=self.latest_rev()
|
|
self.update(rev=rev)
|
|
|
|
def __write_storelist(self, name, data):
|
|
if len(data) == 0:
|
|
try:
|
|
os.unlink(os.path.join(self.storedir, name))
|
|
except:
|
|
pass
|
|
else:
|
|
store_write_string(self.absdir, name, '%s\n' % '\n'.join(data))
|
|
|
|
def write_conflictlist(self):
|
|
self.__write_storelist('_in_conflict', self.in_conflict)
|
|
|
|
def updatefile(self, n, revision, mtime=None):
|
|
filename = os.path.join(self.dir, n)
|
|
storefilename = os.path.join(self.storedir, n)
|
|
origfile_tmp = os.path.join(self.storedir, '_in_update', '%s.copy' % n)
|
|
origfile = os.path.join(self.storedir, '_in_update', n)
|
|
if os.path.isfile(filename):
|
|
shutil.copyfile(filename, origfile_tmp)
|
|
os.rename(origfile_tmp, origfile)
|
|
else:
|
|
origfile = None
|
|
|
|
get_source_file(self.apiurl, self.prjname, self.name, n, targetfilename=storefilename,
|
|
revision=revision, progress_obj=self.progress_obj, mtime=mtime, meta=self.meta)
|
|
|
|
shutil.copyfile(storefilename, filename)
|
|
if not origfile is None:
|
|
os.unlink(origfile)
|
|
|
|
def mergefile(self, n, revision, mtime=None):
|
|
filename = os.path.join(self.dir, n)
|
|
storefilename = os.path.join(self.storedir, n)
|
|
myfilename = os.path.join(self.dir, n + '.mine')
|
|
upfilename = os.path.join(self.dir, n + '.r' + self.rev)
|
|
origfile_tmp = os.path.join(self.storedir, '_in_update', '%s.copy' % n)
|
|
origfile = os.path.join(self.storedir, '_in_update', n)
|
|
shutil.copyfile(filename, origfile_tmp)
|
|
os.rename(origfile_tmp, origfile)
|
|
os.rename(filename, myfilename)
|
|
|
|
get_source_file(self.apiurl, self.prjname, self.name, n,
|
|
revision=revision, targetfilename=upfilename,
|
|
progress_obj=self.progress_obj, mtime=mtime, meta=self.meta)
|
|
|
|
if binary_file(myfilename) or binary_file(upfilename):
|
|
# don't try merging
|
|
shutil.copyfile(upfilename, filename)
|
|
shutil.copyfile(upfilename, storefilename)
|
|
os.unlink(origfile)
|
|
self.in_conflict.append(n)
|
|
self.write_conflictlist()
|
|
return 'C'
|
|
else:
|
|
# try merging
|
|
# diff3 OPTIONS... MINE OLDER YOURS
|
|
merge_cmd = 'diff3 -m -E %s %s %s > %s' % (myfilename, storefilename, upfilename, filename)
|
|
# we would rather use the subprocess module, but it is not availablebefore 2.4
|
|
ret = subprocess.call(merge_cmd, shell=True)
|
|
|
|
# "An exit status of 0 means `diff3' was successful, 1 means some
|
|
# conflicts were found, and 2 means trouble."
|
|
if ret == 0:
|
|
# merge was successful... clean up
|
|
shutil.copyfile(upfilename, storefilename)
|
|
os.unlink(upfilename)
|
|
os.unlink(myfilename)
|
|
os.unlink(origfile)
|
|
return 'G'
|
|
elif ret == 1:
|
|
# unsuccessful merge
|
|
shutil.copyfile(upfilename, storefilename)
|
|
os.unlink(origfile)
|
|
self.in_conflict.append(n)
|
|
self.write_conflictlist()
|
|
return 'C'
|
|
else:
|
|
raise oscerr.ExtRuntimeError('diff3 failed with exit code: %s' % ret, merge_cmd)
|
|
|
|
def update_local_filesmeta(self, revision=None):
|
|
"""
|
|
Update the local _files file in the store.
|
|
It is replaced with the version pulled from upstream.
|
|
"""
|
|
meta = self.get_files_meta(revision=revision)
|
|
store_write_string(self.absdir, '_files', meta + '\n')
|
|
|
|
def get_files_meta(self, revision='latest', skip_service=True):
|
|
fm = show_files_meta(self.apiurl, self.prjname, self.name, revision=revision, meta=self.meta)
|
|
# look for "too large" files according to size limit and mark them
|
|
root = ET.fromstring(fm)
|
|
for e in root.findall('entry'):
|
|
size = e.get('size')
|
|
if size and self.size_limit and int(size) > self.size_limit \
|
|
or skip_service and (e.get('name').startswith('_service:') or e.get('name').startswith('_service_')):
|
|
e.set('skipped', 'true')
|
|
return ET.tostring(root)
|
|
|
|
def update_datastructs(self):
|
|
"""
|
|
Update the internal data structures if the local _files
|
|
file has changed (e.g. update_local_filesmeta() has been
|
|
called).
|
|
"""
|
|
import fnmatch
|
|
files_tree = read_filemeta(self.dir)
|
|
files_tree_root = files_tree.getroot()
|
|
|
|
self.rev = files_tree_root.get('rev')
|
|
self.srcmd5 = files_tree_root.get('srcmd5')
|
|
|
|
self.linkinfo = Linkinfo()
|
|
self.linkinfo.read(files_tree_root.find('linkinfo'))
|
|
|
|
self.filenamelist = []
|
|
self.filelist = []
|
|
self.skipped = []
|
|
for node in files_tree_root.findall('entry'):
|
|
try:
|
|
f = File(node.get('name'),
|
|
node.get('md5'),
|
|
int(node.get('size')),
|
|
int(node.get('mtime')))
|
|
if node.get('skipped'):
|
|
self.skipped.append(f.name)
|
|
f.skipped = True
|
|
except:
|
|
# okay, a very old version of _files, which didn't contain any metadata yet...
|
|
f = File(node.get('name'), '', 0, 0)
|
|
self.filelist.append(f)
|
|
self.filenamelist.append(f.name)
|
|
|
|
self.to_be_added = read_tobeadded(self.absdir)
|
|
self.to_be_deleted = read_tobedeleted(self.absdir)
|
|
self.in_conflict = read_inconflict(self.absdir)
|
|
self.linkrepair = os.path.isfile(os.path.join(self.storedir, '_linkrepair'))
|
|
self.size_limit = read_sizelimit(self.dir)
|
|
self.meta = self.ismetamode()
|
|
|
|
# gather unversioned files, but ignore some stuff
|
|
self.excluded = []
|
|
for i in os.listdir(self.dir):
|
|
for j in conf.config['exclude_glob']:
|
|
if fnmatch.fnmatch(i, j):
|
|
self.excluded.append(i)
|
|
break
|
|
self.filenamelist_unvers = [ i for i in os.listdir(self.dir)
|
|
if i not in self.excluded
|
|
if i not in self.filenamelist ]
|
|
|
|
def islink(self):
|
|
"""tells us if the package is a link (has 'linkinfo').
|
|
A package with linkinfo is a package which links to another package.
|
|
Returns True if the package is a link, otherwise False."""
|
|
return self.linkinfo.islink()
|
|
|
|
def isexpanded(self):
|
|
"""tells us if the package is a link which is expanded.
|
|
Returns True if the package is expanded, otherwise False."""
|
|
return self.linkinfo.isexpanded()
|
|
|
|
def islinkrepair(self):
|
|
"""tells us if we are repairing a broken source link."""
|
|
return self.linkrepair
|
|
|
|
def ispulled(self):
|
|
"""tells us if we have pulled a link."""
|
|
return os.path.isfile(os.path.join(self.storedir, '_pulled'))
|
|
|
|
def isfrozen(self):
|
|
"""tells us if the link is frozen."""
|
|
return os.path.isfile(os.path.join(self.storedir, '_frozenlink'))
|
|
|
|
def ismetamode(self):
|
|
"""tells us if the package is in meta mode"""
|
|
return os.path.isfile(os.path.join(self.storedir, '_meta_mode'))
|
|
|
|
def get_pulled_srcmd5(self):
|
|
pulledrev = None
|
|
for line in open(os.path.join(self.storedir, '_pulled'), 'r'):
|
|
pulledrev = line.strip()
|
|
return pulledrev
|
|
|
|
def haslinkerror(self):
|
|
"""
|
|
Returns True if the link is broken otherwise False.
|
|
If the package is not a link it returns False.
|
|
"""
|
|
return self.linkinfo.haserror()
|
|
|
|
def linkerror(self):
|
|
"""
|
|
Returns an error message if the link is broken otherwise None.
|
|
If the package is not a link it returns None.
|
|
"""
|
|
return self.linkinfo.error
|
|
|
|
def update_local_pacmeta(self):
|
|
"""
|
|
Update the local _meta file in the store.
|
|
It is replaced with the version pulled from upstream.
|
|
"""
|
|
meta = ''.join(show_package_meta(self.apiurl, self.prjname, self.name))
|
|
store_write_string(self.absdir, '_meta', meta + '\n')
|
|
|
|
def findfilebyname(self, n):
|
|
for i in self.filelist:
|
|
if i.name == n:
|
|
return i
|
|
|
|
def get_status(self, excluded=False, *exclude_states):
|
|
global store
|
|
todo = self.todo
|
|
if not todo:
|
|
todo = self.filenamelist + self.to_be_added + \
|
|
[i for i in self.filenamelist_unvers if not os.path.isdir(os.path.join(self.absdir, i))]
|
|
if excluded:
|
|
todo.extend([i for i in self.excluded if i != store])
|
|
todo = set(todo)
|
|
res = []
|
|
for fname in sorted(todo):
|
|
st = self.status(fname)
|
|
if not st in exclude_states:
|
|
res.append((st, fname))
|
|
return res
|
|
|
|
def status(self, n):
|
|
"""
|
|
status can be:
|
|
|
|
file storefile file present STATUS
|
|
exists exists in _files
|
|
|
|
x - - 'A' and listed in _to_be_added
|
|
x x - 'R' and listed in _to_be_added
|
|
x x x ' ' if digest differs: 'M'
|
|
and if in conflicts file: 'C'
|
|
x - - '?'
|
|
- x x 'D' and listed in _to_be_deleted
|
|
x x x 'D' and listed in _to_be_deleted (e.g. if deleted file was modified)
|
|
x x x 'C' and listed in _in_conflict
|
|
x - x 'S' and listed in self.skipped
|
|
- - x 'S' and listed in self.skipped
|
|
- x x '!'
|
|
- - - NOT DEFINED
|
|
|
|
"""
|
|
|
|
known_by_meta = False
|
|
exists = False
|
|
exists_in_store = False
|
|
if n in self.filenamelist:
|
|
known_by_meta = True
|
|
if os.path.exists(os.path.join(self.absdir, n)):
|
|
exists = True
|
|
if os.path.exists(os.path.join(self.storedir, n)):
|
|
exists_in_store = True
|
|
|
|
if n in self.to_be_deleted:
|
|
state = 'D'
|
|
elif n in self.in_conflict:
|
|
state = 'C'
|
|
elif n in self.skipped:
|
|
state = 'S'
|
|
elif n in self.to_be_added and exists and exists_in_store:
|
|
state = 'R'
|
|
elif n in self.to_be_added and exists:
|
|
state = 'A'
|
|
elif exists and exists_in_store and known_by_meta:
|
|
if dgst(os.path.join(self.absdir, n)) != self.findfilebyname(n).md5:
|
|
state = 'M'
|
|
else:
|
|
state = ' '
|
|
elif n in self.to_be_added and not exists:
|
|
state = '!'
|
|
elif not exists and exists_in_store and known_by_meta and not n in self.to_be_deleted:
|
|
state = '!'
|
|
elif exists and not exists_in_store and not known_by_meta:
|
|
state = '?'
|
|
elif not exists_in_store and known_by_meta:
|
|
# XXX: this codepath shouldn't be reached (we restore the storefile
|
|
# in update_datastructs)
|
|
raise oscerr.PackageInternalError(self.prjname, self.name,
|
|
'error: file \'%s\' is known by meta but no storefile exists.\n'
|
|
'This might be caused by an old wc format. Please backup your current\n'
|
|
'wc and checkout the package again. Afterwards copy all files (except the\n'
|
|
'.osc/ dir) into the new package wc.' % n)
|
|
else:
|
|
# this case shouldn't happen (except there was a typo in the filename etc.)
|
|
raise oscerr.OscIOError(None, 'osc: \'%s\' is not under version control' % n)
|
|
|
|
return state
|
|
|
|
def get_diff(self, revision=None, ignoreUnversioned=False):
|
|
import tempfile
|
|
diff_hdr = 'Index: %s\n'
|
|
diff_hdr += '===================================================================\n'
|
|
kept = []
|
|
added = []
|
|
deleted = []
|
|
def diff_add_delete(fname, add, revision):
|
|
diff = []
|
|
diff.append(diff_hdr % fname)
|
|
tmpfile = None
|
|
origname = fname
|
|
if add:
|
|
diff.append('--- %s\t(revision 0)\n' % fname)
|
|
rev = 'revision 0'
|
|
if revision and not fname in self.to_be_added:
|
|
rev = 'working copy'
|
|
diff.append('+++ %s\t(%s)\n' % (fname, rev))
|
|
fname = os.path.join(self.absdir, fname)
|
|
else:
|
|
diff.append('--- %s\t(revision %s)\n' % (fname, revision or self.rev))
|
|
diff.append('+++ %s\t(working copy)\n' % fname)
|
|
fname = os.path.join(self.storedir, fname)
|
|
|
|
try:
|
|
if revision is not None and not add:
|
|
(fd, tmpfile) = tempfile.mkstemp(prefix='osc_diff')
|
|
get_source_file(self.apiurl, self.prjname, self.name, origname, tmpfile, revision)
|
|
fname = tmpfile
|
|
if binary_file(fname):
|
|
what = 'added'
|
|
if not add:
|
|
what = 'deleted'
|
|
diff = diff[:1]
|
|
diff.append('Binary file \'%s\' %s.\n' % (origname, what))
|
|
return diff
|
|
tmpl = '+%s'
|
|
ltmpl = '@@ -0,0 +1,%d @@\n'
|
|
if not add:
|
|
tmpl = '-%s'
|
|
ltmpl = '@@ -1,%d +0,0 @@\n'
|
|
lines = [tmpl % i for i in open(fname, 'r').readlines()]
|
|
if len(lines):
|
|
diff.append(ltmpl % len(lines))
|
|
if not lines[-1].endswith('\n'):
|
|
lines.append('\n\\ No newline at end of file\n')
|
|
diff.extend(lines)
|
|
finally:
|
|
if tmpfile is not None:
|
|
os.close(fd)
|
|
os.unlink(tmpfile)
|
|
return diff
|
|
|
|
if revision is None:
|
|
todo = self.todo or [i for i in self.filenamelist if not i in self.to_be_added]+self.to_be_added
|
|
for fname in todo:
|
|
if fname in self.to_be_added and self.status(fname) == 'A':
|
|
added.append(fname)
|
|
elif fname in self.to_be_deleted:
|
|
deleted.append(fname)
|
|
elif fname in self.filenamelist:
|
|
kept.append(self.findfilebyname(fname))
|
|
elif fname in self.to_be_added and self.status(fname) == '!':
|
|
raise oscerr.OscIOError(None, 'file \'%s\' is marked as \'A\' but does not exist\n'\
|
|
'(either add the missing file or revert it)' % fname)
|
|
elif not ignoreUnversioned:
|
|
raise oscerr.OscIOError(None, 'file \'%s\' is not under version control' % fname)
|
|
else:
|
|
fm = self.get_files_meta(revision=revision)
|
|
root = ET.fromstring(fm)
|
|
rfiles = self.__get_files(root)
|
|
# swap added and deleted
|
|
kept, deleted, added, services = self.__get_rev_changes(rfiles)
|
|
added = [f.name for f in added]
|
|
added.extend([f for f in self.to_be_added if not f in kept])
|
|
deleted = [f.name for f in deleted]
|
|
deleted.extend(self.to_be_deleted)
|
|
for f in added[:]:
|
|
if f in deleted:
|
|
added.remove(f)
|
|
deleted.remove(f)
|
|
# print kept, added, deleted
|
|
for f in kept:
|
|
state = self.status(f.name)
|
|
if state in ('S', '?', '!'):
|
|
continue
|
|
elif state == ' ' and revision is None:
|
|
continue
|
|
elif revision and self.findfilebyname(f.name).md5 == f.md5 and state != 'M':
|
|
continue
|
|
yield [diff_hdr % f.name]
|
|
if revision is None:
|
|
yield get_source_file_diff(self.absdir, f.name, self.rev)
|
|
else:
|
|
tmpfile = None
|
|
diff = []
|
|
try:
|
|
(fd, tmpfile) = tempfile.mkstemp(prefix='osc_diff')
|
|
get_source_file(self.apiurl, self.prjname, self.name, f.name, tmpfile, revision)
|
|
diff = get_source_file_diff(self.absdir, f.name, revision,
|
|
os.path.basename(tmpfile), os.path.dirname(tmpfile), f.name)
|
|
finally:
|
|
if tmpfile is not None:
|
|
os.close(fd)
|
|
os.unlink(tmpfile)
|
|
yield diff
|
|
|
|
for f in added:
|
|
yield diff_add_delete(f, True, revision)
|
|
for f in deleted:
|
|
yield diff_add_delete(f, False, revision)
|
|
|
|
def merge(self, otherpac):
|
|
self.todo += otherpac.todo
|
|
|
|
def __str__(self):
|
|
r = """
|
|
name: %s
|
|
prjname: %s
|
|
workingdir: %s
|
|
localfilelist: %s
|
|
linkinfo: %s
|
|
rev: %s
|
|
'todo' files: %s
|
|
""" % (self.name,
|
|
self.prjname,
|
|
self.dir,
|
|
'\n '.join(self.filenamelist),
|
|
self.linkinfo,
|
|
self.rev,
|
|
self.todo)
|
|
|
|
return r
|
|
|
|
|
|
def read_meta_from_spec(self, spec = None):
|
|
import glob
|
|
if spec:
|
|
specfile = spec
|
|
else:
|
|
# scan for spec files
|
|
speclist = glob.glob(os.path.join(self.dir, '*.spec'))
|
|
if len(speclist) == 1:
|
|
specfile = speclist[0]
|
|
elif len(speclist) > 1:
|
|
print 'the following specfiles were found:'
|
|
for filename in speclist:
|
|
print filename
|
|
print 'please specify one with --specfile'
|
|
sys.exit(1)
|
|
else:
|
|
print 'no specfile was found - please specify one ' \
|
|
'with --specfile'
|
|
sys.exit(1)
|
|
|
|
data = read_meta_from_spec(specfile, 'Summary', 'Url', '%description')
|
|
self.summary = data.get('Summary', '')
|
|
self.url = data.get('Url', '')
|
|
self.descr = data.get('%description', '')
|
|
|
|
|
|
def update_package_meta(self, force=False):
|
|
"""
|
|
for the updatepacmetafromspec subcommand
|
|
argument force supress the confirm question
|
|
"""
|
|
|
|
m = ''.join(show_package_meta(self.apiurl, self.prjname, self.name))
|
|
|
|
root = ET.fromstring(m)
|
|
root.find('title').text = self.summary
|
|
root.find('description').text = ''.join(self.descr)
|
|
url = root.find('url')
|
|
if url == None:
|
|
url = ET.SubElement(root, 'url')
|
|
url.text = self.url
|
|
|
|
u = makeurl(self.apiurl, ['source', self.prjname, self.name, '_meta'])
|
|
mf = metafile(u, ET.tostring(root))
|
|
|
|
if not force:
|
|
print '*' * 36, 'old', '*' * 36
|
|
print m
|
|
print '*' * 36, 'new', '*' * 36
|
|
print ET.tostring(root)
|
|
print '*' * 72
|
|
repl = raw_input('Write? (y/N/e) ')
|
|
else:
|
|
repl = 'y'
|
|
|
|
if repl == 'y':
|
|
mf.sync()
|
|
elif repl == 'e':
|
|
mf.edit()
|
|
|
|
mf.discard()
|
|
|
|
def mark_frozen(self):
|
|
store_write_string(self.absdir, '_frozenlink', '')
|
|
print
|
|
print "The link in this package is currently broken. Checking"
|
|
print "out the last working version instead; please use 'osc pull'"
|
|
print "to repair the link."
|
|
print
|
|
|
|
def unmark_frozen(self):
|
|
if os.path.exists(os.path.join(self.storedir, '_frozenlink')):
|
|
os.unlink(os.path.join(self.storedir, '_frozenlink'))
|
|
|
|
def latest_rev(self, include_service_files=False):
|
|
if self.islinkrepair():
|
|
upstream_rev = show_upstream_xsrcmd5(self.apiurl, self.prjname, self.name, linkrepair=1, meta=self.meta, include_service_files=include_service_files)
|
|
elif self.islink() and self.isexpanded():
|
|
if self.isfrozen() or self.ispulled():
|
|
upstream_rev = show_upstream_xsrcmd5(self.apiurl, self.prjname, self.name, linkrev=self.linkinfo.srcmd5, meta=self.meta, include_service_files=include_service_files)
|
|
else:
|
|
try:
|
|
upstream_rev = show_upstream_xsrcmd5(self.apiurl, self.prjname, self.name, meta=self.meta, include_service_files=include_service_files)
|
|
except:
|
|
try:
|
|
upstream_rev = show_upstream_xsrcmd5(self.apiurl, self.prjname, self.name, linkrev=self.linkinfo.srcmd5, meta=self.meta, include_service_files=include_service_files)
|
|
except:
|
|
upstream_rev = show_upstream_xsrcmd5(self.apiurl, self.prjname, self.name, linkrev="base", meta=self.meta, include_service_files=include_service_files)
|
|
self.mark_frozen()
|
|
else:
|
|
upstream_rev = show_upstream_rev(self.apiurl, self.prjname, self.name, meta=self.meta, include_service_files=include_service_files)
|
|
return upstream_rev
|
|
|
|
def __get_files(self, fmeta_root):
|
|
f = []
|
|
if fmeta_root.get('rev') is None and len(fmeta_root.findall('entry')) > 0:
|
|
raise oscerr.APIError('missing rev attribute in _files:\n%s' % ''.join(ET.tostring(fmeta_root)))
|
|
for i in fmeta_root.findall('entry'):
|
|
skipped = i.get('skipped') is not None
|
|
f.append(File(i.get('name'), i.get('md5'),
|
|
int(i.get('size')), int(i.get('mtime')), skipped))
|
|
return f
|
|
|
|
def __get_rev_changes(self, revfiles):
|
|
kept = []
|
|
added = []
|
|
deleted = []
|
|
services = []
|
|
revfilenames = []
|
|
for f in revfiles:
|
|
revfilenames.append(f.name)
|
|
# treat skipped like deleted files
|
|
if f.skipped:
|
|
if f.name.startswith('_service:'):
|
|
services.append(f)
|
|
else:
|
|
deleted.append(f)
|
|
continue
|
|
# treat skipped like added files
|
|
# problem: this overwrites existing files during the update
|
|
# (because skipped files aren't in self.filenamelist_unvers)
|
|
if f.name in self.filenamelist and not f.name in self.skipped:
|
|
kept.append(f)
|
|
else:
|
|
added.append(f)
|
|
for f in self.filelist:
|
|
if not f.name in revfilenames:
|
|
deleted.append(f)
|
|
|
|
return kept, added, deleted, services
|
|
|
|
def update(self, rev = None, service_files = False, size_limit = None):
|
|
import tempfile
|
|
rfiles = []
|
|
# size_limit is only temporary for this update
|
|
old_size_limit = self.size_limit
|
|
if not size_limit is None:
|
|
self.size_limit = int(size_limit)
|
|
if os.path.isfile(os.path.join(self.storedir, '_in_update', '_files')):
|
|
print 'resuming broken update...'
|
|
root = ET.parse(os.path.join(self.storedir, '_in_update', '_files')).getroot()
|
|
rfiles = self.__get_files(root)
|
|
kept, added, deleted, services = self.__get_rev_changes(rfiles)
|
|
# check if we aborted in the middle of a file update
|
|
broken_file = os.listdir(os.path.join(self.storedir, '_in_update'))
|
|
broken_file.remove('_files')
|
|
if len(broken_file) == 1:
|
|
origfile = os.path.join(self.storedir, '_in_update', broken_file[0])
|
|
wcfile = os.path.join(self.absdir, broken_file[0])
|
|
origfile_md5 = dgst(origfile)
|
|
origfile_meta = self.findfilebyname(broken_file[0])
|
|
if origfile.endswith('.copy'):
|
|
# ok it seems we aborted at some point during the copy process
|
|
# (copy process == copy wcfile to the _in_update dir). remove file+continue
|
|
os.unlink(origfile)
|
|
elif self.findfilebyname(broken_file[0]) is None:
|
|
# should we remove this file from _in_update? if we don't
|
|
# the user has no chance to continue without removing the file manually
|
|
raise oscerr.PackageInternalError(self.prjname, self.name,
|
|
'\'%s\' is not known by meta but exists in \'_in_update\' dir')
|
|
elif os.path.isfile(wcfile) and dgst(wcfile) != origfile_md5:
|
|
(fd, tmpfile) = tempfile.mkstemp(dir=self.absdir, prefix=broken_file[0]+'.')
|
|
os.close(fd)
|
|
os.rename(wcfile, tmpfile)
|
|
os.rename(origfile, wcfile)
|
|
print 'warning: it seems you modified \'%s\' after the broken ' \
|
|
'update. Restored original file and saved modified version ' \
|
|
'to \'%s\'.' % (wcfile, tmpfile)
|
|
elif not os.path.isfile(wcfile):
|
|
# this is strange... because it existed before the update. restore it
|
|
os.rename(origfile, wcfile)
|
|
else:
|
|
# everything seems to be ok
|
|
os.unlink(origfile)
|
|
elif len(broken_file) > 1:
|
|
raise oscerr.PackageInternalError(self.prjname, self.name, 'too many files in \'_in_update\' dir')
|
|
tmp = rfiles[:]
|
|
for f in tmp:
|
|
if os.path.exists(os.path.join(self.storedir, f.name)):
|
|
if dgst(os.path.join(self.storedir, f.name)) == f.md5:
|
|
if f in kept:
|
|
kept.remove(f)
|
|
elif f in added:
|
|
added.remove(f)
|
|
# this can't happen
|
|
elif f in deleted:
|
|
deleted.remove(f)
|
|
if not service_files:
|
|
services = []
|
|
self.__update(kept, added, deleted, services, ET.tostring(root), root.get('rev'))
|
|
os.unlink(os.path.join(self.storedir, '_in_update', '_files'))
|
|
os.rmdir(os.path.join(self.storedir, '_in_update'))
|
|
# ok everything is ok (hopefully)...
|
|
fm = self.get_files_meta(revision=rev)
|
|
root = ET.fromstring(fm)
|
|
rfiles = self.__get_files(root)
|
|
store_write_string(self.absdir, '_files', fm + '\n', subdir='_in_update')
|
|
kept, added, deleted, services = self.__get_rev_changes(rfiles)
|
|
if not service_files:
|
|
services = []
|
|
self.__update(kept, added, deleted, services, fm, root.get('rev'))
|
|
os.unlink(os.path.join(self.storedir, '_in_update', '_files'))
|
|
if os.path.isdir(os.path.join(self.storedir, '_in_update')):
|
|
os.rmdir(os.path.join(self.storedir, '_in_update'))
|
|
self.size_limit = old_size_limit
|
|
|
|
def __update(self, kept, added, deleted, services, fm, rev):
|
|
pathn = getTransActPath(self.dir)
|
|
# check for conflicts with existing files
|
|
for f in added:
|
|
if f.name in self.filenamelist_unvers:
|
|
raise oscerr.PackageFileConflict(self.prjname, self.name, f.name,
|
|
'failed to add file \'%s\' file/dir with the same name already exists' % f.name)
|
|
# ok, the update can't fail due to existing files
|
|
for f in added:
|
|
self.updatefile(f.name, rev, f.mtime)
|
|
print statfrmt('A', os.path.join(pathn, f.name))
|
|
for f in deleted:
|
|
# if the storefile doesn't exist we're resuming an aborted update:
|
|
# the file was already deleted but we cannot know this
|
|
# OR we're processing a _service: file (simply keep the file)
|
|
if os.path.isfile(os.path.join(self.storedir, f.name)) and self.status(f.name) != 'M':
|
|
# if self.status(f.name) != 'M':
|
|
self.delete_localfile(f.name)
|
|
self.delete_storefile(f.name)
|
|
print statfrmt('D', os.path.join(pathn, f.name))
|
|
if f.name in self.to_be_deleted:
|
|
self.to_be_deleted.remove(f.name)
|
|
self.write_deletelist()
|
|
|
|
for f in kept:
|
|
state = self.status(f.name)
|
|
# print f.name, state
|
|
if state == 'M' and self.findfilebyname(f.name).md5 == f.md5:
|
|
# remote file didn't change
|
|
pass
|
|
elif state == 'M':
|
|
# try to merge changes
|
|
merge_status = self.mergefile(f.name, rev, f.mtime)
|
|
print statfrmt(merge_status, os.path.join(pathn, f.name))
|
|
elif state == '!':
|
|
self.updatefile(f.name, rev, f.mtime)
|
|
print 'Restored \'%s\'' % os.path.join(pathn, f.name)
|
|
elif state == 'C':
|
|
get_source_file(self.apiurl, self.prjname, self.name, f.name,
|
|
targetfilename=os.path.join(self.storedir, f.name), revision=rev,
|
|
progress_obj=self.progress_obj, mtime=f.mtime, meta=self.meta)
|
|
print 'skipping \'%s\' (this is due to conflicts)' % f.name
|
|
elif state == 'D' and self.findfilebyname(f.name).md5 != f.md5:
|
|
# XXX: in the worst case we might end up with f.name being
|
|
# in _to_be_deleted and in _in_conflict... this needs to be checked
|
|
if os.path.exists(os.path.join(self.absdir, f.name)):
|
|
merge_status = self.mergefile(f.name, rev, f.mtime)
|
|
print statfrmt(merge_status, os.path.join(pathn, f.name))
|
|
if merge_status == 'C':
|
|
# state changes from delete to conflict
|
|
self.to_be_deleted.remove(f.name)
|
|
self.write_deletelist()
|
|
else:
|
|
# XXX: we cannot recover this case because we've no file
|
|
# to backup
|
|
self.updatefile(f.name, rev, f.mtime)
|
|
print statfrmt('U', os.path.join(pathn, f.name))
|
|
elif state == ' ' and self.findfilebyname(f.name).md5 != f.md5:
|
|
self.updatefile(f.name, rev, f.mtime)
|
|
print statfrmt('U', os.path.join(pathn, f.name))
|
|
|
|
# checkout service files
|
|
for f in services:
|
|
get_source_file(self.apiurl, self.prjname, self.name, f.name,
|
|
targetfilename=os.path.join(self.absdir, f.name), revision=rev,
|
|
progress_obj=self.progress_obj, mtime=f.mtime, meta=self.meta)
|
|
print statfrmt('A', os.path.join(pathn, f.name))
|
|
store_write_string(self.absdir, '_files', fm + '\n')
|
|
if not self.meta:
|
|
self.update_local_pacmeta()
|
|
self.update_datastructs()
|
|
|
|
print 'At revision %s.' % self.rev
|
|
|
|
def run_source_services(self, mode=None, singleservice=None, verbose=None):
|
|
if self.name.startswith("_"):
|
|
return 0
|
|
curdir = os.getcwd()
|
|
os.chdir(self.absdir) # e.g. /usr/lib/obs/service/verify_file fails if not inside the project dir.
|
|
si = Serviceinfo()
|
|
if os.path.exists('_service'):
|
|
if self.filenamelist.count('_service') or self.filenamelist_unvers.count('_service'):
|
|
service = ET.parse(os.path.join(self.absdir, '_service')).getroot()
|
|
si.read(service)
|
|
si.getProjectGlobalServices(self.apiurl, self.prjname, self.name)
|
|
r = si.execute(self.absdir, mode, singleservice, verbose)
|
|
os.chdir(curdir)
|
|
return r
|
|
|
|
def revert(self, filename):
|
|
if not filename in self.filenamelist and not filename in self.to_be_added:
|
|
raise oscerr.OscIOError(None, 'file \'%s\' is not under version control' % filename)
|
|
elif filename in self.skipped:
|
|
raise oscerr.OscIOError(None, 'file \'%s\' is marked as skipped and cannot be reverted' % filename)
|
|
if filename in self.filenamelist and not os.path.exists(os.path.join(self.storedir, filename)):
|
|
raise oscerr.PackageInternalError('file \'%s\' is listed in filenamelist but no storefile exists' % filename)
|
|
state = self.status(filename)
|
|
if not (state == 'A' or state == '!' and filename in self.to_be_added):
|
|
shutil.copyfile(os.path.join(self.storedir, filename), os.path.join(self.absdir, filename))
|
|
if state == 'D':
|
|
self.to_be_deleted.remove(filename)
|
|
self.write_deletelist()
|
|
elif state == 'C':
|
|
self.clear_from_conflictlist(filename)
|
|
elif state in ('A', 'R') or state == '!' and filename in self.to_be_added:
|
|
self.to_be_added.remove(filename)
|
|
self.write_addlist()
|
|
|
|
@staticmethod
|
|
def init_package(apiurl, project, package, dir, size_limit=None, meta=False, progress_obj=None):
|
|
global store
|
|
|
|
if not os.path.exists(dir):
|
|
os.mkdir(dir)
|
|
elif not os.path.isdir(dir):
|
|
raise oscerr.OscIOError(None, 'error: \'%s\' is no directory' % dir)
|
|
if os.path.exists(os.path.join(dir, store)):
|
|
raise oscerr.OscIOError(None, 'error: \'%s\' is already an initialized osc working copy' % dir)
|
|
else:
|
|
os.mkdir(os.path.join(dir, store))
|
|
store_write_project(dir, project)
|
|
store_write_string(dir, '_package', package + '\n')
|
|
store_write_apiurl(dir, apiurl)
|
|
if meta:
|
|
store_write_string(dir, '_meta_mode', '')
|
|
if size_limit:
|
|
store_write_string(dir, '_size_limit', str(size_limit) + '\n')
|
|
store_write_string(dir, '_files', '<directory />' + '\n')
|
|
store_write_string(dir, '_osclib_version', __store_version__ + '\n')
|
|
return Package(dir, progress_obj=progress_obj, size_limit=size_limit)
|
|
|
|
|
|
class AbstractState:
|
|
"""
|
|
Base class which represents state-like objects (<review />, <state />).
|
|
"""
|
|
def __init__(self, tag):
|
|
self.__tag = tag
|
|
|
|
def get_node_attrs(self):
|
|
"""return attributes for the tag/element"""
|
|
raise NotImplementedError()
|
|
|
|
def get_node_name(self):
|
|
"""return tag/element name"""
|
|
return self.__tag
|
|
|
|
def get_comment(self):
|
|
"""return data from <comment /> tag"""
|
|
raise NotImplementedError()
|
|
|
|
def to_xml(self):
|
|
"""serialize object to XML"""
|
|
root = ET.Element(self.get_node_name())
|
|
for attr in self.get_node_attrs():
|
|
val = getattr(self, attr)
|
|
if not val is None:
|
|
root.set(attr, val)
|
|
if self.get_comment():
|
|
ET.SubElement(root, 'comment').text = self.get_comment()
|
|
return root
|
|
|
|
def to_str(self):
|
|
"""return "pretty" XML data"""
|
|
root = self.to_xml()
|
|
xmlindent(root)
|
|
return ET.tostring(root)
|
|
|
|
|
|
class ReviewState(AbstractState):
|
|
"""Represents the review state in a request"""
|
|
def __init__(self, review_node):
|
|
if not review_node.get('state'):
|
|
raise oscerr.APIError('invalid review node (state attr expected): %s' % \
|
|
ET.tostring(review_node))
|
|
AbstractState.__init__(self, review_node.tag)
|
|
self.state = review_node.get('state')
|
|
self.by_user = review_node.get('by_user')
|
|
self.by_group = review_node.get('by_group')
|
|
self.by_project = review_node.get('by_project')
|
|
self.by_package = review_node.get('by_package')
|
|
self.who = review_node.get('who')
|
|
self.when = review_node.get('when')
|
|
self.comment = ''
|
|
if not review_node.find('comment') is None and \
|
|
review_node.find('comment').text:
|
|
self.comment = review_node.find('comment').text.strip()
|
|
|
|
def get_node_attrs(self):
|
|
return ('state', 'by_user', 'by_group', 'by_project', 'by_package', 'who', 'when')
|
|
|
|
def get_comment(self):
|
|
return self.comment
|
|
|
|
|
|
class RequestState(AbstractState):
|
|
"""Represents the state of a request"""
|
|
def __init__(self, state_node):
|
|
if not state_node.get('name'):
|
|
raise oscerr.APIError('invalid request state node (name attr expected): %s' % \
|
|
ET.tostring(state_node))
|
|
AbstractState.__init__(self, state_node.tag)
|
|
self.name = state_node.get('name')
|
|
self.who = state_node.get('who')
|
|
self.when = state_node.get('when')
|
|
self.comment = ''
|
|
if not state_node.find('comment') is None and \
|
|
state_node.find('comment').text:
|
|
self.comment = state_node.find('comment').text.strip()
|
|
|
|
def get_node_attrs(self):
|
|
return ('name', 'who', 'when')
|
|
|
|
def get_comment(self):
|
|
return self.comment
|
|
|
|
|
|
class Action:
|
|
"""
|
|
Represents a <action /> element of a Request.
|
|
This class is quite common so that it can be used for all different
|
|
action types. Note: instances only provide attributes for their specific
|
|
type.
|
|
Examples:
|
|
r = Action('set_bugowner', tgt_project='foo', person_name='buguser')
|
|
# available attributes: r.type (== 'set_bugowner'), r.tgt_project (== 'foo'), r.tgt_package (== None)
|
|
r.to_str() ->
|
|
<action type="set_bugowner">
|
|
<target project="foo" />
|
|
<person name="buguser" />
|
|
</action>
|
|
##
|
|
r = Action('delete', tgt_project='foo', tgt_package='bar')
|
|
# available attributes: r.type (== 'delete'), r.tgt_project (== 'foo'), r.tgt_package (=='bar')
|
|
r.to_str() ->
|
|
<action type="delete">
|
|
<target package="bar" project="foo" />
|
|
</action>
|
|
"""
|
|
|
|
# allowed types + the corresponding (allowed) attributes
|
|
type_args = {'submit': ('src_project', 'src_package', 'src_rev', 'tgt_project', 'tgt_package', 'opt_sourceupdate',
|
|
'acceptinfo_rev', 'acceptinfo_srcmd5', 'acceptinfo_xsrcmd5', 'acceptinfo_osrcmd5',
|
|
'acceptinfo_oxsrcmd5', 'opt_updatelink'),
|
|
'add_role': ('tgt_project', 'tgt_package', 'person_name', 'person_role', 'group_name', 'group_role'),
|
|
'set_bugowner': ('tgt_project', 'tgt_package', 'person_name'), # obsoleted by add_role
|
|
'maintenance_release': ('src_project', 'src_package', 'src_rev', 'tgt_project', 'tgt_package', 'person_name'),
|
|
'maintenance_incident': ('src_project', 'tgt_project', 'person_name', 'opt_sourceupdate'),
|
|
'delete': ('tgt_project', 'tgt_package'),
|
|
'change_devel': ('src_project', 'src_package', 'tgt_project', 'tgt_package')}
|
|
# attribute prefix to element name map (only needed for abbreviated attributes)
|
|
prefix_to_elm = {'src': 'source', 'tgt': 'target', 'opt': 'options'}
|
|
|
|
def __init__(self, type, **kwargs):
|
|
if not type in Action.type_args.keys():
|
|
raise oscerr.WrongArgs('invalid action type: \'%s\'' % type)
|
|
self.type = type
|
|
for i in kwargs.keys():
|
|
if not i in Action.type_args[type]:
|
|
raise oscerr.WrongArgs('invalid argument: \'%s\'' % i)
|
|
# set all type specific attributes
|
|
for i in Action.type_args[type]:
|
|
if kwargs.has_key(i):
|
|
setattr(self, i, kwargs[i])
|
|
else:
|
|
setattr(self, i, None)
|
|
|
|
def to_xml(self):
|
|
"""
|
|
Serialize object to XML.
|
|
The xml tag names and attributes are constructed from the instance's attributes.
|
|
Example:
|
|
self.group_name -> tag name is "group", attribute name is "name"
|
|
self.src_project -> tag name is "source" (translated via prefix_to_elm dict),
|
|
attribute name is "project"
|
|
Attributes prefixed with "opt_" need a special handling, the resulting xml should
|
|
look like this: opt_updatelink -> <options><updatelink>value</updatelink></options>.
|
|
Attributes which are "None" will be skipped.
|
|
"""
|
|
root = ET.Element('action', type=self.type)
|
|
for i in Action.type_args[self.type]:
|
|
prefix, attr = i.split('_', 1)
|
|
val = getattr(self, i)
|
|
if val is None:
|
|
continue
|
|
elm = root.find(Action.prefix_to_elm.get(prefix, prefix))
|
|
if elm is None:
|
|
elm = ET.Element(Action.prefix_to_elm.get(prefix, prefix))
|
|
root.append(elm)
|
|
if prefix == 'opt':
|
|
ET.SubElement(elm, attr).text = val
|
|
else:
|
|
elm.set(attr, val)
|
|
return root
|
|
|
|
def to_str(self):
|
|
"""return "pretty" XML data"""
|
|
root = self.to_xml()
|
|
xmlindent(root)
|
|
return ET.tostring(root)
|
|
|
|
@staticmethod
|
|
def from_xml(action_node):
|
|
"""create action from XML"""
|
|
if action_node is None or \
|
|
not action_node.get('type') in Action.type_args.keys() or \
|
|
not action_node.tag in ('action', 'submit'):
|
|
raise oscerr.WrongArgs('invalid argument')
|
|
elm_to_prefix = dict([(i[1], i[0]) for i in Action.prefix_to_elm.items()])
|
|
kwargs = {}
|
|
for node in action_node:
|
|
prefix = elm_to_prefix.get(node.tag, node.tag)
|
|
if prefix == 'opt':
|
|
data = [('opt_%s' % opt.tag, opt.text.strip()) for opt in node if opt.text]
|
|
else:
|
|
data = [('%s_%s' % (prefix, k), v) for k, v in node.items()]
|
|
kwargs.update(dict(data))
|
|
return Action(action_node.get('type'), **kwargs)
|
|
|
|
|
|
class Request:
|
|
"""Represents a request (<request />)"""
|
|
|
|
def __init__(self):
|
|
self._init_attributes()
|
|
|
|
def _init_attributes(self):
|
|
"""initialize attributes with default values"""
|
|
self.reqid = None
|
|
self.title = ''
|
|
self.description = ''
|
|
self.state = None
|
|
self.actions = []
|
|
self.statehistory = []
|
|
self.reviews = []
|
|
|
|
def read(self, root):
|
|
"""read in a request"""
|
|
self._init_attributes()
|
|
if not root.get('id'):
|
|
raise oscerr.APIError('invalid request: %s\n' % ET.tostring(root))
|
|
self.reqid = root.get('id')
|
|
if root.find('state') is None:
|
|
raise oscerr.APIError('invalid request (state expected): %s\n' % ET.tostring(root))
|
|
self.state = RequestState(root.find('state'))
|
|
action_nodes = root.findall('action')
|
|
if not action_nodes:
|
|
# check for old-style requests
|
|
for i in root.findall('submit'):
|
|
i.set('type', 'submit')
|
|
action_nodes.append(i)
|
|
for action in action_nodes:
|
|
self.actions.append(Action.from_xml(action))
|
|
for review in root.findall('review'):
|
|
self.reviews.append(ReviewState(review))
|
|
for hist_state in root.findall('history'):
|
|
self.statehistory.append(RequestState(hist_state))
|
|
if not root.find('title') is None:
|
|
self.title = root.find('title').text.strip()
|
|
if not root.find('description') is None and root.find('description').text:
|
|
self.description = root.find('description').text.strip()
|
|
|
|
def add_action(self, type, **kwargs):
|
|
"""add a new action to the request"""
|
|
self.actions.append(Action(type, **kwargs))
|
|
|
|
def get_actions(self, *types):
|
|
"""
|
|
get all actions with a specific type
|
|
(if types is empty return all actions)
|
|
"""
|
|
if not types:
|
|
return self.actions
|
|
return [i for i in self.actions if i.type in types]
|
|
|
|
def get_creator(self):
|
|
"""return the creator of the request"""
|
|
if len(self.statehistory):
|
|
return self.statehistory[0].who
|
|
return self.state.who
|
|
|
|
def to_xml(self):
|
|
"""serialize object to XML"""
|
|
root = ET.Element('request')
|
|
if not self.reqid is None:
|
|
root.set('id', self.reqid)
|
|
for action in self.actions:
|
|
root.append(action.to_xml())
|
|
if not self.state is None:
|
|
root.append(self.state.to_xml())
|
|
for review in self.reviews:
|
|
root.append(review.to_xml())
|
|
for hist in self.statehistory:
|
|
root.append(hist.to_xml())
|
|
if self.title:
|
|
ET.SubElement(root, 'title').text = self.title
|
|
if self.description:
|
|
ET.SubElement(root, 'description').text = self.description
|
|
return root
|
|
|
|
def to_str(self):
|
|
"""return "pretty" XML data"""
|
|
root = self.to_xml()
|
|
xmlindent(root)
|
|
return ET.tostring(root)
|
|
|
|
@staticmethod
|
|
def format_review(review, show_srcupdate=False):
|
|
"""
|
|
format a review depending on the reviewer's type.
|
|
A dict which contains the formatted str's is returned.
|
|
"""
|
|
|
|
d = {'state': '%s:' % review.state}
|
|
if review.by_package:
|
|
d['by'] = '%s/%s' % (review.by_project, review.by_package)
|
|
d['type'] = 'Package'
|
|
elif review.by_project:
|
|
d['by'] = '%s' % review.by_project
|
|
d['type'] = 'Project'
|
|
elif review.by_group:
|
|
d['by'] = '%s' % review.by_group
|
|
d['type'] = 'Group'
|
|
else:
|
|
d['by'] = '%s' % review.by_user
|
|
d['type'] = 'User'
|
|
if review.who:
|
|
d['by'] += '(%s)' % review.who
|
|
return d
|
|
|
|
@staticmethod
|
|
def format_action(action, show_srcupdate=False):
|
|
"""
|
|
format an action depending on the action's type.
|
|
A dict which contains the formatted str's is returned.
|
|
"""
|
|
def prj_pkg_join(prj, pkg):
|
|
if not pkg:
|
|
return prj or ''
|
|
return '%s/%s' % (prj, pkg)
|
|
|
|
d = {'type': '%s:' % action.type}
|
|
if action.type == 'set_bugowner':
|
|
d['source'] = action.person_name
|
|
d['target'] = prj_pkg_join(action.tgt_project, action.tgt_package)
|
|
elif action.type == 'change_devel':
|
|
d['source'] = prj_pkg_join(action.tgt_project, action.tgt_package)
|
|
d['target'] = 'developed in %s' % prj_pkg_join(action.src_project, action.src_package)
|
|
elif action.type == 'maintenance_incident':
|
|
d['source'] = '%s ->' % action.src_project
|
|
d['target'] = action.tgt_project
|
|
srcupdate = ' '
|
|
if action.opt_sourceupdate and show_srcupdate:
|
|
srcupdate = '(%s)' % action.opt_sourceupdate
|
|
elif action.type == 'maintenance_release':
|
|
d['source'] = '%s ->' % prj_pkg_join(action.src_project, action.src_package)
|
|
d['target'] = prj_pkg_join(action.tgt_project, action.tgt_package)
|
|
elif action.type == 'submit':
|
|
srcupdate = ' '
|
|
if action.opt_sourceupdate and show_srcupdate:
|
|
srcupdate = '(%s)' % action.opt_sourceupdate
|
|
d['source'] = '%s%s ->' % (prj_pkg_join(action.src_project, action.src_package), srcupdate)
|
|
tgt_package = action.tgt_package
|
|
if action.src_package == action.tgt_package:
|
|
tgt_package = ''
|
|
d['target'] = prj_pkg_join(action.tgt_project, tgt_package)
|
|
elif action.type == 'add_role':
|
|
roles = []
|
|
if action.person_name and action.person_role:
|
|
roles.append('person: %s as %s' % (action.person_name, action.person_role))
|
|
if action.group_name and action.group_role:
|
|
roles.append('group: %s as %s' % (action.group_name, action.group_role))
|
|
d['source'] = ', '.join(roles)
|
|
d['target'] = prj_pkg_join(action.tgt_project, action.tgt_package)
|
|
elif action.type == 'delete':
|
|
d['source'] = ''
|
|
d['target'] = prj_pkg_join(action.tgt_project, action.tgt_package)
|
|
return d
|
|
|
|
def list_view(self):
|
|
"""return "list view" format"""
|
|
import textwrap
|
|
lines = ['%6s State:%-10s By:%-12s When:%-19s' % (self.reqid, self.state.name, self.state.who, self.state.when)]
|
|
tmpl = ' %(type)-16s %(source)-50s %(target)s'
|
|
for action in self.actions:
|
|
lines.append(tmpl % Request.format_action(action))
|
|
tmpl = ' Review by %(type)-10s is %(state)-10s %(by)-50s'
|
|
for review in self.reviews:
|
|
lines.append(tmpl % Request.format_review(review))
|
|
history = ['%s(%s)' % (hist.name, hist.who) for hist in self.statehistory]
|
|
if history:
|
|
lines.append(' From: %s' % ' -> '.join(history))
|
|
if self.description:
|
|
lines.append(textwrap.fill(self.description, width=80, initial_indent=' Descr: ',
|
|
subsequent_indent=' '))
|
|
lines.append(textwrap.fill(self.state.comment, width=80, initial_indent=' Comment: ',
|
|
subsequent_indent=' '))
|
|
return '\n'.join(lines)
|
|
|
|
def __str__(self):
|
|
"""return "detailed" format"""
|
|
lines = ['Request: #%s\n' % self.reqid]
|
|
for action in self.actions:
|
|
tmpl = ' %(type)-13s %(source)s %(target)s'
|
|
if action.type == 'delete':
|
|
# remove 1 whitespace because source is empty
|
|
tmpl = ' %(type)-12s %(source)s %(target)s'
|
|
lines.append(tmpl % Request.format_action(action, show_srcupdate=True))
|
|
lines.append('\n\nMessage:')
|
|
if self.description:
|
|
lines.append(self.description)
|
|
else:
|
|
lines.append('<no message>')
|
|
if self.state:
|
|
lines.append('\nState: %-10s %-12s %s' % (self.state.name, self.state.when, self.state.who))
|
|
lines.append('Comment: %s' % (self.state.comment or '<no comment>'))
|
|
|
|
indent = '\n '
|
|
tmpl = '%(state)-10s %(by)-50s %(when)-12s %(who)-20s %(comment)s'
|
|
reviews = []
|
|
for review in reversed(self.reviews):
|
|
d = {'state': review.state}
|
|
if review.by_user:
|
|
d['by'] = "User: " + review.by_user
|
|
if review.by_group:
|
|
d['by'] = "Group: " + review.by_group
|
|
if review.by_package:
|
|
d['by'] = "Package: " + review.by_project + "/" + review.by_package
|
|
elif review.by_project:
|
|
d['by'] = "Project: " + review.by_project
|
|
d['when'] = review.when or ''
|
|
d['who'] = review.who or ''
|
|
d['comment'] = review.comment or ''
|
|
reviews.append(tmpl % d)
|
|
if reviews:
|
|
lines.append('\nReview: %s' % indent.join(reviews))
|
|
|
|
tmpl = '%(name)-10s %(when)-12s %(who)s'
|
|
histories = []
|
|
for hist in reversed(self.statehistory):
|
|
d = {'name': hist.name, 'when': hist.when,
|
|
'who': hist.who}
|
|
histories.append(tmpl % d)
|
|
if histories:
|
|
lines.append('\nHistory: %s' % indent.join(histories))
|
|
|
|
return '\n'.join(lines)
|
|
|
|
def __cmp__(self, other):
|
|
return cmp(int(self.reqid), int(other.reqid))
|
|
|
|
def create(self, apiurl):
|
|
"""create a new request"""
|
|
u = makeurl(apiurl, ['request'], query='cmd=create')
|
|
f = http_POST(u, data=self.to_str())
|
|
root = ET.fromstring(f.read())
|
|
self.read(root)
|
|
|
|
def shorttime(t):
|
|
"""format time as Apr 02 18:19
|
|
or Apr 02 2005
|
|
depending on whether it is in the current year
|
|
"""
|
|
import time
|
|
|
|
if time.localtime()[0] == time.localtime(t)[0]:
|
|
# same year
|
|
return time.strftime('%b %d %H:%M',time.localtime(t))
|
|
else:
|
|
return time.strftime('%b %d %Y',time.localtime(t))
|
|
|
|
|
|
def is_project_dir(d):
|
|
global store
|
|
|
|
return os.path.exists(os.path.join(d, store, '_project')) and not \
|
|
os.path.exists(os.path.join(d, store, '_package'))
|
|
|
|
|
|
def is_package_dir(d):
|
|
global store
|
|
|
|
return os.path.exists(os.path.join(d, store, '_project')) and \
|
|
os.path.exists(os.path.join(d, store, '_package'))
|
|
|
|
def parse_disturl(disturl):
|
|
"""Parse a disturl, returns tuple (apiurl, project, source, repository,
|
|
revision), else raises an oscerr.WrongArgs exception
|
|
"""
|
|
|
|
global DISTURL_RE
|
|
|
|
m = DISTURL_RE.match(disturl)
|
|
if not m:
|
|
raise oscerr.WrongArgs("`%s' does not look like disturl" % disturl)
|
|
|
|
apiurl = m.group('apiurl')
|
|
if apiurl.split('.')[0] != 'api':
|
|
apiurl = 'https://api.' + ".".join(apiurl.split('.')[1:])
|
|
return (apiurl, m.group('project'), m.group('source'), m.group('repository'), m.group('revision'))
|
|
|
|
def parse_buildlogurl(buildlogurl):
|
|
"""Parse a build log url, returns a tuple (apiurl, project, package,
|
|
repository, arch), else raises oscerr.WrongArgs exception"""
|
|
|
|
global BUILDLOGURL_RE
|
|
|
|
m = BUILDLOGURL_RE.match(buildlogurl)
|
|
if not m:
|
|
raise oscerr.WrongArgs('\'%s\' does not look like url with a build log' % buildlogurl)
|
|
|
|
return (m.group('apiurl'), m.group('project'), m.group('package'), m.group('repository'), m.group('arch'))
|
|
|
|
def slash_split(l):
|
|
"""Split command line arguments like 'foo/bar' into 'foo' 'bar'.
|
|
This is handy to allow copy/paste a project/package combination in this form.
|
|
|
|
Trailing slashes are removed before the split, because the split would
|
|
otherwise give an additional empty string.
|
|
"""
|
|
r = []
|
|
for i in l:
|
|
i = i.rstrip('/')
|
|
r += i.split('/')
|
|
return r
|
|
|
|
def expand_proj_pack(args, idx=0, howmany=0):
|
|
"""looks for occurance of '.' at the position idx.
|
|
If howmany is 2, both proj and pack are expanded together
|
|
using the current directory, or none of them, if not possible.
|
|
If howmany is 0, proj is expanded if possible, then, if there
|
|
is no idx+1 element in args (or args[idx+1] == '.'), pack is also
|
|
expanded, if possible.
|
|
If howmany is 1, only proj is expanded if possible.
|
|
|
|
If args[idx] does not exists, an implicit '.' is assumed.
|
|
if not enough elements up to idx exist, an error is raised.
|
|
|
|
See also parseargs(args), slash_split(args), findpacs(args)
|
|
All these need unification, somehow.
|
|
"""
|
|
|
|
# print args,idx,howmany
|
|
|
|
if len(args) < idx:
|
|
raise oscerr.WrongArgs('not enough argument, expected at least %d' % idx)
|
|
|
|
if len(args) == idx:
|
|
args += '.'
|
|
if args[idx+0] == '.':
|
|
if howmany == 0 and len(args) > idx+1:
|
|
if args[idx+1] == '.':
|
|
# we have two dots.
|
|
# remove one dot and make sure to expand both proj and pack
|
|
args.pop(idx+1)
|
|
howmany = 2
|
|
else:
|
|
howmany = 1
|
|
# print args,idx,howmany
|
|
|
|
args[idx+0] = store_read_project('.')
|
|
if howmany == 0:
|
|
try:
|
|
package = store_read_package('.')
|
|
args.insert(idx+1, package)
|
|
except:
|
|
pass
|
|
elif howmany == 2:
|
|
package = store_read_package('.')
|
|
args.insert(idx+1, package)
|
|
return args
|
|
|
|
|
|
def findpacs(files, progress_obj=None):
|
|
"""collect Package objects belonging to the given files
|
|
and make sure each Package is returned only once"""
|
|
pacs = []
|
|
for f in files:
|
|
p = filedir_to_pac(f, progress_obj)
|
|
known = None
|
|
for i in pacs:
|
|
if i.name == p.name:
|
|
known = i
|
|
break
|
|
if known:
|
|
i.merge(p)
|
|
else:
|
|
pacs.append(p)
|
|
return pacs
|
|
|
|
|
|
def filedir_to_pac(f, progress_obj=None):
|
|
"""Takes a working copy path, or a path to a file inside a working copy,
|
|
and returns a Package object instance
|
|
|
|
If the argument was a filename, add it onto the "todo" list of the Package """
|
|
|
|
if os.path.isdir(f):
|
|
wd = f
|
|
p = Package(wd, progress_obj=progress_obj)
|
|
else:
|
|
wd = os.path.dirname(f) or os.curdir
|
|
p = Package(wd, progress_obj=progress_obj)
|
|
p.todo = [ os.path.basename(f) ]
|
|
return p
|
|
|
|
|
|
def read_filemeta(dir):
|
|
global store
|
|
|
|
msg = '\'%s\' is not a valid working copy.' % dir
|
|
filesmeta = os.path.join(dir, store, '_files')
|
|
if not is_package_dir(dir):
|
|
raise oscerr.NoWorkingCopy(msg)
|
|
if not os.path.isfile(filesmeta):
|
|
raise oscerr.NoWorkingCopy('%s (%s does not exist)' % (msg, filesmeta))
|
|
|
|
try:
|
|
r = ET.parse(filesmeta)
|
|
except SyntaxError, e:
|
|
raise oscerr.NoWorkingCopy('%s\nWhen parsing .osc/_files, the following error was encountered:\n%s' % (msg, e))
|
|
return r
|
|
|
|
def store_readlist(dir, name):
|
|
global store
|
|
|
|
r = []
|
|
if os.path.exists(os.path.join(dir, store, name)):
|
|
r = [line.strip() for line in open(os.path.join(dir, store, name), 'r')]
|
|
return r
|
|
|
|
def read_tobeadded(dir):
|
|
return store_readlist(dir, '_to_be_added')
|
|
|
|
def read_tobedeleted(dir):
|
|
return store_readlist(dir, '_to_be_deleted')
|
|
|
|
def read_sizelimit(dir):
|
|
global store
|
|
|
|
r = None
|
|
fname = os.path.join(dir, store, '_size_limit')
|
|
|
|
if os.path.exists(fname):
|
|
r = open(fname).readline().strip()
|
|
|
|
if r is None or not r.isdigit():
|
|
return None
|
|
return int(r)
|
|
|
|
def read_inconflict(dir):
|
|
return store_readlist(dir, '_in_conflict')
|
|
|
|
def parseargs(list_of_args):
|
|
"""Convenience method osc's commandline argument parsing.
|
|
|
|
If called with an empty tuple (or list), return a list containing the current directory.
|
|
Otherwise, return a list of the arguments."""
|
|
if list_of_args:
|
|
return list(list_of_args)
|
|
else:
|
|
return [os.curdir]
|
|
|
|
|
|
def statfrmt(statusletter, filename):
|
|
return '%s %s' % (statusletter, filename)
|
|
|
|
|
|
def pathjoin(a, *p):
|
|
"""Join two or more pathname components, inserting '/' as needed. Cut leading ./"""
|
|
path = os.path.join(a, *p)
|
|
if path.startswith('./'):
|
|
path = path[2:]
|
|
return path
|
|
|
|
|
|
def makeurl(baseurl, l, query=[]):
|
|
"""Given a list of path compoments, construct a complete URL.
|
|
|
|
Optional parameters for a query string can be given as a list, as a
|
|
dictionary, or as an already assembled string.
|
|
In case of a dictionary, the parameters will be urlencoded by this
|
|
function. In case of a list not -- this is to be backwards compatible.
|
|
"""
|
|
|
|
if conf.config['verbose'] > 1:
|
|
print 'makeurl:', baseurl, l, query
|
|
|
|
if type(query) == type(list()):
|
|
query = '&'.join(query)
|
|
elif type(query) == type(dict()):
|
|
query = urlencode(query)
|
|
|
|
scheme, netloc = urlsplit(baseurl)[0:2]
|
|
return urlunsplit((scheme, netloc, '/'.join(l), query, ''))
|
|
|
|
|
|
def http_request(method, url, headers={}, data=None, file=None, timeout=100):
|
|
"""wrapper around urllib2.urlopen for error handling,
|
|
and to support additional (PUT, DELETE) methods"""
|
|
|
|
filefd = None
|
|
|
|
if conf.config['http_debug']:
|
|
print >>sys.stderr, '\n\n--', method, url
|
|
|
|
if method == 'POST' and not file and not data:
|
|
# adding data to an urllib2 request transforms it into a POST
|
|
data = ''
|
|
|
|
req = urllib2.Request(url)
|
|
api_host_options = {}
|
|
if conf.is_known_apiurl(url):
|
|
# ok no external request
|
|
urllib2.install_opener(conf._build_opener(url))
|
|
api_host_options = conf.get_apiurl_api_host_options(url)
|
|
for header, value in api_host_options['http_headers']:
|
|
req.add_header(header, value)
|
|
|
|
req.get_method = lambda: method
|
|
|
|
# POST requests are application/x-www-form-urlencoded per default
|
|
# since we change the request into PUT, we also need to adjust the content type header
|
|
if method == 'PUT' or (method == 'POST' and data):
|
|
req.add_header('Content-Type', 'application/octet-stream')
|
|
|
|
if type(headers) == type({}):
|
|
for i in headers.keys():
|
|
print headers[i]
|
|
req.add_header(i, headers[i])
|
|
|
|
if file and not data:
|
|
size = os.path.getsize(file)
|
|
if size < 1024*512:
|
|
data = open(file, 'rb').read()
|
|
else:
|
|
import mmap
|
|
filefd = open(file, 'rb')
|
|
try:
|
|
if sys.platform[:3] != 'win':
|
|
data = mmap.mmap(filefd.fileno(), os.path.getsize(file), mmap.MAP_SHARED, mmap.PROT_READ)
|
|
else:
|
|
data = mmap.mmap(filefd.fileno(), os.path.getsize(file))
|
|
data = buffer(data)
|
|
except EnvironmentError, e:
|
|
if e.errno == 19:
|
|
sys.exit('\n\n%s\nThe file \'%s\' could not be memory mapped. It is ' \
|
|
'\non a filesystem which does not support this.' % (e, file))
|
|
elif hasattr(e, 'winerror') and e.winerror == 5:
|
|
# falling back to the default io
|
|
data = open(file, 'rb').read()
|
|
else:
|
|
raise
|
|
|
|
if conf.config['debug']: print >>sys.stderr, method, url
|
|
|
|
old_timeout = socket.getdefaulttimeout()
|
|
# XXX: dirty hack as timeout doesn't work with python-m2crypto
|
|
if old_timeout != timeout and not api_host_options.get('sslcertck'):
|
|
socket.setdefaulttimeout(timeout)
|
|
try:
|
|
fd = urllib2.urlopen(req, data=data)
|
|
finally:
|
|
if old_timeout != timeout and not api_host_options.get('sslcertck'):
|
|
socket.setdefaulttimeout(old_timeout)
|
|
if hasattr(conf.cookiejar, 'save'):
|
|
conf.cookiejar.save(ignore_discard=True)
|
|
|
|
if filefd: filefd.close()
|
|
|
|
return fd
|
|
|
|
|
|
def http_GET(*args, **kwargs): return http_request('GET', *args, **kwargs)
|
|
def http_POST(*args, **kwargs): return http_request('POST', *args, **kwargs)
|
|
def http_PUT(*args, **kwargs): return http_request('PUT', *args, **kwargs)
|
|
def http_DELETE(*args, **kwargs): return http_request('DELETE', *args, **kwargs)
|
|
|
|
|
|
def check_store_version(dir):
|
|
global store
|
|
|
|
versionfile = os.path.join(dir, store, '_osclib_version')
|
|
try:
|
|
v = open(versionfile).read().strip()
|
|
except:
|
|
v = ''
|
|
|
|
if v == '':
|
|
msg = 'Error: "%s" is not an osc package working copy.' % os.path.abspath(dir)
|
|
if os.path.exists(os.path.join(dir, '.svn')):
|
|
msg = msg + '\nTry svn instead of osc.'
|
|
raise oscerr.NoWorkingCopy(msg)
|
|
|
|
if v != __store_version__:
|
|
if v in ['0.2', '0.3', '0.4', '0.5', '0.6', '0.7', '0.8', '0.9', '0.95', '0.96', '0.97', '0.98', '0.99']:
|
|
# version is fine, no migration needed
|
|
f = open(versionfile, 'w')
|
|
f.write(__store_version__ + '\n')
|
|
f.close()
|
|
return
|
|
msg = 'The osc metadata of your working copy "%s"' % dir
|
|
msg += '\nhas __store_version__ = %s, but it should be %s' % (v, __store_version__)
|
|
msg += '\nPlease do a fresh checkout or update your client. Sorry about the inconvenience.'
|
|
raise oscerr.WorkingCopyWrongVersion, msg
|
|
|
|
|
|
def meta_get_packagelist(apiurl, prj, deleted=None):
|
|
|
|
query = {}
|
|
if deleted:
|
|
query['deleted'] = 1
|
|
|
|
u = makeurl(apiurl, ['source', prj], query)
|
|
f = http_GET(u)
|
|
root = ET.parse(f).getroot()
|
|
return [ node.get('name') for node in root.findall('entry') ]
|
|
|
|
|
|
def meta_get_filelist(apiurl, prj, package, verbose=False, expand=False, revision=None, meta=False):
|
|
"""return a list of file names,
|
|
or a list File() instances if verbose=True"""
|
|
|
|
query = {}
|
|
if expand:
|
|
query['expand'] = 1
|
|
if meta:
|
|
query['meta'] = 1
|
|
if revision:
|
|
query['rev'] = revision
|
|
else:
|
|
query['rev'] = 'latest'
|
|
|
|
u = makeurl(apiurl, ['source', prj, package], query=query)
|
|
f = http_GET(u)
|
|
root = ET.parse(f).getroot()
|
|
|
|
if not verbose:
|
|
return [ node.get('name') for node in root.findall('entry') ]
|
|
|
|
else:
|
|
l = []
|
|
# rev = int(root.get('rev')) # don't force int. also allow srcmd5 here.
|
|
rev = root.get('rev')
|
|
for node in root.findall('entry'):
|
|
f = File(node.get('name'),
|
|
node.get('md5'),
|
|
int(node.get('size')),
|
|
int(node.get('mtime')))
|
|
f.rev = rev
|
|
l.append(f)
|
|
return l
|
|
|
|
|
|
def meta_get_project_list(apiurl, deleted=None):
|
|
query = {}
|
|
if deleted:
|
|
query['deleted'] = 1
|
|
|
|
u = makeurl(apiurl, ['source'], query)
|
|
f = http_GET(u)
|
|
root = ET.parse(f).getroot()
|
|
return sorted([ node.get('name') for node in root if node.get('name')])
|
|
|
|
|
|
def show_project_meta(apiurl, prj):
|
|
url = makeurl(apiurl, ['source', prj, '_meta'])
|
|
f = http_GET(url)
|
|
return f.readlines()
|
|
|
|
|
|
def show_project_conf(apiurl, prj):
|
|
url = makeurl(apiurl, ['source', prj, '_config'])
|
|
f = http_GET(url)
|
|
return f.readlines()
|
|
|
|
|
|
def show_package_trigger_reason(apiurl, prj, pac, repo, arch):
|
|
url = makeurl(apiurl, ['build', prj, repo, arch, pac, '_reason'])
|
|
try:
|
|
f = http_GET(url)
|
|
return f.read()
|
|
except urllib2.HTTPError, e:
|
|
e.osc_msg = 'Error getting trigger reason for project \'%s\' package \'%s\'' % (prj, pac)
|
|
raise
|
|
|
|
|
|
def show_package_meta(apiurl, prj, pac, meta=False):
|
|
query = {}
|
|
if meta:
|
|
query['meta'] = 1
|
|
|
|
# packages like _pattern and _project do not have a _meta file
|
|
if pac.startswith('_pattern') or pac.startswith('_project'):
|
|
return ""
|
|
|
|
url = makeurl(apiurl, ['source', prj, pac, '_meta'], query)
|
|
try:
|
|
f = http_GET(url)
|
|
return f.readlines()
|
|
except urllib2.HTTPError, e:
|
|
e.osc_msg = 'Error getting meta for project \'%s\' package \'%s\'' % (prj, pac)
|
|
raise
|
|
|
|
|
|
def show_attribute_meta(apiurl, prj, pac, subpac, attribute, with_defaults, with_project):
|
|
path=[]
|
|
path.append('source')
|
|
path.append(prj)
|
|
if pac:
|
|
path.append(pac)
|
|
if pac and subpac:
|
|
path.append(subpac)
|
|
path.append('_attribute')
|
|
if attribute:
|
|
path.append(attribute)
|
|
query=[]
|
|
if with_defaults:
|
|
query.append("with_default=1")
|
|
if with_project:
|
|
query.append("with_project=1")
|
|
url = makeurl(apiurl, path, query)
|
|
try:
|
|
f = http_GET(url)
|
|
return f.readlines()
|
|
except urllib2.HTTPError, e:
|
|
e.osc_msg = 'Error getting meta for project \'%s\' package \'%s\'' % (prj, pac)
|
|
raise
|
|
|
|
|
|
def show_develproject(apiurl, prj, pac, xml_node=False):
|
|
m = show_package_meta(apiurl, prj, pac)
|
|
node = ET.fromstring(''.join(m)).find('devel')
|
|
if not node is None:
|
|
if xml_node:
|
|
return node
|
|
return node.get('project')
|
|
return None
|
|
|
|
|
|
def show_package_disabled_repos(apiurl, prj, pac):
|
|
m = show_package_meta(apiurl, prj, pac)
|
|
#FIXME: don't work if all repos of a project are disabled and only some are enabled since <disable/> is empty
|
|
try:
|
|
root = ET.fromstring(''.join(m))
|
|
elm = root.find('build')
|
|
r = [ node.get('repository') for node in elm.findall('disable')]
|
|
return r
|
|
except:
|
|
return None
|
|
|
|
|
|
def show_pattern_metalist(apiurl, prj):
|
|
url = makeurl(apiurl, ['source', prj, '_pattern'])
|
|
try:
|
|
f = http_GET(url)
|
|
tree = ET.parse(f)
|
|
except urllib2.HTTPError, e:
|
|
e.osc_msg = 'show_pattern_metalist: Error getting pattern list for project \'%s\'' % prj
|
|
raise
|
|
r = [ node.get('name') for node in tree.getroot() ]
|
|
r.sort()
|
|
return r
|
|
|
|
|
|
def show_pattern_meta(apiurl, prj, pattern):
|
|
url = makeurl(apiurl, ['source', prj, '_pattern', pattern])
|
|
try:
|
|
f = http_GET(url)
|
|
return f.readlines()
|
|
except urllib2.HTTPError, e:
|
|
e.osc_msg = 'show_pattern_meta: Error getting pattern \'%s\' for project \'%s\'' % (pattern, prj)
|
|
raise
|
|
|
|
|
|
class metafile:
|
|
"""metafile that can be manipulated and is stored back after manipulation."""
|
|
def __init__(self, url, input, change_is_required=False, file_ext='.xml'):
|
|
import tempfile
|
|
|
|
self.url = url
|
|
self.change_is_required = change_is_required
|
|
(fd, self.filename) = tempfile.mkstemp(prefix = 'osc_metafile.', suffix = file_ext)
|
|
f = os.fdopen(fd, 'w')
|
|
f.write(''.join(input))
|
|
f.close()
|
|
self.hash_orig = dgst(self.filename)
|
|
|
|
def sync(self):
|
|
if self.change_is_required and self.hash_orig == dgst(self.filename):
|
|
print 'File unchanged. Not saving.'
|
|
os.unlink(self.filename)
|
|
return
|
|
|
|
print 'Sending meta data...'
|
|
# don't do any exception handling... it's up to the caller what to do in case
|
|
# of an exception
|
|
http_PUT(self.url, file=self.filename)
|
|
os.unlink(self.filename)
|
|
print 'Done.'
|
|
|
|
def edit(self):
|
|
try:
|
|
while 1:
|
|
run_editor(self.filename)
|
|
try:
|
|
self.sync()
|
|
break
|
|
except urllib2.HTTPError, e:
|
|
error_help = "%d" % e.code
|
|
if e.headers.get('X-Opensuse-Errorcode'):
|
|
error_help = "%s (%d)" % (e.headers.get('X-Opensuse-Errorcode'), e.code)
|
|
|
|
print >>sys.stderr, 'BuildService API error:', error_help
|
|
# examine the error - we can't raise an exception because we might want
|
|
# to try again
|
|
data = e.read()
|
|
if '<summary>' in data:
|
|
print >>sys.stderr, data.split('<summary>')[1].split('</summary>')[0]
|
|
ri = raw_input('Try again? ([y/N]): ')
|
|
if ri not in ['y', 'Y']:
|
|
break
|
|
finally:
|
|
self.discard()
|
|
|
|
def discard(self):
|
|
if os.path.exists(self.filename):
|
|
print 'discarding %s' % self.filename
|
|
os.unlink(self.filename)
|
|
|
|
|
|
# different types of metadata
|
|
metatypes = { 'prj': { 'path': 'source/%s/_meta',
|
|
'template': new_project_templ,
|
|
'file_ext': '.xml'
|
|
},
|
|
'pkg': { 'path' : 'source/%s/%s/_meta',
|
|
'template': new_package_templ,
|
|
'file_ext': '.xml'
|
|
},
|
|
'attribute': { 'path' : 'source/%s/%s/_meta',
|
|
'template': new_attribute_templ,
|
|
'file_ext': '.xml'
|
|
},
|
|
'prjconf': { 'path': 'source/%s/_config',
|
|
'template': '',
|
|
'file_ext': '.txt'
|
|
},
|
|
'user': { 'path': 'person/%s',
|
|
'template': new_user_template,
|
|
'file_ext': '.xml'
|
|
},
|
|
'pattern': { 'path': 'source/%s/_pattern/%s',
|
|
'template': new_pattern_template,
|
|
'file_ext': '.xml'
|
|
},
|
|
}
|
|
|
|
def meta_exists(metatype,
|
|
path_args=None,
|
|
template_args=None,
|
|
create_new=True,
|
|
apiurl=None):
|
|
|
|
global metatypes
|
|
|
|
if not apiurl:
|
|
apiurl = conf.config['apiurl']
|
|
url = make_meta_url(metatype, path_args, apiurl)
|
|
try:
|
|
data = http_GET(url).readlines()
|
|
except urllib2.HTTPError, e:
|
|
if e.code == 404 and create_new:
|
|
data = metatypes[metatype]['template']
|
|
if template_args:
|
|
data = StringIO(data % template_args).readlines()
|
|
else:
|
|
raise e
|
|
|
|
return data
|
|
|
|
def make_meta_url(metatype, path_args=None, apiurl=None, force=False):
|
|
global metatypes
|
|
|
|
if not apiurl:
|
|
apiurl = conf.config['apiurl']
|
|
if metatype not in metatypes.keys():
|
|
raise AttributeError('make_meta_url(): Unknown meta type \'%s\'' % metatype)
|
|
path = metatypes[metatype]['path']
|
|
|
|
if path_args:
|
|
path = path % path_args
|
|
|
|
query = {}
|
|
if force:
|
|
query = { 'force': '1' }
|
|
|
|
return makeurl(apiurl, [path], query)
|
|
|
|
|
|
def edit_meta(metatype,
|
|
path_args=None,
|
|
data=None,
|
|
template_args=None,
|
|
edit=False,
|
|
force=False,
|
|
change_is_required=False,
|
|
apiurl=None):
|
|
|
|
global metatypes
|
|
|
|
if not apiurl:
|
|
apiurl = conf.config['apiurl']
|
|
if not data:
|
|
data = meta_exists(metatype,
|
|
path_args,
|
|
template_args,
|
|
create_new = metatype != 'prjconf', # prjconf always exists, 404 => unknown prj
|
|
apiurl=apiurl)
|
|
|
|
if edit:
|
|
change_is_required = True
|
|
|
|
url = make_meta_url(metatype, path_args, apiurl, force)
|
|
f=metafile(url, data, change_is_required, metatypes[metatype]['file_ext'])
|
|
|
|
if edit:
|
|
f.edit()
|
|
else:
|
|
f.sync()
|
|
|
|
|
|
def show_files_meta(apiurl, prj, pac, revision=None, expand=False, linkrev=None, linkrepair=False, meta=False):
|
|
query = {}
|
|
if revision:
|
|
query['rev'] = revision
|
|
else:
|
|
query['rev'] = 'latest'
|
|
if linkrev:
|
|
query['linkrev'] = linkrev
|
|
elif conf.config['linkcontrol']:
|
|
query['linkrev'] = 'base'
|
|
if meta:
|
|
query['meta'] = 1
|
|
if expand:
|
|
query['expand'] = 1
|
|
if linkrepair:
|
|
query['emptylink'] = 1
|
|
f = http_GET(makeurl(apiurl, ['source', prj, pac], query=query))
|
|
return f.read()
|
|
|
|
def show_upstream_srcmd5(apiurl, prj, pac, expand=False, revision=None, meta=False, include_service_files=False):
|
|
m = show_files_meta(apiurl, prj, pac, expand=expand, revision=revision, meta=meta)
|
|
et = ET.fromstring(''.join(m))
|
|
if include_service_files:
|
|
try:
|
|
if et.find('serviceinfo') and et.find('serviceinfo').get('xsrcmd5'):
|
|
return et.find('serviceinfo').get('xsrcmd5')
|
|
except:
|
|
pass
|
|
return et.get('srcmd5')
|
|
|
|
|
|
def show_upstream_xsrcmd5(apiurl, prj, pac, revision=None, linkrev=None, linkrepair=False, meta=False, include_service_files=False):
|
|
m = show_files_meta(apiurl, prj, pac, revision=revision, linkrev=linkrev, linkrepair=linkrepair, meta=meta, expand=include_service_files)
|
|
et = ET.fromstring(''.join(m))
|
|
if include_service_files:
|
|
return et.get('srcmd5')
|
|
try:
|
|
# only source link packages have a <linkinfo> element.
|
|
li_node = et.find('linkinfo')
|
|
except:
|
|
return None
|
|
|
|
li = Linkinfo()
|
|
li.read(li_node)
|
|
|
|
if li.haserror():
|
|
raise oscerr.LinkExpandError(prj, pac, li.error)
|
|
return li.xsrcmd5
|
|
|
|
|
|
def show_upstream_rev(apiurl, prj, pac, revision=None, expand=False, linkrev=None, meta=False, include_service_files=False):
|
|
m = show_files_meta(apiurl, prj, pac, revision=revision, expand=expand, linkrev=linkrev, meta=meta)
|
|
et = ET.fromstring(''.join(m))
|
|
if include_service_files:
|
|
try:
|
|
return et.find('serviceinfo').get('xsrcmd5')
|
|
except:
|
|
pass
|
|
return et.get('rev')
|
|
|
|
|
|
def read_meta_from_spec(specfile, *args):
|
|
import codecs, re
|
|
"""
|
|
Read tags and sections from spec file. To read out
|
|
a tag the passed argument mustn't end with a colon. To
|
|
read out a section the passed argument must start with
|
|
a '%'.
|
|
This method returns a dictionary which contains the
|
|
requested data.
|
|
"""
|
|
|
|
if not os.path.isfile(specfile):
|
|
raise oscerr.OscIOError(None, '\'%s\' is not a regular file' % specfile)
|
|
|
|
try:
|
|
lines = codecs.open(specfile, 'r', locale.getpreferredencoding()).readlines()
|
|
except UnicodeDecodeError:
|
|
lines = open(specfile).readlines()
|
|
|
|
tags = []
|
|
sections = []
|
|
spec_data = {}
|
|
|
|
for itm in args:
|
|
if itm.startswith('%'):
|
|
sections.append(itm)
|
|
else:
|
|
tags.append(itm)
|
|
|
|
tag_pat = '(?P<tag>^%s)\s*:\s*(?P<val>.*)'
|
|
for tag in tags:
|
|
m = re.compile(tag_pat % tag, re.I | re.M).search(''.join(lines))
|
|
if m and m.group('val'):
|
|
spec_data[tag] = m.group('val').strip()
|
|
|
|
section_pat = '^%s\s*?$'
|
|
for section in sections:
|
|
m = re.compile(section_pat % section, re.I | re.M).search(''.join(lines))
|
|
if m:
|
|
start = lines.index(m.group()+'\n') + 1
|
|
data = []
|
|
for line in lines[start:]:
|
|
if line.startswith('%'):
|
|
break
|
|
data.append(line)
|
|
spec_data[section] = data
|
|
|
|
return spec_data
|
|
|
|
def get_default_editor():
|
|
import platform
|
|
system = platform.system()
|
|
if system == 'Windows':
|
|
return 'notepad'
|
|
if system == 'Linux':
|
|
try:
|
|
# Python 2.6
|
|
dist = platform.linux_distribution()[0]
|
|
except AttributeError:
|
|
dist = platform.dist()[0]
|
|
if dist == 'debian':
|
|
return 'editor'
|
|
elif dist == 'fedora':
|
|
return 'vi'
|
|
return 'vim'
|
|
return 'vi'
|
|
|
|
def get_default_pager():
|
|
import platform
|
|
system = platform.system()
|
|
if system == 'Windows':
|
|
return 'less'
|
|
if system == 'Linux':
|
|
try:
|
|
# Python 2.6
|
|
dist = platform.linux_distribution()[0]
|
|
except AttributeError:
|
|
dist = platform.dist()[0]
|
|
if dist == 'debian':
|
|
return 'pager'
|
|
return 'less'
|
|
return 'more'
|
|
|
|
def run_pager(message, tmp_suffix=''):
|
|
import tempfile, sys
|
|
|
|
if not message:
|
|
return
|
|
|
|
if not sys.stdout.isatty():
|
|
print message
|
|
else:
|
|
tmpfile = tempfile.NamedTemporaryFile(suffix=tmp_suffix)
|
|
tmpfile.write(message)
|
|
tmpfile.flush()
|
|
pager = os.getenv('PAGER', default=get_default_pager())
|
|
try:
|
|
try:
|
|
subprocess.call('%s %s' % (pager, tmpfile.name), shell=True)
|
|
except OSError, e:
|
|
raise oscerr.ExtRuntimeError('cannot run pager \'%s\': %s' % (pager, e.strerror), pager)
|
|
finally:
|
|
tmpfile.close()
|
|
|
|
def run_editor(filename):
|
|
editor = os.getenv('EDITOR', default=get_default_editor())
|
|
cmd = editor.split(' ')
|
|
cmd.append(filename)
|
|
try:
|
|
return subprocess.call(cmd)
|
|
except OSError, e:
|
|
raise oscerr.ExtRuntimeError('cannot run editor \'%s\': %s' % (editor, e.strerror), editor)
|
|
|
|
def edit_message(footer='', template='', templatelen=30):
|
|
delim = '--This line, and those below, will be ignored--\n'
|
|
import tempfile
|
|
(fd, filename) = tempfile.mkstemp(prefix = 'osc-commitmsg', suffix = '.diff')
|
|
f = os.fdopen(fd, 'w')
|
|
if template != '':
|
|
if not templatelen is None:
|
|
lines = template.splitlines()
|
|
template = '\n'.join(lines[:templatelen])
|
|
if lines[templatelen:]:
|
|
footer = '%s\n\n%s' % ('\n'.join(lines[templatelen:]), footer)
|
|
f.write(template)
|
|
f.write('\n')
|
|
f.write(delim)
|
|
f.write('\n')
|
|
f.write(footer)
|
|
f.close()
|
|
|
|
try:
|
|
while 1:
|
|
run_editor(filename)
|
|
msg = open(filename).read().split(delim)[0].rstrip()
|
|
|
|
if len(msg):
|
|
break
|
|
else:
|
|
ri = raw_input('Log message not specified\n'
|
|
'a)bort, c)ontinue, e)dit: ')
|
|
if ri in 'aA':
|
|
raise oscerr.UserAbort()
|
|
elif ri in 'cC':
|
|
break
|
|
elif ri in 'eE':
|
|
pass
|
|
finally:
|
|
os.unlink(filename)
|
|
return msg
|
|
|
|
def clone_request(apiurl, reqid, msg=None):
|
|
query = {'cmd': 'branch', 'request': reqid}
|
|
url = makeurl(apiurl, ['source'], query)
|
|
r = http_POST(url, data=msg)
|
|
root = ET.fromstring(r.read())
|
|
project = None
|
|
for i in root.findall('data'):
|
|
if i.get('name') == 'targetproject':
|
|
project = i.text.strip()
|
|
if not project:
|
|
raise oscerr.APIError('invalid data from clone request:\n%s\n' % ET.tostring(root))
|
|
return project
|
|
|
|
# create a maintenance release request
|
|
def create_release_request(apiurl, src_project, message=''):
|
|
import cgi
|
|
r = Request()
|
|
# api will complete the request
|
|
r.add_action('maintenance_release', src_project=src_project)
|
|
# XXX: clarify why we need the unicode(...) stuff
|
|
r.description = cgi.escape(unicode(message, 'utf8'))
|
|
r.create(apiurl)
|
|
return r
|
|
|
|
# create a maintenance incident per request
|
|
def create_maintenance_request(apiurl, src_project, tgt_project, opt_sourceupdate, message=''):
|
|
import cgi
|
|
r = Request()
|
|
r.add_action('maintenance_incident', src_project=src_project, tgt_project=tgt_project, opt_sourceupdate = opt_sourceupdate)
|
|
# XXX: clarify why we need the unicode(...) stuff
|
|
r.description = cgi.escape(unicode(message, 'utf8'))
|
|
r.create(apiurl)
|
|
return r
|
|
|
|
# This creates an old style submit request for server api 1.0
|
|
def create_submit_request(apiurl,
|
|
src_project, src_package,
|
|
dst_project=None, dst_package=None,
|
|
message="", orev=None, src_update=None):
|
|
|
|
import cgi
|
|
options_block=""
|
|
if src_update:
|
|
options_block="""<options><sourceupdate>%s</sourceupdate></options> """ % (src_update)
|
|
|
|
# Yes, this kind of xml construction is horrible
|
|
targetxml = ""
|
|
if dst_project:
|
|
packagexml = ""
|
|
if dst_package:
|
|
packagexml = """package="%s" """ %( dst_package )
|
|
targetxml = """<target project="%s" %s /> """ %( dst_project, packagexml )
|
|
# XXX: keep the old template for now in order to work with old obs instances
|
|
xml = """\
|
|
<request type="submit">
|
|
<submit>
|
|
<source project="%s" package="%s" rev="%s"/>
|
|
%s
|
|
%s
|
|
</submit>
|
|
<state name="new"/>
|
|
<description>%s</description>
|
|
</request>
|
|
""" % (src_project,
|
|
src_package,
|
|
orev or show_upstream_rev(apiurl, src_project, src_package),
|
|
targetxml,
|
|
options_block,
|
|
cgi.escape(message))
|
|
|
|
# Don't do cgi.escape(unicode(message, "utf8"))) above.
|
|
# Promoting the string to utf8, causes the post to explode with:
|
|
# uncaught exception: Fatal error: Start tag expected, '<' not found at :1.
|
|
# I guess, my original workaround was not that bad.
|
|
|
|
u = makeurl(apiurl, ['request'], query='cmd=create')
|
|
f = http_POST(u, data=xml)
|
|
|
|
root = ET.parse(f).getroot()
|
|
return root.get('id')
|
|
|
|
|
|
def get_request(apiurl, reqid):
|
|
u = makeurl(apiurl, ['request', reqid])
|
|
f = http_GET(u)
|
|
root = ET.parse(f).getroot()
|
|
|
|
r = Request()
|
|
r.read(root)
|
|
return r
|
|
|
|
|
|
def change_review_state(apiurl, reqid, newstate, by_user='', by_group='', by_project='', by_package='', message='', supersed=None):
|
|
query = {'cmd': 'changereviewstate', 'newstate': newstate }
|
|
if by_user:
|
|
query['by_user'] = by_user
|
|
if by_group:
|
|
query['by_group'] = by_group
|
|
if by_project:
|
|
query['by_project'] = by_project
|
|
if by_package:
|
|
query['by_package'] = by_package
|
|
if supersed:
|
|
query['superseded_by'] = supersed
|
|
u = makeurl(apiurl, ['request', reqid], query=query)
|
|
f = http_POST(u, data=message)
|
|
root = ET.parse(f).getroot()
|
|
return root.get('code')
|
|
|
|
def change_request_state(apiurl, reqid, newstate, message='', supersed=None, force=False):
|
|
query={'cmd': 'changestate', 'newstate': newstate }
|
|
if supersed:
|
|
query['superseded_by'] = supersed
|
|
if force:
|
|
query['force'] = "1"
|
|
u = makeurl(apiurl,
|
|
['request', reqid], query=query)
|
|
f = http_POST(u, data=message)
|
|
|
|
r = f.read()
|
|
if r.startswith('<status code="'):
|
|
r = r.split('<status code="')[1]
|
|
r = r.split('" />')[0]
|
|
|
|
return r
|
|
|
|
def change_request_state_template(req, newstate):
|
|
if not len(req.actions):
|
|
return ''
|
|
action = req.actions[0]
|
|
tmpl_name = '%srequest_%s_template' % (action.type, newstate)
|
|
tmpl = conf.config.get(tmpl_name, '')
|
|
tmpl = tmpl.replace('\\t', '\t').replace('\\n', '\n')
|
|
data = {'reqid': req.reqid, 'type': action.type, 'who': req.get_creator()}
|
|
if req.actions[0].type == 'submit':
|
|
data.update({'src_project': action.src_project,
|
|
'src_package': action.src_package, 'src_rev': action.src_rev,
|
|
'dst_project': action.tgt_project, 'dst_package': action.tgt_package,
|
|
'tgt_project': action.tgt_project, 'tgt_package': action.tgt_package})
|
|
try:
|
|
return tmpl % data
|
|
except KeyError, e:
|
|
print >>sys.stderr, 'error: cannot interpolate \'%s\' in \'%s\'' % (e.args[0], tmpl_name)
|
|
return ''
|
|
|
|
def get_review_list(apiurl, project='', package='', byuser='', bygroup='', byproject='', bypackage='', states=('new')):
|
|
xpath = ''
|
|
xpath = xpath_join(xpath, 'state/@name=\'review\'', inner=True)
|
|
if not 'all' in states:
|
|
for state in states:
|
|
xpath = xpath_join(xpath, 'review/@state=\'%s\'' % state, inner=True)
|
|
if byuser:
|
|
xpath = xpath_join(xpath, 'review/@by_user=\'%s\'' % byuser, op='and')
|
|
if bygroup:
|
|
xpath = xpath_join(xpath, 'review/@by_group=\'%s\'' % bygroup, op='and')
|
|
if bypackage:
|
|
xpath = xpath_join(xpath, 'review/[@by_project=\'%s\' and @by_package=\'%s\']' % (byproject, bypackage), op='and')
|
|
elif byproject:
|
|
xpath = xpath_join(xpath, 'review/@by_project=\'%s\'' % byproject, op='and')
|
|
|
|
# XXX: we cannot use the '|' in the xpath expression because it is not supported
|
|
# in the backend
|
|
todo = {}
|
|
if project:
|
|
todo['project'] = project
|
|
if package:
|
|
todo['package'] = package
|
|
for kind, val in todo.iteritems():
|
|
xpath_base = 'action/target/@%(kind)s=\'%(val)s\' or ' \
|
|
'submit/target/@%(kind)s=\'%(val)s\''
|
|
|
|
if conf.config['include_request_from_project']:
|
|
xpath_base = xpath_join(xpath_base, 'action/source/@%(kind)s=\'%(val)s\' or ' \
|
|
'submit/source/@%(kind)s=\'%(val)s\'', op='or', inner=True)
|
|
xpath = xpath_join(xpath, xpath_base % {'kind': kind, 'val': val}, op='and', nexpr_parentheses=True)
|
|
|
|
if conf.config['verbose'] > 1:
|
|
print '[ %s ]' % xpath
|
|
res = search(apiurl, request=xpath)
|
|
collection = res['request']
|
|
requests = []
|
|
for root in collection.findall('request'):
|
|
r = Request()
|
|
r.read(root)
|
|
requests.append(r)
|
|
return requests
|
|
|
|
def get_request_list(apiurl, project='', package='', req_who='', req_state=('new','review',), req_type=None, exclude_target_projects=[]):
|
|
xpath = ''
|
|
if not 'all' in req_state:
|
|
for state in req_state:
|
|
xpath = xpath_join(xpath, 'state/@name=\'%s\'' % state, inner=True)
|
|
if req_who:
|
|
xpath = xpath_join(xpath, '(state/@who=\'%(who)s\' or history/@who=\'%(who)s\')' % {'who': req_who}, op='and')
|
|
|
|
# XXX: we cannot use the '|' in the xpath expression because it is not supported
|
|
# in the backend
|
|
todo = {}
|
|
if project:
|
|
todo['project'] = project
|
|
if package:
|
|
todo['package'] = package
|
|
for kind, val in todo.iteritems():
|
|
xpath_base = 'action/target/@%(kind)s=\'%(val)s\' or ' \
|
|
'submit/target/@%(kind)s=\'%(val)s\''
|
|
|
|
if conf.config['include_request_from_project']:
|
|
xpath_base = xpath_join(xpath_base, 'action/source/@%(kind)s=\'%(val)s\' or ' \
|
|
'submit/source/@%(kind)s=\'%(val)s\'', op='or', inner=True)
|
|
xpath = xpath_join(xpath, xpath_base % {'kind': kind, 'val': val}, op='and', nexpr_parentheses=True)
|
|
|
|
if req_type:
|
|
xpath = xpath_join(xpath, 'action/@type=\'%s\'' % req_type, op='and')
|
|
for i in exclude_target_projects:
|
|
xpath = xpath_join(xpath, '(not(action/target/@project=\'%(prj)s\' or ' \
|
|
'submit/target/@project=\'%(prj)s\'))' % {'prj': i}, op='and')
|
|
|
|
if conf.config['verbose'] > 1:
|
|
print '[ %s ]' % xpath
|
|
res = search(apiurl, request=xpath)
|
|
collection = res['request']
|
|
requests = []
|
|
for root in collection.findall('request'):
|
|
r = Request()
|
|
r.read(root)
|
|
requests.append(r)
|
|
return requests
|
|
|
|
# old style search, this is to be removed
|
|
def get_user_projpkgs_request_list(apiurl, user, req_state=('new','review',), req_type=None, exclude_projects=[], projpkgs={}):
|
|
"""OBSOLETE: user involved request search is supported by OBS 2.2 server side in a better way
|
|
Return all running requests for all projects/packages where is user is involved"""
|
|
if not projpkgs:
|
|
res = get_user_projpkgs(apiurl, user, exclude_projects=exclude_projects)
|
|
projects = []
|
|
for i in res['project_id'].findall('project'):
|
|
projpkgs[i.get('name')] = []
|
|
projects.append(i.get('name'))
|
|
for i in res['package_id'].findall('package'):
|
|
if not i.get('project') in projects:
|
|
projpkgs.setdefault(i.get('project'), []).append(i.get('name'))
|
|
xpath = ''
|
|
for prj, pacs in projpkgs.iteritems():
|
|
if not len(pacs):
|
|
xpath = xpath_join(xpath, 'action/target/@project=\'%s\'' % prj, inner=True)
|
|
else:
|
|
xp = ''
|
|
for p in pacs:
|
|
xp = xpath_join(xp, 'action/target/@package=\'%s\'' % p, inner=True)
|
|
xp = xpath_join(xp, 'action/target/@project=\'%s\'' % prj, op='and')
|
|
xpath = xpath_join(xpath, xp, inner=True)
|
|
if req_type:
|
|
xpath = xpath_join(xpath, 'action/@type=\'%s\'' % req_type, op='and')
|
|
if not 'all' in req_state:
|
|
xp = ''
|
|
for state in req_state:
|
|
xp = xpath_join(xp, 'state/@name=\'%s\'' % state, inner=True)
|
|
xpath = xpath_join(xp, xpath, op='and', nexpr_parentheses=True)
|
|
res = search(apiurl, request=xpath)
|
|
result = []
|
|
for root in res['request'].findall('request'):
|
|
r = Request()
|
|
r.read(root)
|
|
result.append(r)
|
|
return result
|
|
|
|
def get_request_log(apiurl, reqid):
|
|
r = get_request(apiurl, reqid)
|
|
data = []
|
|
frmt = '-' * 76 + '\n%s | %s | %s\n\n%s'
|
|
r.statehistory.reverse()
|
|
# the description of the request is used for the initial log entry
|
|
# otherwise its comment attribute would contain None
|
|
if len(r.statehistory) >= 1:
|
|
r.statehistory[-1].comment = r.description
|
|
else:
|
|
r.state.comment = r.description
|
|
for state in [ r.state ] + r.statehistory:
|
|
s = frmt % (state.name, state.who, state.when, str(state.comment))
|
|
data.append(s)
|
|
return data
|
|
|
|
|
|
def get_group(apiurl, group):
|
|
u = makeurl(apiurl, ['group', quote_plus(group)])
|
|
try:
|
|
f = http_GET(u)
|
|
return ''.join(f.readlines())
|
|
except urllib2.HTTPError:
|
|
print 'user \'%s\' not found' % group
|
|
return None
|
|
|
|
def get_user_meta(apiurl, user):
|
|
u = makeurl(apiurl, ['person', quote_plus(user)])
|
|
try:
|
|
f = http_GET(u)
|
|
return ''.join(f.readlines())
|
|
except urllib2.HTTPError:
|
|
print 'user \'%s\' not found' % user
|
|
return None
|
|
|
|
|
|
def get_user_data(apiurl, user, *tags):
|
|
"""get specified tags from the user meta"""
|
|
meta = get_user_meta(apiurl, user)
|
|
data = []
|
|
if meta != None:
|
|
root = ET.fromstring(meta)
|
|
for tag in tags:
|
|
try:
|
|
if root.find(tag).text != None:
|
|
data.append(root.find(tag).text)
|
|
else:
|
|
# tag is empty
|
|
data.append('-')
|
|
except AttributeError:
|
|
# this part is reached if the tags tuple contains an invalid tag
|
|
print 'The xml file for user \'%s\' seems to be broken' % user
|
|
return []
|
|
return data
|
|
|
|
|
|
def download(url, filename, progress_obj = None, mtime = None):
|
|
import tempfile, shutil
|
|
global BUFSIZE
|
|
|
|
o = None
|
|
try:
|
|
prefix = os.path.basename(filename)
|
|
path = os.path.dirname(filename)
|
|
(fd, tmpfile) = tempfile.mkstemp(dir=path, prefix = prefix, suffix = '.osctmp')
|
|
os.chmod(tmpfile, 0644)
|
|
try:
|
|
o = os.fdopen(fd, 'wb')
|
|
for buf in streamfile(url, http_GET, BUFSIZE, progress_obj=progress_obj):
|
|
o.write(buf)
|
|
o.close()
|
|
os.rename(tmpfile, filename)
|
|
except:
|
|
os.unlink(tmpfile)
|
|
raise
|
|
finally:
|
|
if o is not None:
|
|
o.close()
|
|
|
|
if mtime:
|
|
os.utime(filename, (-1, mtime))
|
|
|
|
def get_source_file(apiurl, prj, package, filename, targetfilename=None, revision=None, progress_obj=None, mtime=None, meta=False):
|
|
targetfilename = targetfilename or filename
|
|
query = {}
|
|
if meta:
|
|
query['rev'] = 1
|
|
if revision:
|
|
query['rev'] = revision
|
|
u = makeurl(apiurl, ['source', prj, package, pathname2url(filename.encode(locale.getpreferredencoding(), 'replace'))], query=query)
|
|
download(u, targetfilename, progress_obj, mtime)
|
|
|
|
def get_binary_file(apiurl, prj, repo, arch,
|
|
filename,
|
|
package = None,
|
|
target_filename = None,
|
|
target_mtime = None,
|
|
progress_meter = False):
|
|
progress_obj = None
|
|
if progress_meter:
|
|
from meter import TextMeter
|
|
progress_obj = TextMeter()
|
|
|
|
target_filename = target_filename or filename
|
|
|
|
where = package or '_repository'
|
|
u = makeurl(apiurl, ['build', prj, repo, arch, where, filename])
|
|
download(u, target_filename, progress_obj, target_mtime)
|
|
|
|
def dgst_from_string(str):
|
|
# Python 2.5 depracates the md5 modules
|
|
# Python 2.4 doesn't have hashlib yet
|
|
try:
|
|
import hashlib
|
|
md5_hash = hashlib.md5()
|
|
except ImportError:
|
|
import md5
|
|
md5_hash = md5.new()
|
|
md5_hash.update(str)
|
|
return md5_hash.hexdigest()
|
|
|
|
def dgst(file):
|
|
|
|
#if not os.path.exists(file):
|
|
#return None
|
|
|
|
global BUFSIZE
|
|
|
|
try:
|
|
import hashlib
|
|
md5 = hashlib
|
|
except ImportError:
|
|
import md5
|
|
md5 = md5
|
|
s = md5.md5()
|
|
f = open(file, 'rb')
|
|
while 1:
|
|
buf = f.read(BUFSIZE)
|
|
if not buf: break
|
|
s.update(buf)
|
|
return s.hexdigest()
|
|
f.close()
|
|
|
|
|
|
def binary(s):
|
|
"""return true if a string is binary data using diff's heuristic"""
|
|
if s and '\0' in s[:4096]:
|
|
return True
|
|
return False
|
|
|
|
|
|
def binary_file(fn):
|
|
"""read 4096 bytes from a file named fn, and call binary() on the data"""
|
|
return binary(open(fn, 'rb').read(4096))
|
|
|
|
|
|
def get_source_file_diff(dir, filename, rev, oldfilename = None, olddir = None, origfilename = None):
|
|
"""
|
|
This methods diffs oldfilename against filename (so filename will
|
|
be shown as the new file).
|
|
The variable origfilename is used if filename and oldfilename differ
|
|
in their names (for instance if a tempfile is used for filename etc.)
|
|
"""
|
|
|
|
import difflib
|
|
|
|
global store
|
|
|
|
if not oldfilename:
|
|
oldfilename = filename
|
|
|
|
if not olddir:
|
|
olddir = os.path.join(dir, store)
|
|
|
|
if not origfilename:
|
|
origfilename = filename
|
|
|
|
file1 = os.path.join(olddir, oldfilename) # old/stored original
|
|
file2 = os.path.join(dir, filename) # working copy
|
|
if binary_file(file1) or binary_file(file2):
|
|
return ['Binary file \'%s\' has changed.\n' % origfilename]
|
|
|
|
f1 = f2 = None
|
|
try:
|
|
f1 = open(file1, 'rb')
|
|
s1 = f1.readlines()
|
|
f1.close()
|
|
|
|
f2 = open(file2, 'rb')
|
|
s2 = f2.readlines()
|
|
f2.close()
|
|
finally:
|
|
if f1:
|
|
f1.close()
|
|
if f2:
|
|
f2.close()
|
|
|
|
d = difflib.unified_diff(s1, s2,
|
|
fromfile = '%s\t(revision %s)' % (origfilename, rev), \
|
|
tofile = '%s\t(working copy)' % origfilename)
|
|
d = list(d)
|
|
# python2.7's difflib slightly changed the format
|
|
# adapt old format to the new format
|
|
if len(d) > 1:
|
|
d[0] = d[0].replace(' \n', '\n')
|
|
d[1] = d[1].replace(' \n', '\n')
|
|
|
|
# if file doesn't end with newline, we need to append one in the diff result
|
|
for i, line in enumerate(d):
|
|
if not line.endswith('\n'):
|
|
d[i] += '\n\\ No newline at end of file'
|
|
if i+1 != len(d):
|
|
d[i] += '\n'
|
|
return d
|
|
|
|
def server_diff(apiurl,
|
|
old_project, old_package, old_revision,
|
|
new_project, new_package, new_revision,
|
|
unified=False, missingok=False, meta=False, expand=True):
|
|
query = {'cmd': 'diff'}
|
|
if expand:
|
|
query['expand'] = 1
|
|
if old_project:
|
|
query['oproject'] = old_project
|
|
if old_package:
|
|
query['opackage'] = old_package
|
|
if old_revision:
|
|
query['orev'] = old_revision
|
|
if new_revision:
|
|
query['rev'] = new_revision
|
|
if unified:
|
|
query['unified'] = 1
|
|
if missingok:
|
|
query['missingok'] = 1
|
|
if meta:
|
|
query['meta'] = 1
|
|
|
|
u = makeurl(apiurl, ['source', new_project, new_package], query=query)
|
|
|
|
f = http_POST(u)
|
|
return f.read()
|
|
|
|
def server_diff_noex(apiurl,
|
|
old_project, old_package, old_revision,
|
|
new_project, new_package, new_revision,
|
|
unified=False, missingok=False, meta=False, expand=True):
|
|
try:
|
|
return server_diff(apiurl,
|
|
old_project, old_package, old_revision,
|
|
new_project, new_package, new_revision,
|
|
unified, missingok, meta, expand)
|
|
except urllib2.HTTPError, e:
|
|
msg = None
|
|
body = None
|
|
try:
|
|
body = e.read()
|
|
if not 'bad link' in body:
|
|
return '# diff failed: ' + body
|
|
except:
|
|
return '# diff failed with unknown error'
|
|
|
|
if expand:
|
|
rdiff = "## diff on expanded link not possible, showing unexpanded version\n"
|
|
try:
|
|
rdiff += server_diff_noex(apiurl,
|
|
old_project, old_package, old_revision,
|
|
new_project, new_package, new_revision,
|
|
unified, missingok, meta, False)
|
|
except:
|
|
elm = ET.fromstring(body).find('summary')
|
|
summary = ''
|
|
if not elm is None:
|
|
summary = elm.text
|
|
return 'error: diffing failed: %s' % summary
|
|
return rdiff
|
|
|
|
|
|
def request_diff(apiurl, reqid):
|
|
u = makeurl(apiurl, ['request', reqid], query={'cmd': 'diff'} )
|
|
|
|
f = http_POST(u)
|
|
return f.read()
|
|
|
|
def submit_action_diff(apiurl, action):
|
|
"""diff a single submit action"""
|
|
# backward compatiblity: only a recent api/backend supports the missingok parameter
|
|
try:
|
|
return server_diff(apiurl, action.tgt_project, action.tgt_package, None,
|
|
action.src_project, action.src_package, action.src_rev, True, True)
|
|
except urllib2.HTTPError, e:
|
|
if e.code == 400:
|
|
try:
|
|
return server_diff(apiurl, action.tgt_project, action.tgt_package, None,
|
|
action.src_project, action.src_package, action.src_rev, True, False)
|
|
except urllib2.HTTPError, e:
|
|
if e.code != 404:
|
|
raise e
|
|
root = ET.fromstring(e.read())
|
|
return 'error: \'%s\' does not exist' % root.find('summary').text
|
|
elif e.code == 404:
|
|
root = ET.fromstring(e.read())
|
|
return 'error: \'%s\' does not exist' % root.find('summary').text
|
|
raise e
|
|
|
|
def make_dir(apiurl, project, package, pathname=None, prj_dir=None, package_tracking=True):
|
|
"""
|
|
creates the plain directory structure for a package dir.
|
|
The 'apiurl' parameter is needed for the project dir initialization.
|
|
The 'project' and 'package' parameters specify the name of the
|
|
project and the package. The optional 'pathname' parameter is used
|
|
for printing out the message that a new dir was created (default: 'prj_dir/package').
|
|
The optional 'prj_dir' parameter specifies the path to the project dir (default: 'project').
|
|
"""
|
|
prj_dir = prj_dir or project
|
|
|
|
# FIXME: carefully test each patch component of prj_dir,
|
|
# if we have a .osc/_files entry at that level.
|
|
# -> if so, we have a package/project clash,
|
|
# and should rename this path component by appending '.proj'
|
|
# and give user a warning message, to discourage such clashes
|
|
|
|
pathname = pathname or getTransActPath(os.path.join(prj_dir, package))
|
|
if is_package_dir(prj_dir):
|
|
# we want this to become a project directory,
|
|
# but it already is a package directory.
|
|
raise oscerr.OscIOError(None, 'checkout_package: package/project clash. Moving myself away not implemented')
|
|
|
|
if not is_project_dir(prj_dir):
|
|
# this directory could exist as a parent direory for one of our earlier
|
|
# checked out sub-projects. in this case, we still need to initialize it.
|
|
print statfrmt('A', prj_dir)
|
|
Project.init_project(apiurl, prj_dir, project, package_tracking)
|
|
|
|
if is_project_dir(os.path.join(prj_dir, package)):
|
|
# the thing exists, but is a project directory and not a package directory
|
|
# FIXME: this should be a warning message to discourage package/project clashes
|
|
raise oscerr.OscIOError(None, 'checkout_package: package/project clash. Moving project away not implemented')
|
|
|
|
if not os.path.exists(os.path.join(prj_dir, package)):
|
|
print statfrmt('A', pathname)
|
|
os.mkdir(os.path.join(prj_dir, package))
|
|
# os.mkdir(os.path.join(prj_dir, package, store))
|
|
|
|
return os.path.join(prj_dir, package)
|
|
|
|
|
|
def checkout_package(apiurl, project, package,
|
|
revision=None, pathname=None, prj_obj=None,
|
|
expand_link=False, prj_dir=None, server_service_files = None, service_files=None, progress_obj=None, size_limit=None, meta=False):
|
|
try:
|
|
# the project we're in might be deleted.
|
|
# that'll throw an error then.
|
|
olddir = os.getcwd()
|
|
except:
|
|
olddir = os.environ.get("PWD")
|
|
|
|
if not prj_dir:
|
|
prj_dir = olddir
|
|
else:
|
|
if sys.platform[:3] == 'win':
|
|
prj_dir = prj_dir[:2] + prj_dir[2:].replace(':', ';')
|
|
else:
|
|
if conf.config['checkout_no_colon']:
|
|
prj_dir = prj_dir.replace(':', '/')
|
|
|
|
root_dots = '.'
|
|
if conf.config['checkout_rooted']:
|
|
if prj_dir[:1] == '/':
|
|
if conf.config['verbose'] > 1:
|
|
print "checkout_rooted ignored for %s" % prj_dir
|
|
# ?? should we complain if not is_project_dir(prj_dir) ??
|
|
else:
|
|
# if we are inside a project or package dir, ascend to parent
|
|
# directories, so that all projects are checked out relative to
|
|
# the same root.
|
|
if is_project_dir(".."):
|
|
# if we are in a package dir, goto parent.
|
|
# Hmm, with 'checkout_no_colon' in effect, we have directory levels that
|
|
# do not easily reveal the fact, that they are part of a project path.
|
|
# At least this test should find that the parent of 'home/username/branches'
|
|
# is a project (hack alert). Also goto parent in this case.
|
|
root_dots = "../"
|
|
elif is_project_dir("../.."):
|
|
# testing two levels is better than one.
|
|
# May happen in case of checkout_no_colon, or
|
|
# if project roots were previously inconsistent
|
|
root_dots = "../../"
|
|
if is_project_dir(root_dots):
|
|
if conf.config['checkout_no_colon']:
|
|
oldproj = store_read_project(root_dots)
|
|
n = len(oldproj.split(':'))
|
|
else:
|
|
n = 1
|
|
root_dots = root_dots + "../" * n
|
|
|
|
if root_dots != '.':
|
|
if conf.config['verbose']:
|
|
print "found root of %s at %s" % (oldproj, root_dots)
|
|
prj_dir = root_dots + prj_dir
|
|
|
|
if not pathname:
|
|
pathname = getTransActPath(os.path.join(prj_dir, package))
|
|
|
|
# before we create directories and stuff, check if the package actually
|
|
# exists
|
|
show_package_meta(apiurl, project, package, meta)
|
|
|
|
isfrozen = False
|
|
if expand_link:
|
|
# try to read from the linkinfo
|
|
# if it is a link we use the xsrcmd5 as the revision to be
|
|
# checked out
|
|
try:
|
|
x = show_upstream_xsrcmd5(apiurl, project, package, revision=revision, meta=meta, include_service_files=server_service_files)
|
|
except:
|
|
x = show_upstream_xsrcmd5(apiurl, project, package, revision=revision, meta=meta, linkrev='base', include_service_files=server_service_files)
|
|
if x:
|
|
isfrozen = True
|
|
if x:
|
|
revision = x
|
|
directory = make_dir(apiurl, project, package, pathname, prj_dir, conf.config['do_package_tracking'])
|
|
p = Package.init_package(apiurl, project, package, directory, size_limit, meta, progress_obj)
|
|
if isfrozen:
|
|
p.mark_frozen()
|
|
if conf.config['do_package_tracking']:
|
|
# check if we can re-use an existing project object
|
|
if prj_obj is None:
|
|
prj_obj = Project(prj_dir)
|
|
prj_obj.set_state(p.name, ' ')
|
|
prj_obj.write_packages()
|
|
p.update(revision, server_service_files, size_limit)
|
|
if service_files:
|
|
print 'Running all source services local'
|
|
p.run_source_services()
|
|
|
|
def replace_pkg_meta(pkgmeta, new_name, new_prj, keep_maintainers = False,
|
|
dst_userid = None, keep_develproject = False):
|
|
"""
|
|
update pkgmeta with new new_name and new_prj and set calling user as the
|
|
only maintainer (unless keep_maintainers is set). Additionally remove the
|
|
develproject entry (<devel />) unless keep_develproject is true.
|
|
"""
|
|
root = ET.fromstring(''.join(pkgmeta))
|
|
root.set('name', new_name)
|
|
root.set('project', new_prj)
|
|
if not keep_maintainers:
|
|
for person in root.findall('person'):
|
|
root.remove(person)
|
|
if not keep_develproject:
|
|
for dp in root.findall('devel'):
|
|
root.remove(dp)
|
|
return ET.tostring(root)
|
|
|
|
def link_to_branch(apiurl, project, package):
|
|
"""
|
|
convert a package with a _link + project.diff to a branch
|
|
"""
|
|
|
|
if '_link' in meta_get_filelist(apiurl, project, package):
|
|
u = makeurl(apiurl, ['source', project, package], 'cmd=linktobranch')
|
|
http_POST(u)
|
|
else:
|
|
raise oscerr.OscIOError(None, 'no _link file inside project \'%s\' package \'%s\'' % (project, package))
|
|
|
|
def link_pac(src_project, src_package, dst_project, dst_package, force, rev='', cicount='', disable_publish = False):
|
|
"""
|
|
create a linked package
|
|
- "src" is the original package
|
|
- "dst" is the "link" package that we are creating here
|
|
"""
|
|
meta_change = False
|
|
dst_meta = ''
|
|
apiurl = conf.config['apiurl']
|
|
try:
|
|
dst_meta = meta_exists(metatype='pkg',
|
|
path_args=(quote_plus(dst_project), quote_plus(dst_package)),
|
|
template_args=None,
|
|
create_new=False, apiurl=apiurl)
|
|
root = ET.fromstring(''.join(dst_meta))
|
|
if root.get('project') != dst_project:
|
|
# The source comes from a different project via a project link, we need to create this instance
|
|
meta_change = True
|
|
except:
|
|
meta_change = True
|
|
|
|
if meta_change:
|
|
src_meta = show_package_meta(apiurl, src_project, src_package)
|
|
dst_meta = replace_pkg_meta(src_meta, dst_package, dst_project)
|
|
|
|
if disable_publish:
|
|
meta_change = True
|
|
root = ET.fromstring(''.join(dst_meta))
|
|
elm = root.find('publish')
|
|
if not elm:
|
|
elm = ET.SubElement(root, 'publish')
|
|
elm.clear()
|
|
ET.SubElement(elm, 'disable')
|
|
dst_meta = ET.tostring(root)
|
|
|
|
if meta_change:
|
|
edit_meta('pkg',
|
|
path_args=(dst_project, dst_package),
|
|
data=dst_meta)
|
|
# create the _link file
|
|
# but first, make sure not to overwrite an existing one
|
|
if '_link' in meta_get_filelist(apiurl, dst_project, dst_package):
|
|
if force:
|
|
print >>sys.stderr, 'forced overwrite of existing _link file'
|
|
else:
|
|
print >>sys.stderr
|
|
print >>sys.stderr, '_link file already exists...! Aborting'
|
|
sys.exit(1)
|
|
|
|
if rev:
|
|
rev = 'rev="%s"' % rev
|
|
else:
|
|
rev = ''
|
|
|
|
if cicount:
|
|
cicount = 'cicount="%s"' % cicount
|
|
else:
|
|
cicount = ''
|
|
|
|
print 'Creating _link...',
|
|
|
|
project = ''
|
|
if src_project != dst_project:
|
|
project = 'project="%s"' % src_project
|
|
|
|
link_template = """\
|
|
<link %s package="%s" %s %s>
|
|
<patches>
|
|
<!-- <apply name="patch" /> apply a patch on the source directory -->
|
|
<!-- <topadd>%%define build_with_feature_x 1</topadd> add a line on the top (spec file only) -->
|
|
<!-- <add>file.patch</add> add a patch to be applied after %%setup (spec file only) -->
|
|
<!-- <delete>filename</delete> delete a file -->
|
|
</patches>
|
|
</link>
|
|
""" % (project, src_package, rev, cicount)
|
|
|
|
u = makeurl(apiurl, ['source', dst_project, dst_package, '_link'])
|
|
http_PUT(u, data=link_template)
|
|
print 'Done.'
|
|
|
|
def aggregate_pac(src_project, src_package, dst_project, dst_package, repo_map = {}, disable_publish = False, nosources = False):
|
|
"""
|
|
aggregate package
|
|
- "src" is the original package
|
|
- "dst" is the "aggregate" package that we are creating here
|
|
- "map" is a dictionary SRC => TARGET repository mappings
|
|
"""
|
|
meta_change = False
|
|
dst_meta = ''
|
|
apiurl = conf.config['apiurl']
|
|
try:
|
|
dst_meta = meta_exists(metatype='pkg',
|
|
path_args=(quote_plus(dst_project), quote_plus(dst_package)),
|
|
template_args=None,
|
|
create_new=False, apiurl=apiurl)
|
|
root = ET.fromstring(''.join(dst_meta))
|
|
if root.get('project') != dst_project:
|
|
# The source comes from a different project via a project link, we need to create this instance
|
|
meta_change = True
|
|
except:
|
|
meta_change = True
|
|
|
|
if meta_change:
|
|
src_meta = show_package_meta(apiurl, src_project, src_package)
|
|
dst_meta = replace_pkg_meta(src_meta, dst_package, dst_project)
|
|
meta_change = True
|
|
|
|
if disable_publish:
|
|
meta_change = True
|
|
root = ET.fromstring(''.join(dst_meta))
|
|
elm = root.find('publish')
|
|
if not elm:
|
|
elm = ET.SubElement(root, 'publish')
|
|
elm.clear()
|
|
ET.SubElement(elm, 'disable')
|
|
dst_meta = ET.tostring(root)
|
|
if meta_change:
|
|
edit_meta('pkg',
|
|
path_args=(dst_project, dst_package),
|
|
data=dst_meta)
|
|
|
|
# create the _aggregate file
|
|
# but first, make sure not to overwrite an existing one
|
|
if '_aggregate' in meta_get_filelist(apiurl, dst_project, dst_package):
|
|
print >>sys.stderr
|
|
print >>sys.stderr, '_aggregate file already exists...! Aborting'
|
|
sys.exit(1)
|
|
|
|
print 'Creating _aggregate...',
|
|
aggregate_template = """\
|
|
<aggregatelist>
|
|
<aggregate project="%s">
|
|
""" % (src_project)
|
|
|
|
aggregate_template += """\
|
|
<package>%s</package>
|
|
""" % ( src_package)
|
|
|
|
if nosources:
|
|
aggregate_template += """\
|
|
<nosources />
|
|
"""
|
|
for src, tgt in repo_map.iteritems():
|
|
aggregate_template += """\
|
|
<repository target="%s" source="%s" />
|
|
""" % (tgt, src)
|
|
|
|
aggregate_template += """\
|
|
</aggregate>
|
|
</aggregatelist>
|
|
"""
|
|
|
|
u = makeurl(apiurl, ['source', dst_project, dst_package, '_aggregate'])
|
|
http_PUT(u, data=aggregate_template)
|
|
print 'Done.'
|
|
|
|
|
|
def attribute_branch_pkg(apiurl, attribute, maintained_update_project_attribute, package, targetproject, return_existing=False, force=False, noaccess=False):
|
|
"""
|
|
Branch packages defined via attributes (via API call)
|
|
"""
|
|
query = { 'cmd': 'branch' }
|
|
query['attribute'] = attribute
|
|
if targetproject:
|
|
query['target_project'] = targetproject
|
|
if force:
|
|
query['force'] = "1"
|
|
if noaccess:
|
|
query['noaccess'] = "1"
|
|
if package:
|
|
query['package'] = package
|
|
if maintained_update_project_attribute:
|
|
query['update_project_attribute'] = maintained_update_project_attribute
|
|
|
|
u = makeurl(apiurl, ['source'], query=query)
|
|
f = None
|
|
try:
|
|
f = http_POST(u)
|
|
except urllib2.HTTPError, e:
|
|
msg = ''.join(e.readlines())
|
|
msg = msg.split('<summary>')[1]
|
|
msg = msg.split('</summary>')[0]
|
|
raise oscerr.APIError(msg)
|
|
|
|
r = f.read()
|
|
r = r.split('targetproject">')[1]
|
|
r = r.split('</data>')[0]
|
|
return r
|
|
|
|
|
|
def branch_pkg(apiurl, src_project, src_package, nodevelproject=False, rev=None, target_project=None, target_package=None, return_existing=False, msg='', force=False, noaccess=False):
|
|
"""
|
|
Branch a package (via API call)
|
|
"""
|
|
query = { 'cmd': 'branch' }
|
|
if nodevelproject:
|
|
query['ignoredevel'] = '1'
|
|
if force:
|
|
query['force'] = '1'
|
|
if noaccess:
|
|
query['noaccess'] = '1'
|
|
if rev:
|
|
query['rev'] = rev
|
|
if target_project:
|
|
query['target_project'] = target_project
|
|
if target_package:
|
|
query['target_package'] = target_package
|
|
if msg:
|
|
query['comment'] = msg
|
|
u = makeurl(apiurl, ['source', src_project, src_package], query=query)
|
|
try:
|
|
f = http_POST(u)
|
|
except urllib2.HTTPError, e:
|
|
if not return_existing:
|
|
raise
|
|
root = ET.fromstring(e.read())
|
|
summary = root.find('summary')
|
|
if summary is None:
|
|
raise oscerr.APIError('unexpected response:\n%s' % ET.tostring(root))
|
|
m = re.match(r"branch target package already exists: (\S+)/(\S+)", summary.text)
|
|
if not m:
|
|
e.msg += '\n' + summary.text
|
|
raise
|
|
return (True, m.group(1), m.group(2), None, None)
|
|
|
|
data = {}
|
|
for i in ET.fromstring(f.read()).findall('data'):
|
|
data[i.get('name')] = i.text
|
|
return (False, data.get('targetproject', None), data.get('targetpackage', None),
|
|
data.get('sourceproject', None), data.get('sourcepackage', None))
|
|
|
|
|
|
def copy_pac(src_apiurl, src_project, src_package,
|
|
dst_apiurl, dst_project, dst_package,
|
|
client_side_copy = False,
|
|
keep_maintainers = False,
|
|
keep_develproject = False,
|
|
expand = False,
|
|
revision = None,
|
|
comment = None):
|
|
"""
|
|
Create a copy of a package.
|
|
|
|
Copying can be done by downloading the files from one package and commit
|
|
them into the other by uploading them (client-side copy) --
|
|
or by the server, in a single api call.
|
|
"""
|
|
|
|
if not (src_apiurl == dst_apiurl and src_project == dst_project \
|
|
and src_package == dst_package):
|
|
src_meta = show_package_meta(src_apiurl, src_project, src_package)
|
|
dst_userid = conf.get_apiurl_usr(dst_apiurl)
|
|
src_meta = replace_pkg_meta(src_meta, dst_package, dst_project, keep_maintainers,
|
|
dst_userid, keep_develproject)
|
|
|
|
print 'Sending meta data...'
|
|
u = makeurl(dst_apiurl, ['source', dst_project, dst_package, '_meta'])
|
|
http_PUT(u, data=src_meta)
|
|
|
|
print 'Copying files...'
|
|
if not client_side_copy:
|
|
query = {'cmd': 'copy', 'oproject': src_project, 'opackage': src_package }
|
|
if expand:
|
|
query['expand'] = '1'
|
|
if revision:
|
|
query['orev'] = revision
|
|
if comment:
|
|
query['comment'] = comment
|
|
u = makeurl(dst_apiurl, ['source', dst_project, dst_package], query=query)
|
|
f = http_POST(u)
|
|
return f.read()
|
|
|
|
else:
|
|
# copy one file after the other
|
|
import tempfile
|
|
query = {'rev': 'upload'}
|
|
revision = show_upstream_srcmd5(src_apiurl, src_project, src_package, expand=expand, revision=revision)
|
|
for n in meta_get_filelist(src_apiurl, src_project, src_package, expand=expand, revision=revision):
|
|
if n.startswith('_service:') or n.startswith('_service_'):
|
|
continue
|
|
print ' ', n
|
|
tmpfile = None
|
|
try:
|
|
(fd, tmpfile) = tempfile.mkstemp(prefix='osc-copypac')
|
|
get_source_file(src_apiurl, src_project, src_package, n, targetfilename=tmpfile, revision=revision)
|
|
u = makeurl(dst_apiurl, ['source', dst_project, dst_package, pathname2url(n)], query=query)
|
|
http_PUT(u, file = tmpfile)
|
|
finally:
|
|
if not tmpfile is None:
|
|
os.unlink(tmpfile)
|
|
if comment:
|
|
query['comment'] = comment
|
|
query['cmd'] = 'commit'
|
|
u = makeurl(dst_apiurl, ['source', dst_project, dst_package], query=query)
|
|
http_POST(u)
|
|
return 'Done.'
|
|
|
|
|
|
def undelete_package(apiurl, prj, pac, msg=None):
|
|
query={'cmd': 'undelete'}
|
|
if msg:
|
|
query['comment'] = msg
|
|
else:
|
|
query['comment'] = 'undeleted via osc'
|
|
u = makeurl(apiurl, ['source', prj, pac], query)
|
|
http_POST(u)
|
|
|
|
def undelete_project(apiurl, prj, msg=None):
|
|
query={'cmd': 'undelete'}
|
|
if msg:
|
|
query['comment'] = msg
|
|
else:
|
|
query['comment'] = 'undeleted via osc'
|
|
u = makeurl(apiurl, ['source', prj], query)
|
|
http_POST(u)
|
|
|
|
|
|
def delete_package(apiurl, prj, pac, force=False, msg=None):
|
|
query = {}
|
|
if force:
|
|
query['force'] = "1"
|
|
u = makeurl(apiurl, ['source', prj, pac], query)
|
|
http_DELETE(u)
|
|
|
|
def delete_project(apiurl, prj, force=False, msg=None):
|
|
query = {}
|
|
if force:
|
|
query['force'] = "1"
|
|
if msg:
|
|
query['comment'] = msg
|
|
u = makeurl(apiurl, ['source', prj], query)
|
|
http_DELETE(u)
|
|
|
|
def delete_files(apiurl, prj, pac, files):
|
|
for filename in files:
|
|
u = makeurl(apiurl, ['source', prj, pac, filename], query={'comment': 'removed %s' % (filename, )})
|
|
http_DELETE(u)
|
|
|
|
|
|
# old compat lib call
|
|
def get_platforms(apiurl):
|
|
return get_repositories(apiurl)
|
|
|
|
def get_repositories(apiurl):
|
|
f = http_GET(makeurl(apiurl, ['platform']))
|
|
tree = ET.parse(f)
|
|
r = [ node.get('name') for node in tree.getroot() ]
|
|
r.sort()
|
|
return r
|
|
|
|
|
|
def get_distibutions(apiurl, discon=False):
|
|
r = []
|
|
|
|
# FIXME: this is just a naming convention on api.opensuse.org, but not a general valid apparoach
|
|
if discon:
|
|
result_line_templ = '%(name)-25s %(project)s'
|
|
f = http_GET(makeurl(apiurl, ['build']))
|
|
root = ET.fromstring(''.join(f))
|
|
|
|
for node in root.findall('entry'):
|
|
if node.get('name').startswith('DISCONTINUED:'):
|
|
rmap = {}
|
|
rmap['name'] = node.get('name').replace('DISCONTINUED:','').replace(':', ' ')
|
|
rmap['project'] = node.get('name')
|
|
r.append (result_line_templ % rmap)
|
|
|
|
r.insert(0,'distribution project')
|
|
r.insert(1,'------------ -------')
|
|
|
|
else:
|
|
result_line_templ = '%(name)-25s %(project)-25s %(repository)-25s %(reponame)s'
|
|
f = http_GET(makeurl(apiurl, ['distributions']))
|
|
root = ET.fromstring(''.join(f))
|
|
|
|
for node in root.findall('distribution'):
|
|
rmap = {}
|
|
for node2 in node.findall('name'):
|
|
rmap['name'] = node2.text
|
|
for node3 in node.findall('project'):
|
|
rmap['project'] = node3.text
|
|
for node4 in node.findall('repository'):
|
|
rmap['repository'] = node4.text
|
|
for node5 in node.findall('reponame'):
|
|
rmap['reponame'] = node5.text
|
|
r.append(result_line_templ % rmap)
|
|
|
|
r.insert(0,'distribution project repository reponame')
|
|
r.insert(1,'------------ ------- ---------- --------')
|
|
|
|
return r
|
|
|
|
|
|
# old compat lib call
|
|
def get_platforms_of_project(apiurl, prj):
|
|
return get_repositories_of_project(apiurl, prj)
|
|
|
|
def get_repositories_of_project(apiurl, prj):
|
|
f = show_project_meta(apiurl, prj)
|
|
root = ET.fromstring(''.join(f))
|
|
|
|
r = [ node.get('name') for node in root.findall('repository')]
|
|
return r
|
|
|
|
|
|
class Repo:
|
|
repo_line_templ = '%-15s %-10s'
|
|
|
|
def __init__(self, name, arch):
|
|
self.name = name
|
|
self.arch = arch
|
|
|
|
def __str__(self):
|
|
return self.repo_line_templ % (self.name, self.arch)
|
|
|
|
def get_repos_of_project(apiurl, prj):
|
|
f = show_project_meta(apiurl, prj)
|
|
root = ET.fromstring(''.join(f))
|
|
|
|
for node in root.findall('repository'):
|
|
for node2 in node.findall('arch'):
|
|
yield Repo(node.get('name'), node2.text)
|
|
|
|
def get_binarylist(apiurl, prj, repo, arch, package=None, verbose=False):
|
|
what = package or '_repository'
|
|
u = makeurl(apiurl, ['build', prj, repo, arch, what])
|
|
f = http_GET(u)
|
|
tree = ET.parse(f)
|
|
if not verbose:
|
|
return [ node.get('filename') for node in tree.findall('binary')]
|
|
else:
|
|
l = []
|
|
for node in tree.findall('binary'):
|
|
f = File(node.get('filename'),
|
|
None,
|
|
int(node.get('size')),
|
|
int(node.get('mtime')))
|
|
l.append(f)
|
|
return l
|
|
|
|
|
|
def get_binarylist_published(apiurl, prj, repo, arch):
|
|
u = makeurl(apiurl, ['published', prj, repo, arch])
|
|
f = http_GET(u)
|
|
tree = ET.parse(f)
|
|
r = [ node.get('name') for node in tree.findall('entry')]
|
|
return r
|
|
|
|
|
|
def show_results_meta(apiurl, prj, package=None, lastbuild=None, repository=[], arch=[]):
|
|
query = {}
|
|
if package:
|
|
query['package'] = package
|
|
if lastbuild:
|
|
query['lastbuild'] = 1
|
|
u = makeurl(apiurl, ['build', prj, '_result'], query=query)
|
|
for repo in repository:
|
|
u = u + '&repository=%s' % repo
|
|
for a in arch:
|
|
u = u + '&arch=%s' % a
|
|
f = http_GET(u)
|
|
return f.readlines()
|
|
|
|
|
|
def show_prj_results_meta(apiurl, prj):
|
|
u = makeurl(apiurl, ['build', prj, '_result'])
|
|
f = http_GET(u)
|
|
return f.readlines()
|
|
|
|
|
|
def get_package_results(apiurl, prj, package, lastbuild=None, repository=[], arch=[]):
|
|
""" return a package results as a list of dicts """
|
|
r = []
|
|
|
|
f = show_results_meta(apiurl, prj, package, lastbuild, repository, arch)
|
|
root = ET.fromstring(''.join(f))
|
|
|
|
for node in root.findall('result'):
|
|
rmap = {}
|
|
rmap['project'] = rmap['prj'] = prj
|
|
rmap['pkg'] = rmap['package'] = rmap['pac'] = package
|
|
rmap['repository'] = rmap['repo'] = rmap['rep'] = node.get('repository')
|
|
rmap['arch'] = node.get('arch')
|
|
rmap['state'] = node.get('state')
|
|
rmap['dirty'] = node.get('dirty')
|
|
|
|
rmap['details'] = ''
|
|
statusnode = node.find('status')
|
|
if statusnode != None:
|
|
rmap['code'] = statusnode.get('code', '')
|
|
else:
|
|
rmap['code'] = ''
|
|
|
|
if rmap['code'] in ('unresolvable', 'expansion error', 'broken', 'blocked', 'finished'):
|
|
details = statusnode.find('details')
|
|
if details != None:
|
|
rmap['details'] = details.text
|
|
|
|
rmap['dirty'] = rmap['dirty'] == 'true'
|
|
|
|
r.append(rmap)
|
|
return r
|
|
|
|
def format_results(results, format):
|
|
"""apply selected format on each dict in results and return it as a list of strings"""
|
|
return [format % r for r in results]
|
|
|
|
def get_results(apiurl, prj, package, lastbuild=None, repository=[], arch=[], verbose=False):
|
|
r = []
|
|
result_line_templ = '%(rep)-20s %(arch)-10s %(status)s'
|
|
|
|
for res in get_package_results(apiurl, prj, package, lastbuild, repository, arch):
|
|
res['status'] = res['code']
|
|
if verbose and res['details'] != '':
|
|
if res['status'] in ('unresolvable', 'expansion error'):
|
|
lines = res['details'].split(',')
|
|
res['status'] += ': ' + '\n '.join(lines)
|
|
|
|
else:
|
|
res['status'] += ': %s' % (res['details'], )
|
|
if res['dirty']:
|
|
if verbose:
|
|
res['status'] = 'outdated (was: %s)' % res['status']
|
|
else:
|
|
res['status'] += '*'
|
|
|
|
r.append(result_line_templ % res)
|
|
|
|
return r
|
|
|
|
def get_prj_results(apiurl, prj, hide_legend=False, csv=False, status_filter=None, name_filter=None, arch=None, repo=None, vertical=None, show_excluded=None):
|
|
#print '----------------------------------------'
|
|
global buildstatus_symbols
|
|
|
|
r = []
|
|
|
|
f = show_prj_results_meta(apiurl, prj)
|
|
root = ET.fromstring(''.join(f))
|
|
|
|
pacs = []
|
|
# sequence of (repo,arch) tuples
|
|
targets = []
|
|
# {package: {(repo,arch): status}}
|
|
status = {}
|
|
if root.find('result') == None:
|
|
return []
|
|
for results in root.findall('result'):
|
|
for node in results:
|
|
pacs.append(node.get('package'))
|
|
pacs = sorted(list(set(pacs)))
|
|
for node in root.findall('result'):
|
|
# filter architecture and repository
|
|
if arch != None and node.get('arch') not in arch:
|
|
continue
|
|
if repo != None and node.get('repository') not in repo:
|
|
continue
|
|
if node.get('dirty') == "true":
|
|
state = "outdated"
|
|
else:
|
|
state = node.get('state')
|
|
tg = (node.get('repository'), node.get('arch'), state)
|
|
targets.append(tg)
|
|
for pacnode in node.findall('status'):
|
|
pac = pacnode.get('package')
|
|
if pac not in status:
|
|
status[pac] = {}
|
|
status[pac][tg] = pacnode.get('code')
|
|
targets.sort()
|
|
|
|
# filter option
|
|
if status_filter or name_filter or not show_excluded:
|
|
|
|
pacs_to_show = []
|
|
targets_to_show = []
|
|
|
|
#filtering for Package Status
|
|
if status_filter:
|
|
if status_filter in buildstatus_symbols.values():
|
|
for txt, sym in buildstatus_symbols.items():
|
|
if sym == status_filter:
|
|
filt_txt = txt
|
|
for pkg in status.keys():
|
|
for repo in status[pkg].keys():
|
|
if status[pkg][repo] == filt_txt:
|
|
if not name_filter:
|
|
pacs_to_show.append(pkg)
|
|
targets_to_show.append(repo)
|
|
elif name_filter in pkg:
|
|
pacs_to_show.append(pkg)
|
|
|
|
#filtering for Package Name
|
|
elif name_filter:
|
|
for pkg in pacs:
|
|
if name_filter in pkg:
|
|
pacs_to_show.append(pkg)
|
|
|
|
#filter non building states
|
|
elif not show_excluded:
|
|
enabled = {}
|
|
for pkg in status.keys():
|
|
showpkg = False
|
|
for repo in status[pkg].keys():
|
|
if status[pkg][repo] != "excluded":
|
|
enabled[repo] = 1
|
|
showpkg = True
|
|
|
|
if showpkg:
|
|
pacs_to_show.append(pkg)
|
|
|
|
targets_to_show = enabled.keys()
|
|
|
|
pacs = [ i for i in pacs if i in pacs_to_show ]
|
|
if len(targets_to_show):
|
|
targets = [ i for i in targets if i in targets_to_show ]
|
|
|
|
# csv output
|
|
if csv:
|
|
# TODO: option to disable the table header
|
|
row = ['_'] + ['/'.join(tg) for tg in targets]
|
|
r.append(';'.join(row))
|
|
for pac in pacs:
|
|
row = [pac] + [status[pac][tg] for tg in targets]
|
|
r.append(';'.join(row))
|
|
return r
|
|
|
|
if not vertical:
|
|
# human readable output
|
|
max_pacs = 40
|
|
for startpac in range(0, len(pacs), max_pacs):
|
|
offset = 0
|
|
for pac in pacs[startpac:startpac+max_pacs]:
|
|
r.append(' |' * offset + ' ' + pac)
|
|
offset += 1
|
|
|
|
for tg in targets:
|
|
line = []
|
|
line.append(' ')
|
|
for pac in pacs[startpac:startpac+max_pacs]:
|
|
st = ''
|
|
if not status.has_key(pac) or not status[pac].has_key(tg):
|
|
# for newly added packages, status may be missing
|
|
st = '?'
|
|
else:
|
|
try:
|
|
st = buildstatus_symbols[status[pac][tg]]
|
|
except:
|
|
print 'osc: warn: unknown status \'%s\'...' % status[pac][tg]
|
|
print 'please edit osc/core.py, and extend the buildstatus_symbols dictionary.'
|
|
st = '?'
|
|
buildstatus_symbols[status[pac][tg]] = '?'
|
|
line.append(st)
|
|
line.append(' ')
|
|
line.append(' %s %s (%s)' % tg)
|
|
line = ''.join(line)
|
|
|
|
r.append(line)
|
|
|
|
r.append('')
|
|
else:
|
|
offset = 0
|
|
for tg in targets:
|
|
r.append('| ' * offset + '%s %s (%s)'%tg )
|
|
offset += 1
|
|
|
|
for pac in pacs:
|
|
line = []
|
|
for tg in targets:
|
|
st = ''
|
|
if not status.has_key(pac) or not status[pac].has_key(tg):
|
|
# for newly added packages, status may be missing
|
|
st = '?'
|
|
else:
|
|
try:
|
|
st = buildstatus_symbols[status[pac][tg]]
|
|
except:
|
|
print 'osc: warn: unknown status \'%s\'...' % status[pac][tg]
|
|
print 'please edit osc/core.py, and extend the buildstatus_symbols dictionary.'
|
|
st = '?'
|
|
buildstatus_symbols[status[pac][tg]] = '?'
|
|
line.append(st)
|
|
line.append(' '+pac)
|
|
r.append(' '.join(line))
|
|
|
|
line = []
|
|
for i in range(0, len(targets)):
|
|
line.append(str(i%10))
|
|
r.append(' '.join(line))
|
|
|
|
r.append('')
|
|
|
|
if not hide_legend and len(pacs):
|
|
r.append(' Legend:')
|
|
legend = []
|
|
for i, j in buildstatus_symbols.items():
|
|
if i == "expansion error":
|
|
continue
|
|
legend.append('%3s %-20s' % (j, i))
|
|
legend.append(' ? buildstatus not available (only new packages)')
|
|
|
|
if vertical:
|
|
for i in range(0, len(targets)):
|
|
s = '%1d %s %s (%s)' % (i%10, targets[i][0], targets[i][1], targets[i][2])
|
|
if i < len(legend):
|
|
legend[i] += s
|
|
else:
|
|
legend.append(' '*24 + s)
|
|
|
|
r += legend
|
|
|
|
return r
|
|
|
|
|
|
def streamfile(url, http_meth = http_GET, bufsize=8192, data=None, progress_obj=None, text=None):
|
|
"""
|
|
performs http_meth on url and read bufsize bytes from the response
|
|
until EOF is reached. After each read bufsize bytes are yielded to the
|
|
caller.
|
|
"""
|
|
cl = ''
|
|
retries = 0
|
|
# Repeat requests until we get reasonable Content-Length header
|
|
# Server (or iChain) is corrupting data at some point, see bnc#656281
|
|
while cl == '':
|
|
if retries >= int(conf.config['http_retries']):
|
|
raise oscerr.OscIOError(None, 'Content-Length is empty for %s, protocol violation' % url)
|
|
retries = retries + 1
|
|
if retries > 1 and conf.config['http_debug']:
|
|
print >>sys.stderr, '\n\nRetry %d --' % (retries - 1), url
|
|
f = http_meth.__call__(url, data = data)
|
|
cl = f.info().get('Content-Length')
|
|
|
|
if cl is not None:
|
|
cl = int(cl)
|
|
|
|
if progress_obj:
|
|
import urlparse
|
|
basename = os.path.basename(urlparse.urlsplit(url)[2])
|
|
progress_obj.start(basename=basename, text=text, size=cl)
|
|
data = f.read(bufsize)
|
|
read = len(data)
|
|
while len(data):
|
|
if progress_obj:
|
|
progress_obj.update(read)
|
|
yield data
|
|
data = f.read(bufsize)
|
|
read += len(data)
|
|
if progress_obj:
|
|
progress_obj.end(read)
|
|
f.close()
|
|
|
|
if not cl is None and read != cl:
|
|
raise oscerr.OscIOError(None, 'Content-Length is not matching file size for %s: %i vs %i file size' % (url, cl, read))
|
|
|
|
|
|
def print_buildlog(apiurl, prj, package, repository, arch, offset = 0):
|
|
"""prints out the buildlog on stdout"""
|
|
query = {'nostream' : '1', 'start' : '%s' % offset}
|
|
while True:
|
|
query['start'] = offset
|
|
start_offset = offset
|
|
u = makeurl(apiurl, ['build', prj, repository, arch, package, '_log'], query=query)
|
|
for data in streamfile(u):
|
|
offset += len(data)
|
|
sys.stdout.write(data)
|
|
if start_offset == offset:
|
|
break
|
|
|
|
def get_dependson(apiurl, project, repository, arch, packages=None, reverse=None):
|
|
query = []
|
|
if packages:
|
|
for i in packages:
|
|
query.append('package=%s' % quote_plus(i))
|
|
|
|
if reverse:
|
|
query.append('view=revpkgnames')
|
|
else:
|
|
query.append('view=pkgnames')
|
|
|
|
u = makeurl(apiurl, ['build', project, repository, arch, '_builddepinfo'], query=query)
|
|
f = http_GET(u)
|
|
return f.read()
|
|
|
|
def get_buildinfo(apiurl, prj, package, repository, arch, specfile=None, addlist=None):
|
|
query = []
|
|
if addlist:
|
|
for i in addlist:
|
|
query.append('add=%s' % quote_plus(i))
|
|
|
|
u = makeurl(apiurl, ['build', prj, repository, arch, package, '_buildinfo'], query=query)
|
|
|
|
if specfile:
|
|
f = http_POST(u, data=specfile)
|
|
else:
|
|
f = http_GET(u)
|
|
return f.read()
|
|
|
|
|
|
def get_buildconfig(apiurl, prj, repository):
|
|
u = makeurl(apiurl, ['build', prj, repository, '_buildconfig'])
|
|
f = http_GET(u)
|
|
return f.read()
|
|
|
|
|
|
def get_source_rev(apiurl, project, package, revision=None):
|
|
# API supports ?deleted=1&meta=1&rev=4
|
|
# but not rev=current,rev=latest,rev=top, or anything like this.
|
|
# CAUTION: We have to loop through all rev and find the highest one, if none given.
|
|
|
|
if revision:
|
|
url = makeurl(apiurl, ['source', project, package, '_history'], {'rev':revision})
|
|
else:
|
|
url = makeurl(apiurl, ['source', project, package, '_history'])
|
|
f = http_GET(url)
|
|
xml = ET.parse(f)
|
|
ent = None
|
|
for new in xml.findall('revision'):
|
|
# remember the newest one.
|
|
if not ent:
|
|
ent = new
|
|
elif ent.find('time').text < new.find('time').text:
|
|
ent = new
|
|
if not ent:
|
|
return { 'version': None, 'error':'empty revisionlist: no such package?' }
|
|
e = {}
|
|
for k in ent.keys():
|
|
e[k] = ent.get(k)
|
|
for k in list(ent):
|
|
e[k.tag] = k.text
|
|
return e
|
|
|
|
def get_buildhistory(apiurl, prj, package, repository, arch, format = 'text'):
|
|
import time
|
|
u = makeurl(apiurl, ['build', prj, repository, arch, package, '_history'])
|
|
f = http_GET(u)
|
|
root = ET.parse(f).getroot()
|
|
|
|
r = []
|
|
for node in root.findall('entry'):
|
|
rev = int(node.get('rev'))
|
|
srcmd5 = node.get('srcmd5')
|
|
versrel = node.get('versrel')
|
|
bcnt = int(node.get('bcnt'))
|
|
t = time.localtime(int(node.get('time')))
|
|
t = time.strftime('%Y-%m-%d %H:%M:%S', t)
|
|
|
|
if format == 'csv':
|
|
r.append('%s|%s|%d|%s.%d' % (t, srcmd5, rev, versrel, bcnt))
|
|
else:
|
|
r.append('%s %s %6d %s.%d' % (t, srcmd5, rev, versrel, bcnt))
|
|
|
|
if format == 'text':
|
|
r.insert(0, 'time srcmd5 rev vers-rel.bcnt')
|
|
|
|
return r
|
|
|
|
def print_jobhistory(apiurl, prj, current_package, repository, arch, format = 'text', limit=20):
|
|
import time
|
|
query = {}
|
|
if current_package:
|
|
query['package'] = current_package
|
|
if limit != None and int(limit) > 0:
|
|
query['limit'] = int(limit)
|
|
u = makeurl(apiurl, ['build', prj, repository, arch, '_jobhistory'], query )
|
|
f = http_GET(u)
|
|
root = ET.parse(f).getroot()
|
|
|
|
if format == 'text':
|
|
print "time package reason code build time worker"
|
|
for node in root.findall('jobhist'):
|
|
package = node.get('package')
|
|
worker = node.get('workerid')
|
|
reason = node.get('reason')
|
|
if not reason:
|
|
reason = "unknown"
|
|
code = node.get('code')
|
|
rt = int(node.get('readytime'))
|
|
readyt = time.localtime(rt)
|
|
readyt = time.strftime('%Y-%m-%d %H:%M:%S', readyt)
|
|
st = int(node.get('starttime'))
|
|
et = int(node.get('endtime'))
|
|
endtime = time.strftime('%Y-%m-%d %H:%M:%S', time.localtime(et))
|
|
waittm = time.gmtime(et-st)
|
|
if waittm.tm_hour:
|
|
waitbuild = "%2dh %2dm %2ds" % (waittm.tm_hour, waittm.tm_min, waittm.tm_sec)
|
|
else:
|
|
waitbuild = " %2dm %2ds" % (waittm.tm_min, waittm.tm_sec)
|
|
|
|
if format == 'csv':
|
|
print '%s|%s|%s|%s|%s|%s' % (endtime, package, reason, code, waitbuild, worker)
|
|
else:
|
|
print '%s %-50s %-16s %-16s %-16s %-16s' % (endtime, package[0:49], reason[0:15], code[0:15], waitbuild, worker)
|
|
|
|
|
|
def get_commitlog(apiurl, prj, package, revision, format = 'text', meta = False, deleted = False):
|
|
import time
|
|
|
|
query = {}
|
|
if deleted:
|
|
query['deleted'] = 1
|
|
if meta:
|
|
query['meta'] = 1
|
|
|
|
u = makeurl(apiurl, ['source', prj, package, '_history'], query)
|
|
f = http_GET(u)
|
|
root = ET.parse(f).getroot()
|
|
|
|
r = []
|
|
if format == 'xml':
|
|
r.append('<?xml version="1.0"?>')
|
|
r.append('<log>')
|
|
revisions = root.findall('revision')
|
|
revisions.reverse()
|
|
for node in revisions:
|
|
srcmd5 = node.find('srcmd5').text
|
|
try:
|
|
rev = int(node.get('rev'))
|
|
#vrev = int(node.get('vrev')) # what is the meaning of vrev?
|
|
try:
|
|
if revision and rev != int(revision):
|
|
continue
|
|
except ValueError:
|
|
if revision != srcmd5:
|
|
continue
|
|
except ValueError:
|
|
# this part should _never_ be reached but...
|
|
return [ 'an unexpected error occured - please file a bug' ]
|
|
version = node.find('version').text
|
|
user = node.find('user').text
|
|
try:
|
|
comment = node.find('comment').text.encode(locale.getpreferredencoding(), 'replace')
|
|
except:
|
|
comment = '<no message>'
|
|
try:
|
|
requestid = node.find('requestid').text.encode(locale.getpreferredencoding(), 'replace')
|
|
except:
|
|
requestid = ""
|
|
t = time.localtime(int(node.find('time').text))
|
|
t = time.strftime('%Y-%m-%d %H:%M:%S', t)
|
|
|
|
if format == 'csv':
|
|
s = '%s|%s|%s|%s|%s|%s|%s' % (rev, user, t, srcmd5, version,
|
|
comment.replace('\\', '\\\\').replace('\n', '\\n').replace('|', '\\|'), requestid)
|
|
r.append(s)
|
|
elif format == 'xml':
|
|
r.append('<logentry')
|
|
r.append(' revision="%s" srcmd5="%s">' % (rev, srcmd5))
|
|
r.append('<author>%s</author>' % user)
|
|
r.append('<date>%s</date>' % t)
|
|
r.append('<requestid>%s</requestid>' % requestid)
|
|
r.append('<msg>%s</msg>' %
|
|
comment.replace('&', '&').replace('<', '>').replace('>', '<'))
|
|
r.append('</logentry>')
|
|
else:
|
|
if requestid:
|
|
requestid="rq" + requestid
|
|
s = '-' * 76 + \
|
|
'\nr%s | %s | %s | %s | %s | %s\n' % (rev, user, t, srcmd5, version, requestid) + \
|
|
'\n' + comment
|
|
r.append(s)
|
|
|
|
if format not in ['csv', 'xml']:
|
|
r.append('-' * 76)
|
|
if format == 'xml':
|
|
r.append('</log>')
|
|
return r
|
|
|
|
|
|
def runservice(apiurl, prj, package):
|
|
u = makeurl(apiurl, ['source', prj, package], query={'cmd': 'runservice'})
|
|
|
|
try:
|
|
f = http_POST(u)
|
|
except urllib2.HTTPError, e:
|
|
e.osc_msg = 'could not trigger service run for project \'%s\' package \'%s\'' % (prj, package)
|
|
raise
|
|
|
|
root = ET.parse(f).getroot()
|
|
return root.get('code')
|
|
|
|
|
|
def rebuild(apiurl, prj, package, repo, arch, code=None):
|
|
query = { 'cmd': 'rebuild' }
|
|
if package:
|
|
query['package'] = package
|
|
if repo:
|
|
query['repository'] = repo
|
|
if arch:
|
|
query['arch'] = arch
|
|
if code:
|
|
query['code'] = code
|
|
|
|
u = makeurl(apiurl, ['build', prj], query=query)
|
|
try:
|
|
f = http_POST(u)
|
|
except urllib2.HTTPError, e:
|
|
e.osc_msg = 'could not trigger rebuild for project \'%s\' package \'%s\'' % (prj, package)
|
|
raise
|
|
|
|
root = ET.parse(f).getroot()
|
|
return root.get('code')
|
|
|
|
|
|
def store_read_project(dir):
|
|
global store
|
|
|
|
try:
|
|
p = open(os.path.join(dir, store, '_project')).readlines()[0].strip()
|
|
except IOError:
|
|
msg = 'Error: \'%s\' is not an osc project dir or working copy' % os.path.abspath(dir)
|
|
if os.path.exists(os.path.join(dir, '.svn')):
|
|
msg += '\nTry svn instead of osc.'
|
|
raise oscerr.NoWorkingCopy(msg)
|
|
return p
|
|
|
|
|
|
def store_read_package(dir):
|
|
global store
|
|
|
|
try:
|
|
p = open(os.path.join(dir, store, '_package')).readlines()[0].strip()
|
|
except IOError:
|
|
msg = 'Error: \'%s\' is not an osc package working copy' % os.path.abspath(dir)
|
|
if os.path.exists(os.path.join(dir, '.svn')):
|
|
msg += '\nTry svn instead of osc.'
|
|
raise oscerr.NoWorkingCopy(msg)
|
|
return p
|
|
|
|
def store_read_apiurl(dir, defaulturl=True):
|
|
global store
|
|
|
|
fname = os.path.join(dir, store, '_apiurl')
|
|
try:
|
|
url = open(fname).readlines()[0].strip()
|
|
# this is needed to get a proper apiurl
|
|
# (former osc versions may stored an apiurl with a trailing slash etc.)
|
|
apiurl = conf.urljoin(*conf.parse_apisrv_url(None, url))
|
|
except:
|
|
if not defaulturl:
|
|
if is_project_dir(dir):
|
|
project = store_read_project(dir)
|
|
package = None
|
|
elif is_package_dir(dir):
|
|
project = store_read_project(dir)
|
|
package = None
|
|
else:
|
|
msg = 'Error: \'%s\' is not an osc package working copy' % os.path.abspath(dir)
|
|
raise oscerr.NoWorkingCopy(msg)
|
|
msg = 'Your working copy \'%s\' is in an inconsistent state.\n' \
|
|
'Please run \'osc repairwc %s\' (Note this might _remove_\n' \
|
|
'files from the .osc/ dir). Please check the state\n' \
|
|
'of the working copy afterwards (via \'osc status %s\')' % (dir, dir, dir)
|
|
raise oscerr.WorkingCopyInconsistent(project, package, ['_apiurl'], msg)
|
|
apiurl = conf.config['apiurl']
|
|
return apiurl
|
|
|
|
def store_write_string(dir, file, string, subdir=''):
|
|
global store
|
|
|
|
if subdir and not os.path.isdir(os.path.join(dir, store, subdir)):
|
|
os.mkdir(os.path.join(dir, store, subdir))
|
|
fname = os.path.join(dir, store, subdir, file)
|
|
try:
|
|
f = open(fname + '.new', 'w')
|
|
f.write(string)
|
|
f.close()
|
|
os.rename(fname + '.new', fname)
|
|
except:
|
|
if os.path.exists(fname + '.new'):
|
|
os.unlink(fname + '.new')
|
|
raise
|
|
|
|
def store_write_project(dir, project):
|
|
store_write_string(dir, '_project', project + '\n')
|
|
|
|
def store_write_apiurl(dir, apiurl):
|
|
store_write_string(dir, '_apiurl', apiurl + '\n')
|
|
|
|
def store_unlink_file(dir, file):
|
|
global store
|
|
|
|
try: os.unlink(os.path.join(dir, store, file))
|
|
except: pass
|
|
|
|
def store_read_file(dir, file):
|
|
global store
|
|
|
|
try:
|
|
content = open(os.path.join(dir, store, file)).read()
|
|
return content
|
|
except:
|
|
return None
|
|
|
|
def store_write_initial_packages(dir, project, subelements):
|
|
global store
|
|
|
|
fname = os.path.join(dir, store, '_packages')
|
|
root = ET.Element('project', name=project)
|
|
for elem in subelements:
|
|
root.append(elem)
|
|
ET.ElementTree(root).write(fname)
|
|
|
|
def get_osc_version():
|
|
return __version__
|
|
|
|
|
|
def abortbuild(apiurl, project, package=None, arch=None, repo=None):
|
|
query = { 'cmd': 'abortbuild' }
|
|
if package:
|
|
query['package'] = package
|
|
if arch:
|
|
query['arch'] = arch
|
|
if repo:
|
|
query['repository'] = repo
|
|
u = makeurl(apiurl, ['build', project], query)
|
|
try:
|
|
f = http_POST(u)
|
|
except urllib2.HTTPError, e:
|
|
e.osc_msg = 'abortion failed for project %s' % project
|
|
if package:
|
|
e.osc_msg += ' package %s' % package
|
|
if arch:
|
|
e.osc_msg += ' arch %s' % arch
|
|
if repo:
|
|
e.osc_msg += ' repo %s' % repo
|
|
raise
|
|
|
|
root = ET.parse(f).getroot()
|
|
return root.get('code')
|
|
|
|
|
|
def wipebinaries(apiurl, project, package=None, arch=None, repo=None, code=None):
|
|
query = { 'cmd': 'wipe' }
|
|
if package:
|
|
query['package'] = package
|
|
if arch:
|
|
query['arch'] = arch
|
|
if repo:
|
|
query['repository'] = repo
|
|
if code:
|
|
query['code'] = code
|
|
|
|
u = makeurl(apiurl, ['build', project], query)
|
|
try:
|
|
f = http_POST(u)
|
|
except urllib2.HTTPError, e:
|
|
e.osc_msg = 'wipe binary rpms failed for project %s' % project
|
|
if package:
|
|
e.osc_msg += ' package %s' % package
|
|
if arch:
|
|
e.osc_msg += ' arch %s' % arch
|
|
if repo:
|
|
e.osc_msg += ' repository %s' % repo
|
|
if code:
|
|
e.osc_msg += ' code=%s' % code
|
|
raise
|
|
|
|
root = ET.parse(f).getroot()
|
|
return root.get('code')
|
|
|
|
|
|
def parseRevisionOption(string):
|
|
"""
|
|
returns a tuple which contains the revisions
|
|
"""
|
|
|
|
if string:
|
|
if ':' in string:
|
|
splitted_rev = string.split(':')
|
|
try:
|
|
for i in splitted_rev:
|
|
int(i)
|
|
return splitted_rev
|
|
except ValueError:
|
|
print >>sys.stderr, 'your revision \'%s\' will be ignored' % string
|
|
return None, None
|
|
else:
|
|
if string.isdigit():
|
|
return string, None
|
|
elif string.isalnum() and len(string) == 32:
|
|
# could be an md5sum
|
|
return string, None
|
|
else:
|
|
print >>sys.stderr, 'your revision \'%s\' will be ignored' % string
|
|
return None, None
|
|
else:
|
|
return None, None
|
|
|
|
def checkRevision(prj, pac, revision, apiurl=None, meta=False):
|
|
"""
|
|
check if revision is valid revision, i.e. it is not
|
|
larger than the upstream revision id
|
|
"""
|
|
if len(revision) == 32:
|
|
# there isn't a way to check this kind of revision for validity
|
|
return True
|
|
if not apiurl:
|
|
apiurl = conf.config['apiurl']
|
|
try:
|
|
if int(revision) > int(show_upstream_rev(apiurl, prj, pac, meta)) \
|
|
or int(revision) <= 0:
|
|
return False
|
|
else:
|
|
return True
|
|
except (ValueError, TypeError):
|
|
return False
|
|
|
|
def build_table(col_num, data = [], headline = [], width=1, csv = False):
|
|
"""
|
|
This method builds a simple table.
|
|
Example1: build_table(2, ['foo', 'bar', 'suse', 'osc'], ['col1', 'col2'], 2)
|
|
col1 col2
|
|
foo bar
|
|
suse osc
|
|
"""
|
|
|
|
longest_col = []
|
|
for i in range(col_num):
|
|
longest_col.append(0)
|
|
if headline and not csv:
|
|
data[0:0] = headline
|
|
# find longest entry in each column
|
|
i = 0
|
|
for itm in data:
|
|
if longest_col[i] < len(itm):
|
|
longest_col[i] = len(itm)
|
|
if i == col_num - 1:
|
|
i = 0
|
|
else:
|
|
i += 1
|
|
# calculate length for each column
|
|
for i, row in enumerate(longest_col):
|
|
longest_col[i] = row + width
|
|
# build rows
|
|
row = []
|
|
table = []
|
|
i = 0
|
|
for itm in data:
|
|
if i % col_num == 0:
|
|
i = 0
|
|
row = []
|
|
table.append(row)
|
|
# there is no need to justify the entries of the last column
|
|
# or when generating csv
|
|
if i == col_num -1 or csv:
|
|
row.append(itm)
|
|
else:
|
|
row.append(itm.ljust(longest_col[i]))
|
|
i += 1
|
|
if csv:
|
|
separator = '|'
|
|
else:
|
|
separator = ''
|
|
return [separator.join(row) for row in table]
|
|
|
|
def xpath_join(expr, new_expr, op='or', inner=False, nexpr_parentheses=False):
|
|
"""
|
|
Join two xpath expressions. If inner is False expr will
|
|
be surrounded with parentheses (unless it's not already
|
|
surrounded). If nexpr_parentheses is True new_expr will be
|
|
surrounded with parentheses.
|
|
"""
|
|
if not expr:
|
|
return new_expr
|
|
elif not new_expr:
|
|
return expr
|
|
# NOTE: this is NO syntax check etc. (e.g. if a literal contains a '(' or ')'
|
|
# the check might fail and expr will be surrounded with parentheses or NOT)
|
|
parentheses = not inner
|
|
if not inner and expr.startswith('(') and expr.endswith(')'):
|
|
parentheses = False
|
|
braces = [i for i in expr if i == '(' or i == ')']
|
|
closed = 0
|
|
while len(braces):
|
|
if braces.pop() == ')':
|
|
closed += 1
|
|
continue
|
|
else:
|
|
closed += -1
|
|
while len(braces):
|
|
if braces.pop() == '(':
|
|
closed += -1
|
|
else:
|
|
closed += 1
|
|
if closed != 0:
|
|
parentheses = True
|
|
break
|
|
if parentheses:
|
|
expr = '(%s)' % expr
|
|
if nexpr_parentheses:
|
|
new_expr = '(%s)' % new_expr
|
|
return '%s %s %s' % (expr, op, new_expr)
|
|
|
|
def search(apiurl, **kwargs):
|
|
"""
|
|
Perform a search request. The requests are constructed as follows:
|
|
kwargs = {'kind1' => xpath1, 'kind2' => xpath2, ..., 'kindN' => xpathN}
|
|
GET /search/kind1?match=xpath1
|
|
...
|
|
GET /search/kindN?match=xpathN
|
|
"""
|
|
res = {}
|
|
for urlpath, xpath in kwargs.iteritems():
|
|
u = makeurl(apiurl, ['search', urlpath], ['match=%s' % quote_plus(xpath)])
|
|
f = http_GET(u)
|
|
res[urlpath] = ET.parse(f).getroot()
|
|
return res
|
|
|
|
def set_link_rev(apiurl, project, package, revision='', expand=False, baserev=False):
|
|
"""
|
|
updates the rev attribute of the _link xml. If revision is set to None
|
|
the rev attribute is removed from the _link xml. If revision is set to ''
|
|
the "plain" upstream revision is used (if xsrcmd5 and baserev aren't specified).
|
|
"""
|
|
url = makeurl(apiurl, ['source', project, package, '_link'])
|
|
try:
|
|
f = http_GET(url)
|
|
root = ET.parse(f).getroot()
|
|
except urllib2.HTTPError, e:
|
|
e.osc_msg = 'Unable to get _link file in package \'%s\' for project \'%s\'' % (package, project)
|
|
raise
|
|
|
|
# set revision element
|
|
src_project = root.get('project', project)
|
|
src_package = root.get('package', package)
|
|
linkrev=None
|
|
if baserev:
|
|
linkrev = 'base'
|
|
expand = True
|
|
if revision is None:
|
|
if 'rev' in root.keys():
|
|
del root.attrib['rev']
|
|
elif revision == '' or expand:
|
|
revision = show_upstream_rev(apiurl, src_project, src_package, revision=revision, linkrev=linkrev, expand=expand)
|
|
|
|
if revision:
|
|
root.set('rev', revision)
|
|
|
|
l = ET.tostring(root)
|
|
http_PUT(url, data=l)
|
|
|
|
|
|
def delete_dir(dir):
|
|
# small security checks
|
|
if os.path.islink(dir):
|
|
raise oscerr.OscIOError(None, 'cannot remove linked dir')
|
|
elif os.path.abspath(dir) == '/':
|
|
raise oscerr.OscIOError(None, 'cannot remove \'/\'')
|
|
|
|
for dirpath, dirnames, filenames in os.walk(dir, topdown=False):
|
|
for filename in filenames:
|
|
os.unlink(os.path.join(dirpath, filename))
|
|
for dirname in dirnames:
|
|
os.rmdir(os.path.join(dirpath, dirname))
|
|
os.rmdir(dir)
|
|
|
|
|
|
def delete_storedir(store_dir):
|
|
"""
|
|
This method deletes a store dir.
|
|
"""
|
|
head, tail = os.path.split(store_dir)
|
|
if tail == '.osc':
|
|
delete_dir(store_dir)
|
|
|
|
def unpack_srcrpm(srpm, dir, *files):
|
|
"""
|
|
This method unpacks the passed srpm into the
|
|
passed dir. If arguments are passed to the \'files\' tuple
|
|
only this files will be unpacked.
|
|
"""
|
|
if not is_srcrpm(srpm):
|
|
print >>sys.stderr, 'error - \'%s\' is not a source rpm.' % srpm
|
|
sys.exit(1)
|
|
curdir = os.getcwd()
|
|
if os.path.isdir(dir):
|
|
os.chdir(dir)
|
|
cmd = 'rpm2cpio %s | cpio -i %s &> /dev/null' % (srpm, ' '.join(files))
|
|
ret = subprocess.call(cmd, shell=True)
|
|
if ret != 0:
|
|
print >>sys.stderr, 'error \'%s\' - cannot extract \'%s\'' % (ret, srpm)
|
|
sys.exit(1)
|
|
os.chdir(curdir)
|
|
|
|
def is_rpm(f):
|
|
"""check if the named file is an RPM package"""
|
|
try:
|
|
h = open(f, 'rb').read(4)
|
|
except:
|
|
return False
|
|
|
|
if h == '\xed\xab\xee\xdb':
|
|
return True
|
|
else:
|
|
return False
|
|
|
|
def is_srcrpm(f):
|
|
"""check if the named file is a source RPM"""
|
|
|
|
if not is_rpm(f):
|
|
return False
|
|
|
|
try:
|
|
h = open(f, 'rb').read(8)
|
|
except:
|
|
return False
|
|
|
|
if h[7] == '\x01':
|
|
return True
|
|
else:
|
|
return False
|
|
|
|
def addMaintainer(apiurl, prj, pac, user):
|
|
# for backward compatibility only
|
|
addPerson(apiurl, prj, pac, user)
|
|
|
|
def addPerson(apiurl, prj, pac, user, role="maintainer"):
|
|
""" add a new person to a package or project """
|
|
path = quote_plus(prj),
|
|
kind = 'prj'
|
|
if pac:
|
|
path = path + (quote_plus(pac),)
|
|
kind = 'pkg'
|
|
data = meta_exists(metatype=kind,
|
|
path_args=path,
|
|
template_args=None,
|
|
create_new=False)
|
|
|
|
if data and get_user_meta(apiurl, user) != None:
|
|
root = ET.fromstring(''.join(data))
|
|
found = False
|
|
for person in root.getiterator('person'):
|
|
if person.get('userid') == user and person.get('role') == role:
|
|
found = True
|
|
print "user already exists"
|
|
break
|
|
if not found:
|
|
# the xml has a fixed structure
|
|
root.insert(2, ET.Element('person', role=role, userid=user))
|
|
print 'user \'%s\' added to \'%s\'' % (user, pac or prj)
|
|
edit_meta(metatype=kind,
|
|
path_args=path,
|
|
data=ET.tostring(root))
|
|
else:
|
|
print "osc: an error occured"
|
|
|
|
def delMaintainer(apiurl, prj, pac, user):
|
|
# for backward compatibility only
|
|
delPerson(apiurl, prj, pac, user)
|
|
|
|
def delPerson(apiurl, prj, pac, user, role="maintainer"):
|
|
""" delete a person from a package or project """
|
|
path = quote_plus(prj),
|
|
kind = 'prj'
|
|
if pac:
|
|
path = path + (quote_plus(pac), )
|
|
kind = 'pkg'
|
|
data = meta_exists(metatype=kind,
|
|
path_args=path,
|
|
template_args=None,
|
|
create_new=False)
|
|
if data and get_user_meta(apiurl, user) != None:
|
|
root = ET.fromstring(''.join(data))
|
|
found = False
|
|
for person in root.getiterator('person'):
|
|
if person.get('userid') == user and person.get('role') == role:
|
|
root.remove(person)
|
|
found = True
|
|
print "user \'%s\' removed" % user
|
|
if found:
|
|
edit_meta(metatype=kind,
|
|
path_args=path,
|
|
data=ET.tostring(root))
|
|
else:
|
|
print "user \'%s\' not found in \'%s\'" % (user, pac or prj)
|
|
else:
|
|
print "an error occured"
|
|
|
|
def setDevelProject(apiurl, prj, pac, dprj, dpkg=None):
|
|
""" set the <devel project="..."> element to package metadata"""
|
|
path = (quote_plus(prj),) + (quote_plus(pac),)
|
|
data = meta_exists(metatype='pkg',
|
|
path_args=path,
|
|
template_args=None,
|
|
create_new=False)
|
|
|
|
if data and show_project_meta(apiurl, dprj) != None:
|
|
root = ET.fromstring(''.join(data))
|
|
if not root.find('devel') != None:
|
|
ET.SubElement(root, 'devel')
|
|
elem = root.find('devel')
|
|
if dprj:
|
|
elem.set('project', dprj)
|
|
else:
|
|
if 'project' in elem.keys():
|
|
del elem.attrib['project']
|
|
if dpkg:
|
|
elem.set('package', dpkg)
|
|
else:
|
|
if 'package' in elem.keys():
|
|
del elem.attrib['package']
|
|
edit_meta(metatype='pkg',
|
|
path_args=path,
|
|
data=ET.tostring(root))
|
|
else:
|
|
print "osc: an error occured"
|
|
|
|
def createPackageDir(pathname, prj_obj=None):
|
|
"""
|
|
create and initialize a new package dir in the given project.
|
|
prj_obj can be a Project() instance.
|
|
"""
|
|
prj_dir, pac_dir = getPrjPacPaths(pathname)
|
|
if is_project_dir(prj_dir):
|
|
global store
|
|
if not os.path.exists(pac_dir+store):
|
|
prj = prj_obj or Project(prj_dir, False)
|
|
Package.init_package(prj.apiurl, prj.name, pac_dir, pac_dir)
|
|
prj.addPackage(pac_dir)
|
|
print statfrmt('A', os.path.normpath(pathname))
|
|
else:
|
|
raise oscerr.OscIOError(None, 'file or directory \'%s\' already exists' % pathname)
|
|
else:
|
|
msg = '\'%s\' is not a working copy' % prj_dir
|
|
if os.path.exists(os.path.join(prj_dir, '.svn')):
|
|
msg += '\ntry svn instead of osc.'
|
|
raise oscerr.NoWorkingCopy(msg)
|
|
|
|
|
|
def stripETxml(node):
|
|
node.tail = None
|
|
if node.text != None:
|
|
node.text = node.text.replace(" ", "").replace("\n", "")
|
|
for child in node.getchildren():
|
|
stripETxml(child)
|
|
|
|
def addGitSource(url):
|
|
service_file = os.path.join(os.getcwd(), '_service')
|
|
addfile = False
|
|
if os.path.exists( service_file ):
|
|
services = ET.parse(os.path.join(os.getcwd(), '_service')).getroot()
|
|
else:
|
|
services = ET.fromstring("<services />")
|
|
addfile = True
|
|
stripETxml( services )
|
|
si = Serviceinfo()
|
|
s = si.addGitUrl(services, url)
|
|
s = si.addRecompressTar(services)
|
|
si.read(s)
|
|
|
|
# for pretty output
|
|
xmlindent(s)
|
|
f = open(service_file, 'wb')
|
|
f.write(ET.tostring(s))
|
|
f.close()
|
|
if addfile:
|
|
addFiles( ['_service'] )
|
|
|
|
def addDownloadUrlService(url):
|
|
service_file = os.path.join(os.getcwd(), '_service')
|
|
addfile = False
|
|
if os.path.exists( service_file ):
|
|
services = ET.parse(os.path.join(os.getcwd(), '_service')).getroot()
|
|
else:
|
|
services = ET.fromstring("<services />")
|
|
addfile = True
|
|
stripETxml( services )
|
|
si = Serviceinfo()
|
|
s = si.addDownloadUrl(services, url)
|
|
si.read(s)
|
|
|
|
# for pretty output
|
|
xmlindent(s)
|
|
f = open(service_file, 'wb')
|
|
f.write(ET.tostring(s))
|
|
f.close()
|
|
if addfile:
|
|
addFiles( ['_service'] )
|
|
|
|
# download file
|
|
path = os.getcwd()
|
|
files = os.listdir(path)
|
|
si.execute(path)
|
|
newfiles = os.listdir(path)
|
|
|
|
# add verify service for new files
|
|
for filename in files:
|
|
newfiles.remove(filename)
|
|
|
|
for filename in newfiles:
|
|
if filename.startswith('_service:download_url:'):
|
|
s = si.addVerifyFile(services, filename)
|
|
|
|
# for pretty output
|
|
xmlindent(s)
|
|
f = open(service_file, 'wb')
|
|
f.write(ET.tostring(s))
|
|
f.close()
|
|
|
|
|
|
def addFiles(filenames, prj_obj = None):
|
|
for filename in filenames:
|
|
if not os.path.exists(filename):
|
|
raise oscerr.OscIOError(None, 'file \'%s\' does not exist' % filename)
|
|
|
|
# init a package dir if we have a normal dir in the "filenames"-list
|
|
# so that it will be find by findpacs() later
|
|
pacs = list(filenames)
|
|
for filename in filenames:
|
|
prj_dir, pac_dir = getPrjPacPaths(filename)
|
|
if not is_package_dir(filename) and os.path.isdir(filename) and is_project_dir(prj_dir) \
|
|
and conf.config['do_package_tracking']:
|
|
prj_name = store_read_project(prj_dir)
|
|
prj_apiurl = store_read_apiurl(prj_dir, defaulturl=False)
|
|
Package.init_package(prj_apiurl, prj_name, pac_dir, filename)
|
|
elif is_package_dir(filename) and conf.config['do_package_tracking']:
|
|
raise oscerr.PackageExists(store_read_project(filename), store_read_package(filename),
|
|
'osc: warning: \'%s\' is already under version control' % filename)
|
|
elif os.path.isdir(filename) and is_project_dir(prj_dir):
|
|
raise oscerr.WrongArgs('osc: cannot add a directory to a project unless ' \
|
|
'\'do_package_tracking\' is enabled in the configuration file')
|
|
elif os.path.isdir(filename):
|
|
print 'skipping directory \'%s\'' % filename
|
|
pacs.remove(filename)
|
|
pacs = findpacs(pacs)
|
|
for pac in pacs:
|
|
if conf.config['do_package_tracking'] and not pac.todo:
|
|
prj = prj_obj or Project(os.path.dirname(pac.absdir), False)
|
|
if pac.name in prj.pacs_unvers:
|
|
prj.addPackage(pac.name)
|
|
print statfrmt('A', getTransActPath(os.path.join(pac.dir, os.pardir, pac.name)))
|
|
for filename in pac.filenamelist_unvers:
|
|
if os.path.isdir(os.path.join(pac.dir, filename)):
|
|
print 'skipping directory \'%s\'' % os.path.join(pac.dir, filename)
|
|
else:
|
|
pac.todo.append(filename)
|
|
elif pac.name in prj.pacs_have:
|
|
print 'osc: warning: \'%s\' is already under version control' % pac.name
|
|
for filename in pac.todo:
|
|
if filename in pac.skipped:
|
|
continue
|
|
if filename in pac.excluded:
|
|
print >>sys.stderr, 'osc: warning: \'%s\' is excluded from a working copy' % filename
|
|
continue
|
|
pac.addfile(filename)
|
|
|
|
def getPrjPacPaths(path):
|
|
"""
|
|
returns the path for a project and a package
|
|
from path. This is needed if you try to add
|
|
or delete packages:
|
|
Examples:
|
|
osc add pac1/: prj_dir = CWD;
|
|
pac_dir = pac1
|
|
osc add /path/to/pac1:
|
|
prj_dir = path/to;
|
|
pac_dir = pac1
|
|
osc add /path/to/pac1/file
|
|
=> this would be an invalid path
|
|
the caller has to validate the returned
|
|
path!
|
|
"""
|
|
# make sure we hddave a dir: osc add bar vs. osc add bar/; osc add /path/to/prj_dir/new_pack
|
|
# filename = os.path.join(tail, '')
|
|
prj_dir, pac_dir = os.path.split(os.path.normpath(path))
|
|
if prj_dir == '':
|
|
prj_dir = os.getcwd()
|
|
return (prj_dir, pac_dir)
|
|
|
|
def getTransActPath(pac_dir):
|
|
"""
|
|
returns the path for the commit and update operations/transactions.
|
|
Normally the "dir" attribute of a Package() object will be passed to
|
|
this method.
|
|
"""
|
|
if pac_dir != '.':
|
|
pathn = os.path.normpath(pac_dir)
|
|
else:
|
|
pathn = ''
|
|
return pathn
|
|
|
|
def get_commit_message_template(pac):
|
|
"""
|
|
Read the difference in .changes file(s) and put them as a template to commit message.
|
|
"""
|
|
diff = []
|
|
template = []
|
|
|
|
if pac.todo:
|
|
todo = pac.todo
|
|
else:
|
|
todo = pac.filenamelist + pac.filenamelist_unvers
|
|
|
|
files = [i for i in todo if i.endswith('.changes') and pac.status(i) in ('A', 'M')]
|
|
|
|
for filename in files:
|
|
if pac.status(filename) == 'M':
|
|
diff += get_source_file_diff(pac.absdir, filename, pac.rev)
|
|
elif pac.status(filename) == 'A':
|
|
f = open(filename, 'r')
|
|
for line in f:
|
|
diff += '+' + line
|
|
f.close()
|
|
|
|
if diff:
|
|
template = parse_diff_for_commit_message(''.join(diff))
|
|
|
|
return template
|
|
|
|
def parse_diff_for_commit_message(diff, template = []):
|
|
date_re = re.compile(r'\+(Mon|Tue|Wed|Thu|Fri|Sat|Sun) ([A-Z][a-z]{2}) ( ?[0-9]|[0-3][0-9]) .*')
|
|
diff = diff.split('\n')
|
|
|
|
# The first four lines contains a header of diff
|
|
for line in diff[3:]:
|
|
# this condition is magical, but it removes all unwanted lines from commit message
|
|
if not(line) or (line and line[0] != '+') or \
|
|
date_re.match(line) or \
|
|
line == '+' or line[0:3] == '+++':
|
|
continue
|
|
|
|
if line == '+-------------------------------------------------------------------':
|
|
template.append('')
|
|
else:
|
|
template.append(line[1:])
|
|
|
|
return template
|
|
|
|
def get_commit_msg(wc_dir, pacs):
|
|
template = store_read_file(wc_dir, '_commit_msg')
|
|
# open editor for commit message
|
|
# but first, produce status and diff to append to the template
|
|
footer = []
|
|
lines = []
|
|
for p in pacs:
|
|
states = sorted(p.get_status(False, ' ', '?'), lambda x, y: cmp(x[1], y[1]))
|
|
changed = [statfrmt(st, os.path.normpath(os.path.join(p.dir, filename))) for st, filename in states]
|
|
if changed:
|
|
footer += changed
|
|
footer.append('\nDiff for working copy: %s' % p.dir)
|
|
footer.extend([''.join(i) for i in p.get_diff(ignoreUnversioned=True)])
|
|
lines.extend(get_commit_message_template(p))
|
|
if template is None:
|
|
if lines and lines[0] == '':
|
|
del lines[0]
|
|
template = '\n'.join(lines)
|
|
msg = ''
|
|
# if footer is empty, there is nothing to commit, and no edit needed.
|
|
if footer:
|
|
msg = edit_message(footer='\n'.join(footer), template=template)
|
|
if msg:
|
|
store_write_string(wc_dir, '_commit_msg', msg + '\n')
|
|
else:
|
|
store_unlink_file(wc_dir, '_commit_msg')
|
|
return msg
|
|
|
|
def print_request_list(apiurl, project, package = None, states = ('new','review',), force = False):
|
|
"""
|
|
prints list of pending requests for the specified project/package if "check_for_request_on_action"
|
|
is enabled in the config or if "force" is set to True
|
|
"""
|
|
if not conf.config['check_for_request_on_action'] and not force:
|
|
return
|
|
requests = get_request_list(apiurl, project, package, req_state=states)
|
|
msg = 'Pending requests for %s: %s (%s)'
|
|
if package is None and len(requests):
|
|
print msg % ('project', project, len(requests))
|
|
elif len(requests):
|
|
print msg % ('package', '/'.join([project, package]), len(requests))
|
|
for r in requests:
|
|
print r.list_view(), '\n'
|
|
|
|
def request_interactive_review(apiurl, request, initial_cmd=''):
|
|
"""review the request interactively"""
|
|
import tempfile, re
|
|
|
|
tmpfile = None
|
|
|
|
def print_request(request):
|
|
print request
|
|
|
|
print_request(request)
|
|
try:
|
|
prompt = '(a)ccept/(d)ecline/(r)evoke/c(l)one/(s)kip/(c)ancel > '
|
|
sr_actions = request.get_actions('submit')
|
|
if sr_actions:
|
|
prompt = 'd(i)ff/(a)ccept/(d)ecline/(r)evoke/(b)uildstatus/c(l)one/(e)dit/(s)kip/(c)ancel > '
|
|
editprj = ''
|
|
orequest = None
|
|
while True:
|
|
if initial_cmd:
|
|
repl = initial_cmd
|
|
initial_cmd = ''
|
|
else:
|
|
repl = raw_input(prompt).strip()
|
|
if repl == 'i' and sr_actions:
|
|
if not orequest is None and tmpfile:
|
|
tmpfile.close()
|
|
tmpfile = None
|
|
if tmpfile is None:
|
|
tmpfile = tempfile.NamedTemporaryFile(suffix='.diff')
|
|
try:
|
|
diff = request_diff(apiurl, request.reqid)
|
|
tmpfile.write(diff)
|
|
except urllib2.HTTPError, e:
|
|
if e.code != 400:
|
|
raise
|
|
# backward compatible diff for old apis
|
|
for action in sr_actions:
|
|
diff = 'old: %s/%s\nnew: %s/%s\n' % (action.src_project, action.src_package,
|
|
action.tgt_project, action.tgt_package)
|
|
diff += submit_action_diff(apiurl, action)
|
|
diff += '\n\n'
|
|
tmpfile.write(diff)
|
|
tmpfile.flush()
|
|
run_editor(tmpfile.name)
|
|
print_request(request)
|
|
elif repl == 's':
|
|
print >>sys.stderr, 'skipping: #%s' % request.reqid
|
|
break
|
|
elif repl == 'c':
|
|
print >>sys.stderr, 'Aborting'
|
|
raise oscerr.UserAbort()
|
|
elif repl == 'b' and sr_actions:
|
|
for action in sr_actions:
|
|
print '%s/%s:' % (action.src_project, action.src_package)
|
|
print '\n'.join(get_results(apiurl, action.src_project, action.src_package))
|
|
elif repl == 'e' and sr_actions:
|
|
if not editprj:
|
|
editprj = clone_request(apiurl, request.reqid, 'osc editrequest')
|
|
orequest = request
|
|
request = edit_submitrequest(apiurl, editprj, orequest, request)
|
|
sr_actions = request.get_actions('submit')
|
|
print_request(request)
|
|
prompt = 'd(i)ff/(a)ccept/(b)uildstatus/(e)dit/(s)kip/(c)ancel > '
|
|
else:
|
|
state_map = {'a': 'accepted', 'd': 'declined', 'r': 'revoked'}
|
|
mo = re.search('^([adrl])(?:\s+-m\s+(.*))?$', repl)
|
|
if mo is None or orequest and mo.group(1) != 'a':
|
|
print >>sys.stderr, 'invalid choice: \'%s\'' % repl
|
|
continue
|
|
state = state_map.get(mo.group(1))
|
|
msg = mo.group(2)
|
|
footer = ''
|
|
msg_template = ''
|
|
if not (state is None or request.state is None):
|
|
footer = 'changing request from state \'%s\' to \'%s\'\n\n' \
|
|
% (request.state.name, state)
|
|
msg_template = change_request_state_template(request, state)
|
|
footer += str(request)
|
|
if tmpfile is not None:
|
|
tmpfile.seek(0)
|
|
# the read bytes probably have a moderate size so the str won't be too large
|
|
footer += '\n\n' + tmpfile.read()
|
|
if msg is None:
|
|
msg = edit_message(footer = footer, template=msg_template)
|
|
else:
|
|
msg = msg.strip('\'').strip('"')
|
|
if not orequest is None:
|
|
request.create(apiurl)
|
|
change_request_state(apiurl, request.reqid, 'accepted', msg)
|
|
repl = raw_input('Supersede original request? (y|N) ')
|
|
if repl in ('y', 'Y'):
|
|
change_request_state(apiurl, orequest.reqid, 'superseded',
|
|
'superseded by %s' % request.reqid, request.reqid)
|
|
elif state is None:
|
|
clone_request(apiurl, request.reqid, msg)
|
|
else:
|
|
reviews = [r for r in request.reviews if r.state == 'new']
|
|
if not reviews:
|
|
change_request_state(apiurl, request.reqid, state, msg)
|
|
break
|
|
print 'Please chose one of the following reviews:'
|
|
for i in range(len(reviews)):
|
|
fmt = Request.format_review(reviews[i])
|
|
print '(%i)' % i, 'by %(type)-10s %(by)s' % fmt
|
|
num = raw_input('> ')
|
|
try:
|
|
num = int(num)
|
|
except ValueError:
|
|
print '\'%s\' is not a number.' % num
|
|
continue
|
|
if num < 0 or num >= len(reviews):
|
|
print 'number \'%s\' out of range.' % num
|
|
continue
|
|
review = reviews[num]
|
|
change_review_state(apiurl, request.reqid, state, by_user=review.by_user,
|
|
by_group=review.by_group, by_project=review.by_project,
|
|
by_package=review.by_package, message=msg)
|
|
break
|
|
finally:
|
|
if tmpfile is not None:
|
|
tmpfile.close()
|
|
|
|
def edit_submitrequest(apiurl, project, orequest, new_request=None):
|
|
"""edit a submit action from orequest/new_request"""
|
|
import tempfile, shutil, subprocess
|
|
actions = orequest.get_actions('submit')
|
|
oactions = actions
|
|
if not orequest is None:
|
|
actions = new_request.get_actions('submit')
|
|
num = 0
|
|
if len(actions) > 1:
|
|
print 'Please chose one of the following submit actions:'
|
|
for i in range(len(actions)):
|
|
fmt = Request.format_action(actions[i])
|
|
print '(%i)' % i, '%(source)s %(target)s' % fmt
|
|
num = raw_input('> ')
|
|
try:
|
|
num = int(num)
|
|
except ValueError:
|
|
raise oscerr.WrongArgs('\'%s\' is not a number.' % num)
|
|
if num < 0 or num >= len(orequest.actions):
|
|
raise oscerr.WrongArgs('number \'%s\' out of range.' % num)
|
|
|
|
# the api replaced ':' with '_' in prj and pkg names (clone request)
|
|
package = '%s.%s' % (oactions[num].src_package.replace(':', '_'),
|
|
oactions[num].src_project.replace(':', '_'))
|
|
tmpdir = None
|
|
cleanup = True
|
|
try:
|
|
tmpdir = tempfile.mkdtemp(prefix='osc_editsr')
|
|
p = Package.init_package(apiurl, project, package, tmpdir)
|
|
p.update()
|
|
shell = os.getenv('SHELL', default='/bin/sh')
|
|
olddir = os.getcwd()
|
|
os.chdir(tmpdir)
|
|
print 'Checked out package \'%s\' to %s. Started a new shell (%s).\n' \
|
|
'Please fix the package and close the shell afterwards.' % (package, tmpdir, shell)
|
|
subprocess.call(shell)
|
|
# the pkg might have uncommitted changes...
|
|
cleanup = False
|
|
os.chdir(olddir)
|
|
# reread data
|
|
p = Package(tmpdir)
|
|
modified = p.get_status(False, ' ', '?', 'S')
|
|
if modified:
|
|
print 'Your working copy has the following modifications:'
|
|
print '\n'.join([statfrmt(st, filename) for st, filename in modified])
|
|
repl = raw_input('Do you want to commit the local changes first? (y|N) ')
|
|
if repl in ('y', 'Y'):
|
|
msg = get_commit_msg(p.absdir, [p])
|
|
p.commit(msg=msg)
|
|
cleanup = True
|
|
finally:
|
|
if cleanup:
|
|
shutil.rmtree(tmpdir)
|
|
else:
|
|
print 'Please remove the dir \'%s\' manually' % tmpdir
|
|
r = Request()
|
|
for action in orequest.get_actions():
|
|
new_action = Action.from_xml(action.to_xml())
|
|
r.actions.append(new_action)
|
|
if new_action.type == 'submit':
|
|
new_action.src_package = '%s.%s' % (action.src_package.replace(':', '_'),
|
|
action.src_project.replace(':', '_'))
|
|
new_action.src_project = project
|
|
# do an implicit cleanup
|
|
new_action.opt_sourceupdate = 'cleanup'
|
|
return r
|
|
|
|
def get_user_projpkgs(apiurl, user, role=None, exclude_projects=[], proj=True, pkg=True, maintained=False, metadata=False):
|
|
"""Return all project/packages where user is involved."""
|
|
xpath = 'person/@userid = \'%s\'' % user
|
|
excl_prj = ''
|
|
excl_pkg = ''
|
|
for i in exclude_projects:
|
|
excl_prj = xpath_join(excl_prj, 'not(@name = \'%s\')' % i, op='and')
|
|
excl_pkg = xpath_join(excl_pkg, 'not(@project = \'%s\')' % i, op='and')
|
|
role_filter_xpath = xpath
|
|
if role:
|
|
xpath = xpath_join(xpath, 'person/@role = \'%s\'' % role, inner=True, op='and')
|
|
xpath_pkg = xpath_join(xpath, excl_pkg, op='and')
|
|
xpath_prj = xpath_join(xpath, excl_prj, op='and')
|
|
|
|
if maintained:
|
|
xpath_pkg = xpath_join(xpath_pkg, '(project/attribute/@name=\'%(attr)s\' or attribute/@name=\'%(attr)s\')' % {'attr': conf.config['maintained_attribute']}, op='and')
|
|
|
|
what = {}
|
|
if pkg:
|
|
if metadata:
|
|
what['package'] = xpath_pkg
|
|
else:
|
|
what['package_id'] = xpath_pkg
|
|
if proj:
|
|
if metadata:
|
|
what['project'] = xpath_prj
|
|
else:
|
|
what['project_id'] = xpath_prj
|
|
try:
|
|
res = search(apiurl, **what)
|
|
except urllib2.HTTPError, e:
|
|
if e.code != 400 or not role_filter_xpath:
|
|
raise e
|
|
# backward compatibility: local role filtering
|
|
what = dict([[kind, role_filter_xpath] for kind in what.keys()])
|
|
if what.has_key('package'):
|
|
what['package'] = xpath_join(role_filter_xpath, excl_pkg, op='and')
|
|
if what.has_key('project'):
|
|
what['project'] = xpath_join(role_filter_xpath, excl_prj, op='and')
|
|
res = search(apiurl, **what)
|
|
filter_role(res, user, role)
|
|
return res
|
|
|
|
def raw_input(*args):
|
|
import __builtin__
|
|
try:
|
|
return __builtin__.raw_input(*args)
|
|
except EOFError:
|
|
# interpret ctrl-d as user abort
|
|
raise oscerr.UserAbort()
|
|
|
|
# backward compatibility: local role filtering
|
|
def filter_role(meta, user, role):
|
|
"""
|
|
remove all project/package nodes if no person node exists
|
|
where @userid=user and @role=role
|
|
"""
|
|
for kind, root in meta.iteritems():
|
|
delete = []
|
|
for node in root.findall(kind):
|
|
found = False
|
|
for p in node.findall('person'):
|
|
if p.get('userid') == user and p.get('role') == role:
|
|
found = True
|
|
break
|
|
if not found:
|
|
delete.append(node)
|
|
for node in delete:
|
|
root.remove(node)
|
|
|
|
def find_default_project(apiurl=None, package=None):
|
|
""""
|
|
look though the list of conf.config['getpac_default_project']
|
|
and find the first project where the given package exists in the build service.
|
|
"""
|
|
if not len(conf.config['getpac_default_project']):
|
|
return None
|
|
candidates = re.split('[, ]+', conf.config['getpac_default_project'])
|
|
if package is None or len(candidates) == 1:
|
|
return candidates[0]
|
|
|
|
# search through the list, where package exists ...
|
|
for prj in candidates:
|
|
try:
|
|
# any fast query will do here.
|
|
show_package_meta(apiurl, prj, package)
|
|
return prj
|
|
except urllib2.HTTPError:
|
|
pass
|
|
return None
|
|
|
|
|
|
|
|
# vim: sw=4 et
|