glib/glib/update-gtranslit.py
Philip Withnall d270b6c3db py: Various flake8 cleanups
None of these are particularly significant, but they do get the CI
output clean.

Signed-off-by: Philip Withnall <pwithnall@endlessos.org>
2020-11-17 15:50:07 +00:00

457 lines
14 KiB
Python
Executable File

#!/usr/bin/env python3
# Run this script like so:
#
# ./update-gtranslit.py /path/to/glibc/localedata/locales > gtranslit-data.h
import os
import sys
localedir = sys.argv[1]
# returns true if the name looks like a POSIX locale name
def looks_like_locale(name):
name, _, variant = name.partition("@")
if "_" not in name:
return False
lang, _, land = name.partition("_")
return len(lang) == 2 or len(lang) == 3 and len(land) == 2
# handles <U1234> style escapes
def unescape(string):
chunks = []
n = len(string)
i = 0
while i < n:
start_escape = string.find("<", i)
if start_escape == -1:
chunks.append(string[i:])
break
assert string[start_escape : (start_escape + 2)] == "<U"
start_escape += 2
end_escape = string.find(">", start_escape)
assert end_escape != -1
chunks.append(chr(int(string[start_escape:end_escape], 16)))
i = end_escape + 1
return "".join(chunks)
# Checks if a string is ascii
def is_ascii(string):
return all(ord(c) < 0x80 for c in string)
# A Mapping is a map from non-ascii strings to ascii strings.
#
# It corresponds to a sequence of one or more mapping lines:
#
# <U00C4> "<U0041><U0308>";"<U0041><U0045>"
#
# in a file.
class Mapping:
def __init__(self):
self.serialised = None
self.mapping = {}
# Scans a string like
#
# <U00C4> "<U0041><U0308>";"<U0041><U0045>" % \
# LATIN CAPITAL LETTER A WITH DIAERESIS.
#
# and adds the first all-ascii choice (or IGNORE) to the mapping
# dictionary, with the origin string as the key. In the case of
# IGNORE, stores the empty string.
def consider_mapping_line(self, line):
key, value, rest = (line + " % comment").split(maxsplit=2)
key = unescape(key)
for alternative in value.split(";"):
if alternative[0] == '"' and alternative[-1] == '"':
unescaped = unescape(alternative[1:-1])
if is_ascii(unescaped):
self.mapping[key] = unescaped
break
elif alternative[0] == "<" and alternative[-1] == ">":
unescaped = unescape(alternative)
if is_ascii(unescaped):
self.mapping[key] = unescaped
break
elif alternative == "IGNORE":
self.mapping[key] = ""
break
# Performs a normal dictionary merge, but ensures that there are no
# conflicting entries between the original dictionary and the requested
# changes
def merge_mapping(self, changes):
for key in changes.mapping:
if key in self.mapping:
assert self.mapping[key] == changes.mapping[key]
self.mapping.update(changes.mapping)
# Can't get much flatter...
def get_flattened(self):
return [self]
def serialise(self, serialiser):
if self.serialised is None:
self.serialised = serialiser.add_mapping(self.mapping)
return self.serialised
# A Chain is a sequence of mappings and chains.
#
# A chain contains another chain whenever "copy" or "include" is
# encountered in a source file.
#
# A chain contains a mapping whenever a sequence of mapping lines:
#
# <U00C4> "<U0041><U0308>";"<U0041><U0045>"
#
# is encountered in a file.
#
# The order of lookup is reverse: later entries override earlier ones.
class Chain:
def __init__(self, name):
self.serialised = None
self.name = name
self.chain = []
self.links = 0
self.read_from_file(os.path.join(localedir, name))
def read_from_file(self, filename):
current_mapping = None
in_lc_ctype = False
in_translit = False
fp = open(filename, encoding="ascii", errors="surrogateescape")
for line in fp:
line = line.strip()
if in_lc_ctype:
if line == "END LC_CTYPE":
break
if line.startswith("copy") or line.startswith("include"):
if current_mapping:
self.chain.append(current_mapping)
copyname = unescape(line.split('"', 3)[1])
copyfile = get_chain(copyname)
self.chain.append(copyfile)
copyfile.links += 1
current_mapping = None
elif line == "translit_start":
in_translit = True
elif line == "translit_end":
in_translit = False
elif in_translit and line.startswith("<U"):
if not current_mapping:
current_mapping = Mapping()
current_mapping.consider_mapping_line(line)
elif line == "" or line.startswith("%"):
pass
elif "default_missing <U003F>":
pass
elif in_translit:
print("unknown line:", line)
assert False
elif line == "LC_CTYPE":
in_lc_ctype = True
if current_mapping:
self.chain.append(current_mapping)
# If there is only one link to this chain, we may as well just
# return the contents of the chain so that they can be merged into
# our sole parent directly. Otherwise, return ourselves.
def get_flattened(self):
if self.links == 1:
return sum((item.get_flattened() for item in self.chain), [])
else:
return [self]
def serialise(self, serialiser):
if self.serialised is None:
# Before we serialise, see if we can optimise a bit
self.chain = sum((item.get_flattened() for item in self.chain), [])
i = 0
while i < len(self.chain) - 1:
if isinstance(self.chain[i], Mapping) and isinstance(
self.chain[i + 1], Mapping
):
# We have two mappings in a row. Try to merge them.
self.chain[i].merge_mapping(self.chain[i + 1])
del self.chain[i + 1]
else:
i += 1
# If all that is left is one item, just serialise that directly
if len(self.chain) == 1:
self.serialised = self.chain[0].serialise(serialiser)
else:
ids = [item.serialise(serialiser) for item in self.chain]
self.serialised = serialiser.add_chain(ids)
return self.serialised
# Chain cache -- allows sharing of common chains
chains = {}
def get_chain(name):
if name not in chains:
chains[name] = Chain(name)
return chains[name]
# Remove the country name from a locale, preserving variant
# eg: 'sr_RS@latin' -> 'sr@latin'
def remove_country(string):
base, at, variant = string.partition("@")
lang, _, land = base.partition("_")
return lang + at + variant
def encode_range(start, end):
assert start <= end
length = end - start
assert start < 0x1000
assert length < 0x8
result = 0x8000 + (length << 12) + start
assert result < 0x10000
return result
def c_pair_array(array):
return "{ " + ", ".join("{ %u, %u }" % pair for pair in array) + " };"
class Serialiser:
def __init__(self):
self.mappings = []
self.chains = []
self.locales = {}
def add_mapping(self, mapping):
if mapping in self.mappings:
mapping_id = self.mappings.index(mapping)
else:
mapping_id = len(self.mappings)
self.mappings.append(mapping)
assert mapping_id < 128
return mapping_id
def add_chain(self, chain):
if chain in self.chains:
chain_id = self.chains.index(chain)
else:
chain_id = len(self.chains)
self.chains.append(chain)
assert chain_id < 128
return 128 + chain_id
def add_locale(self, name, item_id):
self.locales[name] = item_id
def add_default(self, item_id):
self.default = item_id
def optimise_locales(self):
# Check if all regions of a language/variant agree
languages = list(set(remove_country(locale) for locale in self.locales))
for language in languages:
locales = [
locale for locale in self.locales if remove_country(locale) == language
]
item_id = self.locales[locales[0]]
if all(self.locales[locale] == item_id for locale in locales):
self.locales[language] = item_id
for locale in locales:
del self.locales[locale]
# Check if a variant is the same as the non-variant form
# eg: 'de@euro' and 'de'
for variant in list(locale for locale in self.locales if "@" in locale):
base, _, _ = variant.partition("@")
if base in self.locales and self.locales[base] == self.locales[variant]:
del self.locales[variant]
# Eliminate any entries that are just the same as the C locale
for locale in list(self.locales):
if self.locales[locale] == self.default:
del self.locales[locale]
def to_c(self):
src_table = ""
ascii_table = ""
mappings_table = []
mapping_ranges = []
chains_table = []
chain_starts = []
locale_names = ""
locale_index = []
max_lookup = 0
max_localename = 0
for mapping in self.mappings:
mapping_ranges.append((len(mappings_table), len(mapping)))
for key in sorted(mapping):
if len(key) == 1 and ord(key[0]) < 0x8000:
src_range = ord(key[0])
else:
existing = src_table.find(key)
if existing == -1:
start = len(src_table)
assert all(ord(c) <= 0x10FFFF for c in key)
src_table += key
src_range = encode_range(start, len(src_table))
max_lookup = max(max_lookup, len(key))
else:
src_range = encode_range(existing, existing + len(key))
value = mapping[key]
if len(value) == 1 and ord(value[0]) < 0x80:
ascii_range = ord(value[0])
else:
existing = ascii_table.find(value)
if existing == -1:
start = len(ascii_table)
assert all(ord(c) < 0x80 for c in value)
ascii_table += value
ascii_range = encode_range(start, len(ascii_table))
else:
ascii_range = encode_range(existing, existing + len(value))
mappings_table.append((src_range, ascii_range))
for chain in self.chains:
chain_starts.append(len(chains_table))
for item_id in reversed(chain):
assert item_id < 0xFF
chains_table.append(item_id)
chains_table.append(0xFF)
for locale in sorted(self.locales):
max_localename = max(max_localename, len(locale))
name_offset = len(locale_names)
assert all(ord(c) <= 0x7F for c in locale)
locale_names += locale + "\0"
item_id = self.locales[locale]
assert name_offset < 256
assert item_id < 256
locale_index.append((name_offset, item_id))
print("/* Generated by update-gtranslit.py */")
print("#define MAX_KEY_SIZE", max_lookup)
print("#define MAX_LOCALE_NAME", max_localename)
print(
"static const gunichar src_table[] = {",
", ".join(str(ord(c)) for c in src_table),
"};",
)
# cannot do this in plain ascii because of trigraphs... :(
print(
"static const gchar ascii_table[] = {",
", ".join(str(ord(c)) for c in ascii_table),
"};",
)
print(
"static const struct mapping_entry mappings_table[] =",
c_pair_array(mappings_table),
)
print(
"static const struct mapping_range mapping_ranges[] =",
c_pair_array(mapping_ranges),
)
print(
"static const guint8 chains_table[] = {",
", ".join(str(i) for i in chains_table),
"};",
)
print(
"static const guint8 chain_starts[] = {",
", ".join(str(i) for i in chain_starts),
"};",
)
print(
'static const gchar locale_names[] = "'
+ locale_names.replace("\0", "\\0")
+ '";'
)
print(
"static const struct locale_entry locale_index[] = ",
c_pair_array(locale_index),
)
print("static const guint8 default_item_id = %u;" % (self.default,))
def dump(self):
print(self.mappings)
print(self.chains)
print(self.locales)
locales = []
for name in os.listdir(localedir):
if looks_like_locale(name):
chain = get_chain(name)
locales.append(chain)
chain.links += 1
serialiser = Serialiser()
for locale in locales:
serialiser.add_locale(locale.name, locale.serialise(serialiser))
i18n = get_chain("i18n").serialise(serialiser)
combining = get_chain("translit_combining").serialise(serialiser)
serialiser.add_default(serialiser.add_chain([i18n, combining]))
serialiser.optimise_locales()
serialiser.to_c()