Currently a new connection will not be attempted until the previous
one has timed out and as the current API only exposes a single
timeout value in practice it often means that it will wait 30 seconds
(or forever with 0 (the default)) on each connection.
This is unacceptable so we are now trying to follow the behavior
RFC 8305 recommends by making multiple connection attempts if
the connection takes longer than 250ms. The first connection
to make it to completion then wins.
As RFC 8305 recommends we can start multiple DNS queries in parallel
to more quickly make an initial response, especially when one is
particularly slow/broken.
This allows higher levels to have more control over resolving
(ipv4 or ipv6 for now) which allows for optimizations such
as requesting both in parallel as RFC 8305 recommends.
This means the output (including lists of filenames) does not depend on
the order of the input files, which may matter if this tool is invoked
with a glob or some other mechanism that doesn't guarantee an order.
Turns out the fix in commit 93555577c wasn't enough, when using glib as
subproject and the parent project uses only libgio_dep, and include
<gi18n.h>, it won't find libintl.h because it's in the
include_directories of libglib_dep. Fix that by declaring dependencies
explicitly, which is the right thing to do since glib and gobject are
public dependencies of gio. That reflects what we do for the pkg-config
file as well.
Previously, method and signal arguments were sorted by name, which
(assuming you don't happen to give your arguments
lexicographically-ordered names) means the generated signatures were
incorrect when there is more than 1 argument.
While sorting the methods and signals themselves (and properties, and
annotations on all these) is fine, it's easiest to not sort anything.
Since 1217b1bc4f, LICENSE_STR has taken two
parameters, not one. Without this change, running either mode fails
with a traceback like:
Traceback (most recent call last):
File "../gdbus-codegen", line 55, in <module>
sys.exit(codegen_main.codegen_main())
File ".../codegen_main.py", line 294, in codegen_main
gen.generate()
File ".../codegen.py", line 896, in generate
self.generate_body_preamble()
File ".../codegen.py", line 682, in generate_body_preamble
self.outfile.write(LICENSE_STR.format(config.VERSION))
IndexError: tuple index out of range
8916874ee6, which introduced these flags,
was actually merged after that commit, but I assume it was written
beforehand.
This is to ensure that the generated code is still compilable by the
running compiler, and see whether we can read the things in there
properly.
See issue #1580.
glib-compile-resources was updated to generate octal byte
representation in a string, but unfortunately this breaks the build
on Visual Studio when the generated string sequence exceeds 65535
characters, which is the imposed limit on Visual Studio compilers.
To make things work on Visual Studio builds and to not slow down things
on other compilers, generate a code path for Visual Studio an array of
octal byte representations, as well as a code path for other compilers
that use the new string representation of octal values, and let the
compiler take the appropriate code path when compiling the
generated code.
Fixes issue #1580.
While uniqueness is great, sometimes you want to restart
a newer version of the same app. These two flags make that
possible.
We also add a ::name-lost signal, that is emitted when it
happens. The default handler for this signal just calls
g_application_quit(), but applications may want to connect
and do cleanup or state-saving here.
When using glib as subproject we are forced to pass glib_dep,
gobject_dep and gio_dep to any build target. If we pass only gio_dep it
will missing include directory for glib and gobject.
Putting the raw URIs in the documentation comments would not link them,
and the ‘%20’s in the URIs were being parsed by gtk-doc as symbol
references. Fix that by using Markdown to format them correctly as
links.
Signed-off-by: Philip Withnall <withnall@endlessm.com>
Using `%` indicates that you’re linking to a symbol. In these cases we
wanted some nicely formatted literals, or a link to a specific property.
Use backticks for the literals, and link to the property fully.
Signed-off-by: Philip Withnall <withnall@endlessm.com>
Due to the line wrapping, gtk-doc was interpreting this second line as
redefining the @dbus_register documentation.
Signed-off-by: Philip Withnall <withnall@endlessm.com>
Otherwise they fill your real ~/.config/mimeapps.list with rubbish and
race for access to it. This is arguably not good.
Signed-off-by: Philip Withnall <withnall@endlessm.com>
https://gitlab.gnome.org/GNOME/glib/issues/1514
If a D-Bus interface was annotated with o.fd.DBus.Deprecated, then the
corresponding GObject property, in the common GInterface implemented
by the generated GDBusObjectProxies and GDBusObjectSkeletons, to
access the generated code for the D-Bus interface was not being marked
with G_PARAM_DEPRECATED, even though the gtk-doc snippet had the
'Deprecated: ' tag.
G_PARAM_DEPRECATED is older than gdbus-codegen, 2.26 and 2.30
respectively, hence it can be used unconditionally.
https://gitlab.gnome.org/GNOME/glib/merge_requests/485
If a D-Bus interface has a property that's annotated with
o.fd.DBus.Deprecated, then the corresponding GObject property was not
being marked with G_PARAM_DEPRECATED, even though the gtk-doc snippet
had the 'Deprecated: ' tag.
G_PARAM_DEPRECATED is older than gdbus-codegen, 2.26 and 2.30
respectively, hence it can be used unconditionally.
https://gitlab.gnome.org/GNOME/glib/merge_requests/485
Allowing unsafe rehandshakes used to be required for web compatibility,
but this is no longer a concern in 2018. So there should no longer be
compatibility benefits to calling this function. All it does is make
your TLS connection insecure.
Also, rehandshaking no longer exists at all in TLS 1.3.
At some point (maybe soon!) glib-networking will begin ignoring the
rehandshake mode, so let's deprecate it now.
Let's entirely deprecate calling this function for rehandshaking. The
current documentation is OK, but guarantees defined behavior (to attempt
a rehandshake) when TLS 1.2 is in use. But there's no way to force TLS
1.2, and also no way to check which version of TLS is in use. I really
should have deprecated use of this function for rehandshaking entirely
last time I updated it.
Fortunately, there should be no compatibility risk for existing code,
because rehandshaking has no visible effects at the API level.
With the changes to limit GVariant type nesting (commit 7c4e6e9fbe),
it’s now possible to have a valid type signature which is not a valid
GVariant type when enclosed in parentheses (to make it a tuple).
Check for that when parsing the signature field in a D-Bus message.
Includes a unit test.
oss-fuzz#11120
Signed-off-by: Philip Withnall <withnall@endlessm.com>
This test is intended to verify the fix for
https://bugzilla.gnome.org/show_bug.cgi?id=787731, which was that
g_file_query_filesystem_info() would return stale information for the
mount. After replacing a read-only mount with a read-write mount, this
test used to only fail if G_FILE_ATTRIBUTE_FILESYSTEM_READONLY was TRUE
and yet the file could be opened for writing. In particular, if (due to
a test bug) the file really was still on a read-only filesystem, the
test would pass.
Now that we have fixed that bug in the test, we can make a stronger
assertion.
fusermount -z behaves like umount --lazy, which is documented thus:
> Detach the filesystem from the file hierarchy now, and clean up all
> references to this filesystem as soon as it is not busy anymore.
Without this, the call to `fusermount -u` often fails with:
/usr/bin/fusermount: failed to unmount /home/wjt/src/gnome/glib/_build/dir_bindfs_mountpoint: Device or resource busy
which causes the subsequent call to bindfs to fail:
fuse: mountpoint is not empty
fuse: if you are sure this is safe, use the 'nonempty' mount option
It's not clear what is causing the mount to be busy. Inserting a
g_usleep (100 * 1000) before the calls to `fusermount -u` also works to
make the problem go away, but for the purposes of this test the
important point is that the mount is detached from the directory, for
which a lazy unmount is fine.
Fixes#1590.
In practice, fusermount -u often fails:
/usr/bin/fusermount: failed to unmount /home/wjt/src/gnome/glib/_build/dir_bindfs_mountpoint: Device or resource busy
which causes the subsequent calls to bindfs to fail:
fuse: mountpoint is not empty
fuse: if you are sure this is safe, use the 'nonempty' mount option
This may or may not cause the current test run to fail, but it reliably
causes a repeat run of the test to fail. This change causes the current
run to fail instead.
I made a mistake when last updating the documentation in 94a99ae9. I
wrote that, with TLS 1.3, this would perform a rekey instead of a
rehandshake. In fact, that's only true for client connections. For
server connections, it's a no-op.
I was a bit nervous about how to document the behavior anyway, because
we really don't know what behavior will be reasonable with non-GnuTLS
crypto backends. This behavior is reasonable for the GnuTLS backend, but
might not necessarily make sense for OpenSSL. Ideally, we would
discourage API users from doing things which could have unexpected
effects, so instead of documenting what the GnuTLS backend does, I think
it'd be better to document that this is "undefined but not dangerous,"
since of course we want to make sure that existing code that doesn't
know about TLS 1.3 is not broken.
The docs sound like settings list is a thing, and
a ::children-changed signal exists. That is not the
case, and will never be the case at this point, so
stop pretending.
Closes: #1362
Currently, GDBusProxy:g-name-owner only notifies changes to the unique
name owner of the remote object in case the proxy was constructed for a
well-known name.
That sounds like an artificial restriction, and it's convenient to
connect to notify::g-name-owner if a proxy instance has already been
created for an unique name, instead of additionally using
g_bus_watch_name() to track the owner.
To fix this, always connect to NameOwnerChanged after the proxy is
initialized, instead of only doing so when the proxy was constructed for
a well-known name.
https://bugzilla.gnome.org/show_bug.cgi?id=791316https://gitlab.gnome.org/GNOME/glib/issues/1310
test_c_args is defined in the root meson.build with unfiltered list of
compiler flags, then redefined in gio/tests/meson.build after the
subdir() call. Move it before.
The length of the stolen data from a memory output stream is given by
get_data_size() — get_size() can be larger, and hence cause unnecessary
overallocation.
Similarly to g_source_set_name(), this sets a name on a GTask for
debugging and profiling. Importantly, this name is propagated to the
GSource for idle callbacks for the GTask, ending the glorious reign of
`[gio] complete_in_idle_cb`.
The name can be queried using g_task_get_name(). Locking is avoided by
only allowing the name to be set before the GTask is used from another
thread.
Includes tests.
Signed-off-by: Philip Withnall <withnall@endlessm.com>
This is the most degenerate possible test but it does exercise this code
path.
(Tweaked by Philip Withnall <withnall@endlessm.com> to also add the flag
to the autotools build.)
The executable depends only on the generated header file at compile
time, and on the library at link time. So meson can decide to compile
gdbus-peer.c before compiling the library and thus won't have generated
the header yet, causing the build error.
So declare_dependency() should only have the header file in its sources,
to force generating files before compiling gdbus-peer.c without
including objectmanager-gen.c into gdbus-peer's sources.
We use libgdbus_example_objectmanager_dep as dependency for various
tests, but this implies only a link dependency while it doesn't ensure
that the sources are generated.
Make this explicit
mntent-based implementation filter out mounts with device path that was
repeated. Consequently, it is not possible to show such mounts in UI even
with x-gvfs-show, because they are not returned from g_unix_mounts_get.
libmount-based implementation currently doesn't filter out any mounts
which causes issues to our volume monitors. Let's rather mark mounts
which don't point into fs root as system_internal. This approach won't be
affected by mount order as is mntent-based implementation. It will mark
more mounts as system_internal than it is filtered out with mntend-based
implementation, but there will be always possibility to show them in UI
over x-gvfs-show, which was not possible with mntend-based. We can
probably introduce some improvements later to not mark unique mounts as
system internal even if they don't point into fs root...
https://gitlab.gnome.org/GNOME/glib/issues/1271
Currently, there isn't API to determine root path for mounts created
over bind operation (or btrfs subvolumes). This causes issues to our
volume monitors if there is multiple mounts for one device, which can
happen with libmount-based implementation currently. Let's propagate
root path from libmount over g_unix_mount_get_root_path, so we can
handle this somehow in our volume monitors.
https://gitlab.gnome.org/GNOME/glib/issues/1271
Currently, readlink() is used only 12 times when expanding symlinks.
However, kernel uses 40 for this purpose and it is defined as MAXSYMLINKS.
Use that constant if available, or 40. See:
https://github.com/torvalds/linux/include/linux/namei.h.
find_mountpoint_for() uses current file in case of error, because
get_parent() returns NULL for error, but also if parent doesn't exist.
Return "." from get_parent() if parent doesn't exist in order to
differentiate the error state.
Test symlink expansion in find_mountpoint_for() function over
_g_local_file_find_topdir_for(). find_mount_for() is crucial for many
of glocalfile.c functionality (e.g. to determine correct trash location)
and symlink expansion has to work properly.
https://gitlab.gnome.org/GNOME/glib/issues/1522
These were callers which explicitly specified the string length to
g_utf8_validate(), when it couldn’t be negative, and hence should be
able to unconditionally benefit from the increased string handling
length.
At least one call site would have previously silently changed behaviour
if called with strings longer than G_MAXSSIZE in length.
Another call site was passing strlen(string) to g_utf8_validate(), which
seems pointless: just pass -1 instead, and let g_utf8_validate()
calculate the string length. Its behaviour on embedded nul bytes
wouldn’t change, as strlen() stops at the first one.
Signed-off-by: Philip Withnall <withnall@endlessm.com>
This introduces no real functional changes (except when compiling with
G_DISABLE_ASSERT, in which case it fixes the test). Mostly just a code
cleanup.
Signed-off-by: Philip Withnall <withnall@endlessm.com>
The code was checking whether the signature provided by the blob was a
valid D-Bus signature — but that’s a superset of a valid GVariant type
string, since a D-Bus signature is zero or more complete types. A
GVariant type string is exactly one complete type.
This meant that a D-Bus message with a header field containing a variant
with an empty type signature (for example) could cause a critical
warning in the code parsing it.
Fix that by checking whether the string is a valid type string too.
Unit test included.
oss-fuzz#9810
Signed-off-by: Philip Withnall <withnall@endlessm.com>
Parsing a D-Bus message with the signature field in the message header
of type other than ‘g’ (GVariant type signature) would cause a critical
warning. Instead, we should return a runtime error.
Includes a test.
oss-fuzz#9825
Signed-off-by: Philip Withnall <withnall@endlessm.com>
G_FILE_ATTRIBUTE_ACCESS_CAN_TRASH can be set to a wrong value if
its parent dir is a symlink. This is because the find_mountpoint_for()
function tries to find mountpoint for a filepath and expands symlinks
only in parent dirs. But in this case the path is already parent dir
and needs to be expanded first...
Closes: https://gitlab.gnome.org/GNOME/glib/issues/1522
Fedora is using https://fedoraproject.org/wiki/Changes/Annobin
to try to ensure that all objects are built with hardening flags.
Pass down `CFLAGS` to ensure the SystemTap objects use them.
There were tests for invalid UTF-8 output when asynchronously
communicating with a subprocess, but nothing for synchronous
communication. Add such a test, and refine the code as a result.
Signed-off-by: Philip Withnall <withnall@endlessm.com>
Instead of sometimes returning a non-NULL buffer, always return NULL.
However, keep the documentation as explicitly returning undefined values
on failure, so that we can change the behaviour in future if needed.
The return values weren’t defined for failure before, so were
implicitly returning undefined values.
Signed-off-by: Philip Withnall <withnall@endlessm.com>
Put the core readlink() code into a separate
_g_win32_readlink_handle_raw() function that takes a file handle,
can optionally ensure NUL-terminatedness of its output
(for cases where we need a NUL-terminator and do *not* need
to get the exact contents of the symlink as it is stored in FS)
and can either fill a caller-provided buffer *or* allocate
its own buffer, and can also read the reparse tag.
Put the rest of readlink() code into separate
functions that do UTF-16<->UTF-8, strip inconvenient prefix
and open/close the symlink file handle as needed.
Split _g_win32_stat_utf16_no_trailing_slashes() into
two functions - the one that takes a filename and the one
that takes a file descriptor. The part of these functions
that would have been duplicate is now split into the
_g_win32_fill_privatestat() funcion.
Add more comments explaining what each function does.
Only g_win32_readlink_utf8(), which is callable from outside
via private function interface, gets a real doc-comment,
the rest get normal, non-doc comments.
Change all callers to use the new version of the private
g_win32_readlink_utf8() function, which can now NUL-terminate
and allocate on demand - no need to call it in a loop.
Also, the new code should correctly get reparse tag when the
caller does fstat() on a symlink. Do note that this requires
the caller to get a FD for the symlink, not the target. Figuring
out how to do that is up to the caller.
Since symlink info (target path and reparse tag) are now always
read directly, via DeviceIoControl(), we don't need to use
FindFirstFileW() anymore.
On Windows NTFS symlinks are implemented as reparse points,
which are special kinds of files *or directories*. A directory
symlink should link to a directory. A file symlink should link
to a file. Mismatching (such as a file symlink pointing to a
directory) produces symlinks that simply do not function.
Therefore GFileType file vs directory vs symlink distinction is
too simplistic to correctly represent a NTFS filesystem object type.
Since we can't turn back time and choose a better way of representing
file types, make GFileType reflect the file vs directory type on
Windows, meaning that all FS objects are either files or
directories (or shortcuts, which are also files), but never symlinks.
A test for symlinkiness will have to be made via GFileInfo - it
tracks symlinkiness separately from file/directory/whatever.
The st_nlink field of the stat structure has meaning and should
be put into GFileInfo.
The st_mode field is far less meaningful, but could still be used
for some purposes, adjust the comment to clarify that.
When using glib as a meson subproject on Windows the build currently fails
due to too long paths during the build process. See
https://github.com/mesonbuild/meson/issues/4226 for the upstream bug.
To work around the issue shorten the filenames of the generated gdbus files.
The tests were making assumptions about the order of the returned D-Bus
introspection nodes. However, these are semantically unordered and
changes to e.g. GHashTable would break the tests.
Fix this by applying a sort prior to validation.
There are a lot of gbooleans in the private GTask struct, which seems a
bit wasteful. Use a bitfield to compress the struct a bit.
This reduces the size of the struct from 216 bytes to 168 bytes on my
64-bit machine.
One of the fields needs to remain separate, since it’s used from a
TRACE() macro which calls typeof() on it.
Signed-off-by: Philip Withnall <withnall@endlessm.com>
These functions already check to see if a successful result has already
been returned; expand them to also check to see if an error has been
returned.
We can’t just check GTask.result_set, as that’s actually an indicator
for whether the GTask.result member is filled — when
g_task_propagate_*() is called, it’s cleared again. We need a new state
bit.
Signed-off-by: Philip Withnall <withnall@endlessm.com>
https://gitlab.gnome.org/GNOME/glib/issues/1525
This method drops the last reference *it* owns to the GTask, but then
continues to call methods on the GTask. This wasn’t resulting in
failures because a ref in another thread kept the GTask alive, but
that’s quite dodgy.
Signed-off-by: Philip Withnall <withnall@endlessm.com>
The X-Flatpak-RenamedFrom key is used in .desktop files to identify past
names for the desktop file. It is defined to be a list of strings.
However, there was previously no correct way to retrieve a list of
strings from the GKeyFile wrapped by GDesktopAppInfo, short of
re-parsing the file with GKeyFile.
Note that doing something like:
g_strsplit (g_desktop_app_info_get_string (...), ";", -1)
is not correct: the raw value "a\;b;" represents the one-element list
["a;b"], but g_key_file_get_string() rejects the sequence "\;", and so
g_desktop_app_info_get_string() returns NULL in this case. (Of course, a
.desktop file with a semicolon in its name is a pathological case.)
Add g_desktop_app_info_get_string_list(), a trivial wrapper around
g_key_file_get_string_list(), similar to g_desktop_app_info_get_string()
and co.
The change from g_key_file_free() to g_key_file_unref() in the test is
needed because g_key_file_free() clears the contents of the keyfile.
This is fine for all the fields which are eagerly loaded and copied into
GDesktopAppInfo, but not when we want to access arbitrary stuff from the
keyfile.
This avoids the convenience library being treated as though it was
an installed static library (objects not included in the dependent
static library, and convenience library being listed in the pkg-config
metadata), both of which would make static linking impossible.
This is a workaround for meson not having
https://github.com/mesonbuild/meson/pull/3939 merged yet.
Fixes: https://gitlab.gnome.org/GNOME/glib/issues/1536
Signed-off-by: Simon McVittie <smcv@collabora.com>
This makes it easier to debug test failures, by ensuring that g_debug()
and g_test_message() are printed as TAP diagnostics.
Fixes: https://gitlab.gnome.org/GNOME/glib/issues/1528
Signed-off-by: Simon McVittie <smcv@collabora.com>
None of these files starts with a #! line, and they are not native
binary executables, so if a user attempts to execute them as a program,
Unix shells will run them as /bin/sh scripts. This is not going to end
well, since none of them are shell scripts (the gio bash completion
is for bash, which is not a lowest-common-denominator POSIX shell, and
in any case is designed to be sourced rather than executed).
Fixes: #1539
Signed-off-by: Simon McVittie <smcv@collabora.com>
These keep on taking just longer than 30s on my local machine when run
in parallel with the rest of the tests (i.e. with `ninja test`). Testing
them individually, they do terminate correctly.
Signed-off-by: Philip Withnall <withnall@endlessm.com>
g_icon_new_for_string() docs states that it should return a single name
when created with a single name. I add a second condition to this case:
the themed icon must not include default fallbacks (i.e. it must not
have been created with `g_themed_icon_new_with_default_fallbacks()`).
Otherwise the return value of `g_icon_new_for_string()` would not
recreate the same icon list when passed to `g_icon_new_for_string()`
(which would be another documentation inconsistency).
g_icon_new_for_string() is now back to old behavior for this specific
case.
I also revert the unit test for this case, and add a new unit test when
using g_themed_icon_new_with_default_fallbacks() with a single name as
well.
Closes#1513.
Currently this function calls `g_warning()` explicitly. It would be
nicer to properly propagate these failure up to the caller that tried to
initialise us.
`read_netlink_messages()` is the callback attached to the netlink socket
(G_IO_IN). It calls `g_socket_receive_message()`. There is a race
condition that if the socket is closed while there is a pending call, we
will try to receive on a closed socket, which fails.
To avoid this, we switch the order of the operations around: first
destroy the source and then close the socket.
This is not a correct way to check if `g_socket_new_from_fd()` failed.
Instead just see if it returned `NULL` itself.
This was preventing the netlink monitor from being initialised.
Closes#1518
All the other initialisation failure paths set a GError, but this one
didn’t. Set a GError to avoid breaking the invariant that returning
FALSE should always have a GError set.
Signed-off-by: Philip Withnall <withnall@endlessm.com>
https://gitlab.gnome.org/GNOME/glib/issues/1523
This is a speculative fix for epiphany#533, which we think might be
caused by xdg-desktop-portal not ever being started. This service is
started on-demand, not automatically.
1) Remove the non-Windows-only condition for subdir('tests').
2) Add libiphlpapi, libws2_32 and libsecur32 deps, needed for W32 tests.
3) Remove the -no-undefined argument (gcc doesn't understand it,
it *does* understand -Wl,-no-undefined; either way, the test
compiles without this argument just fine; maybe meson adds it
by itself - you can hardly build shared modules without it).
4) Add or fix a number of includes
5) Disable gdbus-objectmanager tests when building with MSVC
(right now these tests don't work on Windows anyway, so the fact
that MSVC can't even build them properly is irrelevant;
most likely gdbus-codegen needs changes to put _GLIB_EXTERN
before each function)
G_FILE_ATTRIBUTE_DOS_IS_MOUNTPOINT allows mountpoints
(NTFS reparse points with IO_REPARSE_TAG_MOUNT_POINT tag) to
be told apart from symlinks (NTFS reparse points with
IO_REPARSE_TAG_SYMLINK tag), even though both are reported
by glib as "symlinks".
G_FILE_ATTRIBUTE_DOS_REPARSE_POINT_TAG allows the exact
reparse tag value to be obtained by the user. This way
even more exotic reparse points can be identified and
handled by the user (glib itself currently has no code
to work with any reparse points that are not symlinks
or mountpoints).
The existing code was generating code with undefined results that modern compilers warn about:
accounts-generated.c:204:23: warning: dereferencing type-punned pointer will break strict-aliasing rules [-Wstrict-aliasing]
(GDBusArgInfo **) &_accounts_accounts_method_info_list_cached_users_OUT_ARG_pointers,
It wasn’t being tested. It should behave the same as
g_list_model_get_item(), so write a wrapper for the two.
This brings the code coverage of glistmodel.c up to 100%.
Signed-off-by: Philip Withnall <withnall@endlessm.com>
When setting the GListStore:item-type property we need to check the
GType is a GObject (or subclass). There was some explicit code for this,
but when actually testing it and looking at the code coverage, it turns
out that the GObject property type check coming from
g_param_spec_gtype() does everything we want, and the custom
g_critical() can never be hit. So turn it into an assertion.
Signed-off-by: Philip Withnall <withnall@endlessm.com>
This is also faster, though I doubt anyone's able to measure it.
The previous code was a more complicted way to do the same thing and it
was likely written the more complicated way because it fell out commit
758d7073a9 when fixing
https://bugzilla.gnome.org/show_bug.cgi?id=795307
There are many cases where a default TLS database is not able to be
defined within the constraints of a system. For example glib-networking
(or glib-openssl) cannot retrieve the default certificate store on iOS
or Android and need to be initialized from a cert file of certificates
bundled with the application.
Previously GStreamer was relying on a custom patch to glib-networking to
populate the default database from the file pointed to by the
CA_CERTIFICATES environment variable however the mechanism that enabled
this was recently remove from glib-networking.
Adding a more generic g_tls_backend_set_default_database() API allows
application developers to override the default database using their own
certificates as well as allowing equivalent functionality on Android/iOS
(or others) as on the default database handling Linux.
Fixes https://gitlab.gnome.org/GNOME/glib-networking/issues/35
The ::network-changed signal is documented to indicate any change in
network configuration, which doesn't necessarily imply a property
change - additional services becoming available after connecting to
a VPN comes to mind for instance.
In order to match the "native" network monitor's behavior, always
emit the signal when it's in response to the 'changed' D-Bus signal.
Also emit the signal unconditionally when loading the initial property
values, to allow clients to differentiate between "offline" meaning
"offline" and "offline" meaning "uninitialized".
With 0d685b4946, we now encode resource
data as a string. Strings have trailing nul terminators. A C compiler
will happily ignore the fact that the nul terminator exceeds the stated
array length, and will drop it — but a C++ compiler won’t, and will
raise:
error: initializer-string for array of chars is too long [-fpermissive]
Fix that by increasing the array length by 1, and subtracting it again
in the GStaticResource struct.
Signed-off-by: Philip Withnall <withnall@endlessm.com>
This reverts commit 3c1902fcf9.
This was accidentally re-added from an old version of the branch before
!265 was merged. It should not have been re-added.
Rearrange the code so we try version 3 first,
falling back to version 2 and then version 1.
We still do a construct-time check to ensure
that we work with unsupported versions.
Note that this also takes care of setting the
initial property values in the version 1 case.
Version 3 of the network monitor portal interface adds
a CanReach method. Use it to implement can_reach.
The docs state that can_reach will either return TRUE
or set an error. So, set an error of G_IO_ERROR_HOST_UNREACHABLE
when the portal returns FALSE for CanReach.
GSettings XML schema files are installed in a well known directory
under Glib's installation directory: `glib-2.0/schemas`. However,
the Glib installation directory might vary, so the exact location of
the schema files might be unknown.
The information regarding this directory has been added to GIO's
pkg-config file, so it can be checked, and also overrided, by using
the command line utility.
The source callback for a GCancellable should have the cancellable itself
as first argument.
This was not the case, and when this code was hit, we were instead trying
to treat the pointer as a CommunicateState reference and thus wrongly
deferencing it, causing a memory error and a crash.
Some Testing revealed encoding resource data with string
escape codes to compile significantly quicker compared
to the same data encoded as an array with hexadecimal numbers.
See #1489
This is a follow-up to commit 614adf8a75,
which started generating two new files as part of the test; they need to
be cleaned up before distcheck will pass.
Ideally, the test should run a temporary directory and wipe that
directory itself before exiting, but that’s a bit of a big change to
make right now. Deferred to
https://gitlab.gnome.org/GNOME/glib/issues/1495.
Signed-off-by: Philip Withnall <withnall@endlessm.com>
In file gio/gtestdbus.c, function watch_parent, there is a loop which
waits for commands sent from the parent process and kills all processes
recorded in 'pids_to_kill' array on parent process exit. The detection
of parent process exit is done by calling g_poll and checking whether
the returned event is G_IO_HUP. However, 'revents' is a bit mask, and
we should use a bitwise-AND check instead of the equality check here.
It seems to work fine on Linux, but it fails on FreeBSD because the
g_poll returns both G_IO_IN and G_IO_HUP on pipe close. This means the
watcher process continues waiting for commands after the parent process
exit, and g_io_channel_read_line returns G_IO_STATUS_EOF with 'command'
set to NULL. Then the watcher process crashes with segfault when calling
sscanf because 'command' is NULL. Since the test result is already
reported by the parent process as 'OK', this kind of crash is likely to
be unnoticed unless someone checks dmesg messages after the test:
pid 57611 (defaultvalue), uid 1001: exited on signal 11
pid 57935 (actions), uid 1001: exited on signal 11
pid 57945 (gdbus-bz627724), uid 1001: exited on signal 11
pid 57952 (gdbus-connection), uid 1001: exited on signal 11
pid 57970 (gdbus-connection-lo), uid 1001: exited on signal 11
pid 57976 (gdbus-connection-sl), uid 1001: exited on signal 11
pid 58039 (gdbus-exit-on-close), uid 1001: exited on signal 11
pid 58043 (gdbus-exit-on-close), uid 1001: exited on signal 11
pid 58047 (gdbus-exit-on-close), uid 1001: exited on signal 11
pid 58051 (gdbus-exit-on-close), uid 1001: exited on signal 11
pid 58055 (gdbus-export), uid 1001: exited on signal 11
pid 58059 (gdbus-introspection), uid 1001: exited on signal 11
pid 58065 (gdbus-names), uid 1001: exited on signal 11
pid 58071 (gdbus-proxy), uid 1001: exited on signal 11
pid 58079 (gdbus-proxy-threads), uid 1001: exited on signal 11
pid 58083 (gdbus-proxy-well-kn), uid 1001: exited on signal 11
pid 58091 (gdbus-test-codegen), uid 1001: exited on signal 11
pid 58095 (gdbus-threading), uid 1001: exited on signal 11
pid 58104 (gmenumodel), uid 1001: exited on signal 11
pid 58108 (gnotification), uid 1001: exited on signal 11
pid 58112 (gdbus-test-codegen-), uid 1001: exited on signal 11
pid 58116 (gapplication), uid 1001: exited on signal 11
pid 58132 (dbus-appinfo), uid 1001: exited on signal 11
If the watcher process crashes before killing the dbus-daemon process
spawned by the parent process, the dbus-daemon process will keep running
after all tests complete. Due to the implementation of 'communicate'
function in Python subprocess, it causes meson to crash. 'communicate'
assumes the stdout and stderr pipes are closed when the child process
exits, but it is not true if processes forked by the child process
doesn't exit. It causes Python subprocess 'communicate' function to
block on the call to poll until the timeout expires even if the test
finishes in a few seconds. Meson assumes the timeout exception always
means the test is still running. It calls 'communicate' again and
crashes because pipes no longer exist.
https://gitlab.gnome.org/Infrastructure/GitLab/issues/286https://github.com/mesonbuild/meson/issues/3967https://bugs.python.org/issue30154
Previously, glocalfile.c would not set file system metadata for
the free/used key for file systems which reported 0 free space. This is
because some file systems don’t set that metadata when you call
statfs(), so we can’t reliably report it. However, some do, and they
can legitimately set f_bavail and f_bfree to 0 if the file system is
full.
In order to avoid that, always set the file system metadata unless the
file system is FUSE or ncpfs.
This is a partial revert of commit 0b9f24c1e1: instead of the changes
made in that commit, I think we should maintain a blacklist of file
systems which are known to not correctly report free space.
Signed-off-by: Philip Withnall <withnall@endlessm.com>
https://gitlab.gnome.org/GNOME/glib/issues/328
Similar issue was fixed with commit f929d148, but it's happening again.
Define G_MESSAGES_DEBUG=all when running CI to ensure we won't regress
anymore.
In order to determine whether to trash a file to the home directory, we
compare its st_dev to our home directory's st_dev field.
This is the wrong thing to do on overlayfs when deleting files, because
st_dev contains the ID of the filesystem providing the file (which can
be the lower or upper filesystem), but directories always return the ID
of the overlayfs. Thus the comparison fails and we are unable to trash
the file.
Fix this by checking st_dev of the parent directory when we are deleting
a file.
Also adjust `test_trash_not_supported` for this - make its st_dev check
look at the parent directory's `st_dev` rather than the temporary file's
own.
Fixes#1027.
The documentation was unclear about what error codes would be returned
on attempting to open an empty or corrupt GVDB file. Previous versions
of the documentation incorrectly said that corrupt GVDB files were
considered equivalent to empty ones.
A recent commit has clarified the documentation to include its error
handling behaviour.
Update the two users of GVDB within GLib, GResource and GSettingsSource,
to follow this change, and add unit tests for them both.
Other users of the GVDB copylib will need to update their copy and make
appropriate changes if they have bugs in their handling of this
situation. dconf is one example of this. GVDB should be updated from
https://gitlab.gnome.org/GNOME/gvdb.
Signed-off-by: Philip Withnall <withnall@endlessm.com>
https://gitlab.gnome.org/GNOME/glib/issues/1454
While mountpoints are *not* symlinks, strictly speaking,
they works in a similar enough way, so consider them to be
symlinks for the purpose of querying local file info.
On Windows st_ctime field is the file creation time.
POSIX mandates that field to be the file state change time.
Naturally, glib code interpreted st_ctime as POSIX suggested,
and the result was bad.
Fix this by introducing special W32-only logic for setting
attributes from st_ctime field.
Fixes issue #1452.
We now send the fallback SCSV, meaning use of this function will cause
modern servers to immediately terminate the connection, so let's warn
API users to expect that behavior and be crystal clear that this
function should only be used as a fallback when a normal connection
attempt has already failed.
Also, the documentation is mostly duplicated between the property and
the function, so let's just reference the function documentation from
the property.
7efd76dd67 added these configure time tests to work around a bug
with older Android. Since the test didn't take Windows into account it
wrongfully applied the workaround on Windows too, breaking the build.
With meson this wasn't an issue since the check is skipped on Windows there
and our CI didn't catch this issue.
Change the test to run on Android only for meson and autotools.
This also makes it clear that the test+code can be dropped again if we stop
supporting older Android versions at some point.
GLib currently tries to use FAM volume monitor for monitoring files
within home on NFS. If FAM support is not available, it fallbacks by
default to GInotifyFileMonitor. I think we should fallback to
GPollFileMonitor instead, because inotify is not reliable on NFS,
which may cause issues for dconf. With this patch, it should be safe to
not build libgiofam and still be sure that dconf works properly if home
is mounted on NFS. I think this might be a first step to remove FAM
support from GLib completely, because gamin is buggy and dead for
several years already. Gamin just polls files on NFS anyway. This
change applies on files only, because GPollFileMonitor seems doesn't
support dirs, however it should be enough for dconf. The other
drawback is that one can't set poll timeout currently. Just a note
that this can still be overwritten by GIO_USE_FILE_MONITOR.
Releasing GVolumeMonitor before g_volume_mount finish cause that
g_volume_get_mount returns NULL, because the mount is not correctly
propagated to the volume.
https://gitlab.gnome.org/GNOME/glib/issues/1458
The gresource code uses libelf if available but that also depends on mmap but isn't
guarded with HAVE_MMAP. This can make the build fail under MSYS2 where a mingw version
of libelf exists but there is no mmap.
Instead of guarting the libelf code with HAVE_LIBELF add a new macro named USE_LIBELF
which is only defined if libelf and mmap support are available.
Also install the mingw libelf version for CI so we catch similar errors in the future.
It is a bug if we distribute files which are generated at build time —
they should be built on the machine which is compiling GLib, not be
shipped in the tarball.
This brings the autotools-generated tarball in line with the
ninja-generated one, with the exception of man pages and gtk-doc HTML
output.
Signed-off-by: Philip Withnall <withnall@endlessm.com>
The documentation claims that g_volume_get_mount should succeed after
g_volume_mount. Let's update mounts before releasing g_volume_mount to
be sure that the mount is added to the corresponding volume. The same
is done in GVfsUDisks2VolumeMonitor.
https://gitlab.gnome.org/GNOME/glib/issues/1458
The network monitor portal interface is changing.
Version 2 is no longer using properties, but getters
instead (this lets the portal apply access control
and avoid sending information to non-networked
sandboxes).
To support both version 1 and 2 of the interface,
we stop using generated code and instead deal with
the api differences in our own code, which is not
too difficult.
Support version 1 as well
The new python module, added with 0.46, works with Python 2 and 3 and
allows to pass a path for the interpreter to use, if the need arises.
Previously the meson build set PYTHON, used in the shebang line of
the scripts installed by glib, to the full path of the interpreter.
The new meson module doesn't expose that atm, but we should set it to
a executable name anyway, and not a full path.
Several of our tools are installed and are used by other projects to
generate code. However, there is no 'install' when projects use glib
as a subproject.
We need some way for glib to 'provide' these tools so that when some
project uses glib as a subproject, find_program('glib-mkenums') will
transparently return the glib-mkenums we just built.
Starting from Meson 0.46, this can be done with the
`meson.override_find_program()` function.
As a bonus, the Meson GNOME module will also use these
'overriden'/'provided' programs instead of looking for them in PATH.
PEP8 says that:
"Comparisons to singletons like None should always be done with is or
is not, never the equality operators."
glib uses a mix of "== None" and "is None". This patch changes all
cases to the latter.
Use g_test_skip() so that the TAP output is correct for the tests,
rather than printing using g_printerr().
Signed-off-by: Philip Withnall <withnall@endlessm.com>
https://gitlab.gnome.org/GNOME/glib/issues/640
Use either g_get_real_time() or g_date_time_new_now_local(). This means
we don’t need to worry about time_t being 32b in future (the year 2038
problem), and it makes the need for error handling a bit more explicit.
Improve the error handling in several cases.
Based on a patch by Niels De Graef
(https://gitlab.gnome.org/GNOME/glib/merge_requests/142).
Signed-off-by: Philip Withnall <withnall@endlessm.com>
https://gitlab.gnome.org/GNOME/glib/issues/1402
Apparently this is needed for building PE libraries. It makes no
difference on Linux, where linking of the GLib symbols in the inotify
file monitor code is done lazily.
Signed-off-by: Philip Withnall <withnall@endlessm.com>
https://gitlab.gnome.org/GNOME/glib/issues/1201
The tests array was being wiped out by an assignment instead of an
append. This adds another 19 tests to what’s typically being run
already.
Signed-off-by: Philip Withnall <withnall@endlessm.com>
This was missed in !137 because some of the GIO tests weren’t being run
under Meson (see following commits).
Signed-off-by: Philip Withnall <withnall@endlessm.com>
Adds tests to cover this case and similar cases for various GResource
methods in future.
Signed-off-by: Philip Withnall <withnall@endlessm.com>
https://gitlab.gnome.org/GNOME/glib/issues/927
When changing the locale using setlocale(), duplicate the old locale
value before updating it, so that we can safely restore the old locale
after running the test.
Signed-off-by: Philip Withnall <withnall@endlessm.com>
The flush data structures were not zero-initialised, which meant the
branch in flush_buffer_thread() was based on an uninitialised condition.
Signed-off-by: Philip Withnall <withnall@endlessm.com>
For the purposes of debugging, it is quite useful for every GSource to
have a name set. Ensure that any GSource we construct inside GLib has a
name set. For GSources which are then returned to the caller, this name
can then be overridden with something even more useful by the caller.
Since this data is only used for debugging, avoid doing any allocations
for it; just use static strings.
https://gitlab.gnome.org/GNOME/glib/issues/1175
This reverts commit 03c324c64a and fixes
the original problem with e004d5f397 that
caused the revert.
We use $(builddir) instead of $(abs_builddir) so that Automake's
dependency generation works correctly.
See !127.
Add an app-launching function which allows standard file descriptors
to be passed to the child process.
This will be used by gnome-shell to pass systemd journal descriptors
as stdout/stderr. gnome-shell's child_setup function can then be
eliminated, which will enable use of the posix_spawn optimized
gspawn codepath for desktop app launching.
In order to use the new posix_spawn gspawn codepath, for more robust
app launching when available memory is low, we need to meet some
conditions.
child_setup needs to be NULL for this optimization to work, so drop
the internal child_setup that is used here. Replace it with a lightweight
wrapper binary (gio-launch-desktop) that sets GIO_LAUNCHED_DESKTOP_FILE_PID
before executing the app.
Adjust PATH for gio tests so that it can execute the new binary from the
build directory.
They’re network file systems, but not system file systems (in the sense
that procfs is a system file system). This fixes them disappearing from
the sidebar in the UI.
Signed-off-by: Philip Withnall <withnall@endlessm.com>
https://gitlab.gnome.org/GNOME/glib/issues/1424
Add a new syntax to override files: if the group name has a ':' in it,
it indicates that we want to override the default values of keys for
only one desktop. For example:
[org.gnome.desktop.interface:Unity]
font-name='Ubuntu 12'
Will override the settings, only if "Unity" is found in
XDG_CURRENT_DESKTOP. Multiple per-desktop overrides can be specified
for a given key: the one which comes first in XDG_CURRENT_DESKTOP will
be used.
https://bugzilla.gnome.org/show_bug.cgi?id=746592
Recognise a new 'd' option in schema keys which gives a dictionary of
per-desktop default values. This dictionary is searched for the items
found in XDG_CURRENT_DESKTOP, in the order. If nothing matches (or if
the option is missing) then the default value is used as before.
This feature was requested by Alberts Muktupāvels and this patch is
based on an approach devised by them.
https://bugzilla.gnome.org/show_bug.cgi?id=746592
There are a couple of different ways (and soon one more) to access the
default value of a key. Clean up the various places that access this to
avoid duplication.
https://bugzilla.gnome.org/show_bug.cgi?id=746592
There were previously no tests for it. These take gmountoperation.c up
to 85.5% coverage of lines.
Signed-off-by: Philip Withnall <withnall@endlessm.com>
https://gitlab.gnome.org/GNOME/glib/issues/1423
check_expected_events is heavily modified in this commit to tolerate
event loss and allow renaming to be reported as creation and deletion.
This fixes test failure on FreeBSD.
Previously, kqueue file monitor only add event sources for directories
regardless of the type of the file being monitored. Doing so may be
possible on inotify, but it is not sufficient on kqueue. Watching a
directory on kqueue doesn't report changes made to files under it, and
we must watch files themselves to get notified. This problem is fixed
by adding a second watch for non-directory file monitors, and the result
is that we are now able to receive 'CHANGED' and 'ATTRIBUTE_CHANGED'
events for non-directory files.
Since having two watches on one file monitor requires many code changes
to work properly, this commit also changes the following things:
- NOTE_ALL macro is now replaced by note_all inline function. Since the
kqueue backend is shared by all BSD operating systems, there are a
few difference between these systems. It is easier to do '#ifdef'
check in a function than in a macro.
- Both g_kqueue_file_monitor_callback and g_kqueue_file_monitor_cancel
now holds a lock before accessing kqueue_sub structs. This fixes a
crash when these two functions are called from different threads,
causing g_kqueue_file_monitor_callback to access freed memory.
- 'mask' variable in g_kqueue_file_monitor_callback is now removed.
The usage of 'mask' was wrong because of the 'mask > 0' check.
'CHANGED' event has value 0 so the 'mask > 0' check made it
impossible to emit 'CHANGED' events.
- kqueue-missing scans can now be triggered from the kqueue event
callback instead of always waiting for 4 seconds.
- Don't remove a file from kqueue on unlink unless its hard link count
has dropped to zero.
- Don't use 'else if' in the check of 'fflags'. It is possible for a
kevent to have multiple flags set.
- Don't use g_file_monitor_emit_event directly. Always use
g_file_monitor_source_handle_event to report events.
Events submitted to g_file_monitor_emit_event are delivered
immediately, but events sent to g_file_monitor_source_handle_event
are scheduled by GLocalFileMonitor. If we mix the two, the order of
events will be wrong and tests will fail.
- Report 'CHANGES_DONE_HINT' immediately after 'CREATED' if the file
created is not a regular file. This is copied from ih_event_callback.
GVfs utils used to have bash completion, which was pretty useful. However,
it hasn't been ported to gio tool unfortunately. GLib provides completion
for various utils already, so it would be nice to provide completion also
for gio tool. I've updated old bash completion code and merged with some
my old unmerged fixes.
The gvfs completion used "gvfs-ls --show-completions" helper. This mentioned
option hasn't been obviously ported to "gio list" and the proposed completion
doesn't add this option in "gio list" to not pollute the codes, but maybe it
is a bit slower as consequence.
The proposed bash completion suggests subcommands, uris and paths including
the remote mounts. It contains some workarounds, especially because of proper
handling of paths with colons and other special chars (like spaces)...
Since it’s deprecated in favour of positional arguments, including it in
the help output is confusing.
Signed-off-by: Philip Withnall <withnall@endlessm.com>
https://bugzilla.gnome.org/show_bug.cgi?id=795304
These generate basic .c and .h files containing the GDBusInterfaceInfo
for a D-Bus introspection XML file, but no other code (no skeletons,
proxies, GObjects, etc.).
This is useful for projects who want to describe their D-Bus interfaces
using introspection XML, but who wish to implement the interfaces
manually (for various reasons, typically because the skeletons generated
by gdbus-codegen are too simplistic and limiting). Previously, these
projects would have had to write the GDBusInterfaceInfo manually, which
is painstaking and error-prone.
The new --interface-info-[body|header] options are very similar to
--[body|header], but mutually exclusive with them.
Signed-off-by: Philip Withnall <withnall@endlessm.com>
https://bugzilla.gnome.org/show_bug.cgi?id=795304
Recursive annotations do seem to be supported, so we should support them
properly in the type system representation. This currently introduces no
behavioural changes, but will be used in upcoming commits.
Signed-off-by: Philip Withnall <withnall@endlessm.com>
https://bugzilla.gnome.org/show_bug.cgi?id=795304
This is to avoid race between dispose() being called on the
GFdoNotificationBackend instance, and any pending operations which are
still waiting on a D-Bus reply when it’s disposed.
(thx to Philip Withnall for pointing that out)
Signed-off-by: Arnaud Rebillout <elboulangero@gmail.com>
The basic test for duplicate icon names was not working fine when taking
into account fallbacks and icon style mix. This fixes it.
Also resolve the few review comments, i.e.: comment explaining
g_themed_icon_update_names() and the priority order applied to icons;
and using guint for 'i'.
Previously, calling:
g_dbus_is_supported_address ("some-imaginary-transport:", NULL)
correctly returned FALSE; but calling:
g_dbus_is_supported_address ("some-imaginary-transport:", &error)
crashed with:
GLib-GIO:ERROR:../gio/gdbusaddress.c:434:g_dbus_is_supported_address:
assertion failed: (ret || (!ret && (error == NULL || *error != NULL)))
This was because, if the address component did not start with a known
transport, no error was set. Fix this, reusing an error string used by
the corresponding else branch in g_dbus_address_connect(), and adjust
the test to pass both NULL and non-NULL GError **s to this function in
every test case. This case:
g_assert (!g_dbus_is_supported_address ("some-imaginary-transport:foo=bar;unix:path=/this/is/valid", NULL));
would have caught this bug with a non-NULL GError **.
The way things were before: a FreedesktopNotification struct is
allocated before the dbus call, and this same struct is possibly re-used
for other dbus calls. If the server becomes unavailable, the callback
will be invoked after the call times out, which leaves a long time where
other dbus calls can happen, re-using the same FreedesktopNotification
as user data. When the first call times out, the callback is invoked,
and the user data is freed. Subsequent calls that used the same user
data will time out later on, and try to free a pointer that was already
freed, hence segfaults.
This bug can be reproduced in Cinnamon 3.6.7, as mentioned in:
<https://github.com/linuxmint/Cinnamon/issues/7491>
This commit fixes that by always allocating a new
FreedesktopNotification before invoking dbus_call(), ensuring that the
callback always have a valid user data.
Signed-off-by: Arnaud Rebillout <elboulangero@gmail.com>
Use a GOnce to make sure we only warn about notifications not being
supported on Windows once, rather than on each attempted notification.
Signed-off-by: Philip Withnall <withnall@endlessm.com>
Helps: https://gitlab.gnome.org/GNOME/glib/issues/1234
On an Android build, API 22, at least, I got a:
> warning: "UNIX_PATH_MAX" redefined
We were currently defining it as:
> #define UNIX_PATH_MAX sizeof (((struct sockaddr_un *) 0)->sun_path)
Whereas Android's headers define this variable of sockaddr_un as:
> char sun_path[UNIX_PATH_MAX];
So by definition, we will still get the right result in the end by just
using the original value of UNIX_PATH_MAX.
C_IN macro was added years ago in bcbaf1bef0, using same value as the
internal code of Android with the reasonning that "some parts of the API
used by the resolver objects is not public in the Android NDK (yet)".
Well since then things are changed, since it is definitely available (at
least on the API 22 of Android which I am using) in the public header
arpa/nameser_compat.h.
Let's just add a #ifndef to handle both cases when you build with an
older or recent API.
Meson has the ability to classify tests according to "suites", a list of
tags. This is especially useful when we want to run specific sets of
tests — e.g. only GLib's tests — instead of the whole test suite. It
also allows us to classify special tests, like "slow" ones, so that we
can only run them when needed.
FIONREAD ioctl on Linux reports the size of payload on UDP sockets.
However, other systems usually add internal header size to the reported
size, which vary between different operating systems and socket types.
To make it work on more systems, we should follow what we do on Windows
instead of using this unreliable FIONREAD ioctl.
This fixes socket test on FreeBSD.
libelf, just like libc, is not a single project. It is an interface
which can be implemented independently by different operating systems.
Therefore, we cannot expect all systems to provide a .pc file, and we
should fallback to cc.find_library and cc.has_function like what we
already do in autotools build.
Non-glibc gettext implementation seems to decide the language from
LC_MESSAGES environment variable instead of LC_MESSAGES locale, so
we should set both environment variable and locale when running tests
which need translation from specific languages.
G_IO_ERROR_NOT_SUPPORTED is used as parameter for g_set_io_error(),
however, errno is expected instead and thus error code is set to 0,
which is wrong. Let's use ENOTSUP instead.
To be honest, I am not sure why, but in some special environments (e.g.
our CI integration) can happen, that file device number is different from
parent device number. Return "Unable to find or create trash directory for
%s" error from g_local_file_trash() in that case and also set
G_FILE_ATTRIBUTE_ACCESS_CAN_TRASH appropriately.
Make the --help output more consistent with the man page, making it more
obvious that --sourcedir only applies to the files referenced in FILE,
not to the location of FILE itself.
Signed-off-by: Philip Withnall <withnall@endlessm.com>
https://gitlab.gnome.org/GNOME/glib/issues/1406
Change G_FILE_ATTRIBUTE_ACCESS_CAN_TRASH logic to be consistent
with recent g_local_file_trash changes, i.e. set this to FALSE for
locations on system-internal mounts.
https://gitlab.gnome.org/GNOME/glib/issues/251
New bugs appears periodically in nautilus/gvfs/glib components that not
all trashed files are shown in trash:///. It used to be problem mostly
for "bind mounts" and btrfs subvolumes only. Currently, it is also
problem for nfs, cifs and other filesystems, which have been recently
added by commmit 0d69462f on the list of system internal filesystems.
This happens because the trash backend doesn't monitor files on system
internal mounts. Such behavior is not against the trash-spec, however,
we should be consistent within GNOME.
This behavior has the nice side-effect that it solves issues with hangs
on network filesystems: https://gitlab.gnome.org/GNOME/glib/issues/605,
because those are currently on the system internal filesystem list.
https://gitlab.gnome.org/GNOME/glib/issues/251
Main changes were:
- take into account that style variants are added to the list of icon
names.
- In the test of icons set with 3 names ("first", "testicon" and
"last"), I also changed "first" to "first-symbolic" so that we can
test in such a case that the variant is the regular icon (only for
this name, out of the 3).
- Finally icon hashes are necessarily changed, and since emblems are
sorted by their hash in emblem icons, I had to invert an order which
was now different in one of the tests.
... a theme icon.
Otherwise fallbacks of the added icon name are not added to the list (if
use-default-fallbacks is set), nor is the regular/symbolic variant. Also
if we do not recreate the finale list from scratch, sorting of icons and
their variants may end up wrong.
To this end, let's keep around the icon names used for initialization,
separate from the finale name list.
When a themed icon is constructed with several input icon names, add the
variants (symbolic as well as level fallbacks) for every icon names, not
only the first one.
The search order is: first icon name, then its level fallbacks, then
second icon name, then its level fallbacks, then all style variants
(symbolic or regular, opposite to requested style) keeping the same
order again.
This fixes the contenttype GIO unit test.
Whatever the preferred icon style is (symbolic, regular or the requested
style), fallbacking to the other style in case of absent variant is
better than not finding any icon at all.
Also style fallbacking should be managed separately from property
"use-default-fallbacks". Default fallbacks are meant for the process of
getting up in context levels (as separated by dashes in icon name). Even
though it also uses dash characters in format, this is a different
concept as the variant of styles.
Without this commit for instance, if an icon only had a symbolic
variant, and the theme had "-gtk-icon-style" set to "regular" while your
GTK+ application requested the regular icon name, you were getting no
icons, and the application would look completely broken.
Now one would at least fallback to the symbolic icon as last resort
(which is infinitely better than having no icons).
Autoconf macro AC_HEADER_MAJOR doesn't define a macro in config.h when
major is defined in sys/types.h. This was not a problem because major
is assumed to be always available. However, commit aefffa3fbc
changes this assumption in order to fix build on systems without major,
which causes code using major to be disabled on systems putting major
in sys/types.h.
This commit defines a new macro MAJOR_IN_TYPES for both autotools and
meson builds to make major useful on these systems again.
gio-querymodules-wrapper.py is copied from glib-networking. This python
wrapper script is needed because meson.build cannot check for DESTDIR
env variable itself, unlike Makefile.am. It is used to update
giomodule.cache file when installing GIO modules like fam.
The previously implementation considered a file to be a mountpoint if
its parent is on a different device. / is its own parent, so by this
definition it is not a mountpoint.
But / is (generally) listed in fstab, and fstab(5) defines the
directories it contains to be mountpoints. This attribute should follow
that definition (and reasonable expectation): the root directory is a
mountpoint.
So, add a special-case for the case where the file's parent has the same
st_dev and st_ino as the file, which is true only at the root.
Test this attribute at / (only on POSIX), /proc (but only on Linux), and
at many files and directories created by the test suite (which cannot be
mountpoints).
Sometimes file monitor events may be slow to emit. Using g_idle_add
makes it less possible for events to be scheduled later than the main
loop quit, preventing test failure caused by missing events.
This fixes test failure on FreeBSD.
The check in _ke_is_excluded, which causes GKqueueFileMonitor to
fallback to GPollFileMonitor when it returns TRUE, was made to prevent
file monitor from blocking unmount of removable drives on systems not
supporting O_EVTONLY flag in open. However, since g_mount_can_unmount
always returns TRUE on Unix-like platforms, the check always returns
TRUE on non-standard mount points, which is very likely to cause all
programs on the desktop to use the polling fallback if GNOME is
installed in a different prefix for development. This makes the desktop
sluggish and results in bad developer experience on *BSD.
Kqueue isn't good at detecting rapid file creation and deletion. It
tends to miss events because events returned by the kernel don't include
filename information. Since the size of struct kevent is fixed, it is
probably not possible to extend the API to include file names without
breaking ABI. Therefore, we disables the test here to avoid test failure
that is impossible to fix in a reliable way.
The test 'file' uses non-standard '--bytes' option when running du,
which may cause error on non-GNU systems. To keep the test working,
we skips the du check as if we don't find a du command when du fails.
-export-dynamic is a libtool flag. It is also supported by GCC as an
undocumented flag, but it is not supported by Clang. Since we don't use
libtool in meson, we should use -Wl,--export-dynamic instead.
In master, it is already possible to build GLib using Visual Studio
using Meson[1] for some time, so we should focus on maintaining only the
Meson build files for building GLib with Visual Studio.
[1]: There are caveats when building with Visual Studio 2008, namely
that one needs to use the mt command to embed the manifests that
are generated with the .exe/DLLs, for all builds, and that in the
case where the compilation hangs on Visual Studio 2008 x64, as a
workaround, should stop the build by terminating all cl.exe tasks
and change the compiler optimization flag from /O2 (full speed) to
/O1 (optimize for size), due to compiler optimization issues.
Change G_FILE_ATTRIBUTE_ACCESS_CAN_TRASH logic to be consistent
with recent g_local_file_trash changes, i.e. set this to FALSE for
locations on system-internal mounts.
https://gitlab.gnome.org/GNOME/glib/issues/251
New bugs appears periodically in nautilus/gvfs/glib components that not
all trashed files are shown in trash:///. It used to be problem mostly
for "bind mounts" and btrfs subvolumes only. Currently, it is also
problem for nfs, cifs and other filesystems, which have been recently
added by commmit 0d69462f on the list of system internal filesystems.
This happens because the trash backend doesn't monitor files on system
internal mounts. Such behavior is not against the trash-spec, however,
we should be consistent within GNOME.
This behavior has the nice side-effect that it solves issues with hangs
on network filesystems: https://gitlab.gnome.org/GNOME/glib/issues/605,
because those are currently on the system internal filesystem list.
https://gitlab.gnome.org/GNOME/glib/issues/251
This test assumes the subprocess does not print anything else on stdout
other than the dbus address, otherwise g_test_trap_assert_stdout()
fails to match. But if the env running tests has G_MESSAGES_DEBUG=all
then it will also print "PATH=%s".
It's a synonym of G_VOLUME_IDENTIFIER_KIND_UNIX_DEVICE.
It doesn't change anything except not feeling dirty from using a wrongly
prefixed constant for the object type.
See: #182
Help is usually printed from tools if no arguments are given and there
is not default action. However "gio mount" and "gio trash" just silently
return. Let's print "No locations given" error and show help consistently.
"gio help COMMAND" shows some arguments with "..." and some with "…",
which looks weird, e.g.:
$ gio help mount
gio mount [OPTION…] [LOCATION...]
Let's use "…" consitently.
- Compiler checks were failing because it were using C compiler to build
objc code.
- xdgmime is needed on osx too.
- -DGIO_COMPILATION must be passed to objc compiler too.
- gapplication doesn't build on osx, it is excluded in autotools too.
We have to be careful when we use add_project_link_arguments(): All
targets are built using link arguments for the C language, except for
libgio on osx which use the objc language, because it contains some ".m"
source files. See https://github.com/mesonbuild/meson/issues/3585.
https://bugzilla.gnome.org/show_bug.cgi?id=796214
Those files got renamed to .c to work around an automake issue, but
Meson needs them to have .m extension. Better rename them at build time
in Makefile.am since that's where the workaround is needed.
https://bugzilla.gnome.org/show_bug.cgi?id=672777
Fixes this build error on macOS when inside an ssh terminal:
Traceback (most recent call last):
File "[...]/gio/tests/gengiotypefuncs.py", line 23, in <module>
for line in f:
File "[...]/lib/python3.6/encodings/ascii.py", line 26, in decode
return codecs.ascii_decode(input, self.errors)[0]
UnicodeDecodeError: 'ascii' codec can't decode byte 0xe2 in position 2625: ordinal not in range(128)
https://bugzilla.gnome.org/show_bug.cgi?id=796328
GVfsUDisks2VolumeMonitor handles x-gvfs-hide/x-gvfs-show mount options
used to overwrite our heuristics whether the mount should be shown, or
hidden. Unfortunately, it works currently only for mounts with
corresponding fstab entries, because the options are read over
g_unix_mount_point_get_options. Let's introduce g_unix_mount_get_options
to allow reading of the options for all sort of mounts (e.g. created
over pam_mount, or manually mounted).
(Minor fixes to the documentation by Philip Withnall
<withnall@endlessm.com>.)
https://bugzilla.gnome.org/show_bug.cgi?id=668132
On non-glibc platforms gettext is provided by extra libintl dependency.
We wrongly thought libintl is an internal dependency and applications
needs to explicitly link on it, but turns out that breaks many
applications and with autotools the .pc generated actually has -lintl in
public "Libs:".
https://bugzilla.gnome.org/show_bug.cgi?id=796085
-z nodelete breaks the libresourceplugin module usage in the resources.c
test, which expects to be able to unload it.
Make the Meson build match what the autotools build does: only pass
glib_link_flags to the headline libraries (glib-2.0, gio-2.0,
gobject-2.0, gthread-2.0, gmodule-2.0) and omit it from all other build
targets.
https://bugzilla.gnome.org/show_bug.cgi?id=788771
gioenumtypes needs to be generated:
In file included from ../../../../external/glib/gio/kqueue/gkqueuefilemonitor.c:37:
In file included from ../../../../external/./glib/gio/glocalfilemonitor.h:25:
In file included from ../../../../external/./glib/gio/gunixmounts.h:24:
../../../../external/./glib/gio/gio.h:86:10: fatal error: 'gio/gioenumtypes.h' file not found
#include <gio/gioenumtypes.h>
^~~~~~~~~~~~~~~~~~~~
https://bugzilla.gnome.org/show_bug.cgi?id=794325
It inserted the new items one position after the given one and inserted all new items
at the same position resulting in the items being in the reverse order of the
input array.
It was decided to make these behavioural changes because this function has according to
https://codesearch.debian.net/search?q=g_list_store_splice only one real user (nautilus)
and it didn't do what one would expect from reading the documentation.
https://bugzilla.gnome.org/show_bug.cgi?id=795307
All those logging functions already add a newline to any message they
print, so there’s no need to add a trailing newline in the message
passed to them.
Signed-off-by: Philip Withnall <withnall@endlessm.com>
Reviewed-by: nobody
The monitor_path comes from g_file_get_path(), so should be freed with
g_free() rather than free(). This makes no difference because they are
the same function in practice, but using free() is a bit confusing.
Signed-off-by: Philip Withnall <withnall@endlessm.com>
Reviewed-by: nobody
This continues one of the const-correctness fixes from the previous
commit (it needed some more transitive fixes), and reverts another of
them, since it was over-zealous.
This fixes CI failure: https://gitlab.gnome.org/GNOME/glib/-/jobs/27125.
Signed-off-by: Philip Withnall <withnall@endlessm.com>
Reviewed-by: nobody
Spotted when temporarily compiling with -Wwrite-strings. This only goes
a small way towards making the code base -Wwrite-strings–clean. It
introduces no functional changes, and fixes no bugs.
Signed-off-by: Philip Withnall <withnall@endlessm.com>
Reviewed-by: nobody
This fix undefined symbol link error when building for non-glibc
platform. Applications must link on libintl, it is not a public
dependency of libglib.
On glibc platforms libintl is a not found dependency and is just ignored
by meson, so it doesn't hurt to always have it.
https://bugzilla.gnome.org/show_bug.cgi?id=795406
Fix various warnings regarding unused variables, duplicated
branches etc by adjusting the ifdeffery and some missing casts.
gnulib triggers -Wduplicated-branches in one of the copied files,
disable as that just makes updating the code harder.
The warning indicating missing features are made none fatal through
pragmas. They still show but don't abort the build.
https://bugzilla.gnome.org/show_bug.cgi?id=793729
Make it a bit more obvious that the changed properties are provided as
an a{sv}.
Signed-off-by: Philip Withnall <withnall@endlessm.com>
Reviewed-by: nobody
g_desktop_app_info_load_from_keyfile() refuses to load .desktop files
where the executable doesn't exist. Therefore whether or not the .desktop
file added in commit 148995544 is actually considered during tests depends
on /usr/bin/flatpak being installed. This isn't a safe assumption to make,
so use /bin/sh to test filtering of "prefix" commands.
https://bugzilla.gnome.org/show_bug.cgi?id=795488
The executable name can be a useful bit of information to match on in
searches where it differs from the name (for example because the latter
is localised), but will produce surprising results where the real appli-
cation is executed by a shared binary (for example interpretors like
gjs or python, or sandboxes like flatpak).
Address this by adding a blacklist of binary names that are ignored
in search.
https://bugzilla.gnome.org/show_bug.cgi?id=795488
Removing the event and closing the related file descriptor must be
done first to make sure the kqueue subsystem delete pending events.
The timeout must be disarmed before freeing the directory dependency
list otherwise it might populate it again.
https://bugzilla.gnome.org/show_bug.cgi?id=795193
This makes it a bit easier for debugging which files were generated from
which introspection XML.
Signed-off-by: Philip Withnall <withnall@endlessm.com>
https://bugzilla.gnome.org/show_bug.cgi?id=650875
Add a test for monitoring an existing local file, with the
WATCH_HARD_LINKS flag specified. This would previously cause a crash;
now it doesn’t.
This test contains a FIXME where I suspect we should be getting some
additional file change notifications from changes made through the hard
link; this requires further follow up and probably further fixes to our
inotify backend.
Signed-off-by: Philip Withnall <withnall@endlessm.com>
https://bugzilla.gnome.org/show_bug.cgi?id=755721
This gets the G_FILE_MONITOR_WATCH_HARD_LINKS flag to the state where it
doesn’t cause crashes, and essentially acts as a no-op. It will not yet
actually monitor for changes made via hard links.
Signed-off-by: Philip Withnall <withnall@endlessm.com>
https://bugzilla.gnome.org/show_bug.cgi?id=755721
The call to _start() fills in the dirname, basename, and filename
arguments according to the following rules:
dir watches: dirname filled
file watches: dirname and basename filled
hardlink: filename filled
This doesn't map to how the current inotify backend works very nicely,
so we need to adjust things a bit when creating our "sub" objects.
https://bugzilla.gnome.org/show_bug.cgi?id=755721
This test will only work on machines which have IPv6 enabled and have a
local IPv6 interface with ID 1. On machines which don’t (such as AWS
servers, which we run CI tests on), the GResolver tests will fail with
G_RESOLVER_ERROR_INVALID. We can’t differentiate this kind of failure
(where we’d want to skip the test) from an actual failure (where we’d
want to fail the test), so the only other option is to drop this
particular test vector. I don’t think it’s a significant loss.
This is the last fix needed to get our CI tests working reliably on
jenkins.gnome.org.
Signed-off-by: Philip Withnall <withnall@endlessm.com>
https://bugzilla.gnome.org/show_bug.cgi?id=795234
There are various reasons why setting up a server might fail; it
reliably fails on AWS with IPv6 addresses (are we binding to the right
address?). Since we’re trying to test GSocket as a client, skip tests
where that happens.
Signed-off-by: Philip Withnall <withnall@endlessm.com>
https://bugzilla.gnome.org/show_bug.cgi?id=795234
While 333 runs is very likely to reproduce the bug, Milan has previously
reproduced it with as few as 9 runs. Since this test will be run by the
CI machinery quite often, a lower number of runs each CI run will still
probably catch any regressions over time.
This reduces the total test runtime from 33s to 2s.
https://bugzilla.gnome.org/show_bug.cgi?id=793727
Signed-off-by: Philip Withnall <withnall@endlessm.com>
Reviewed-by: nobody
When using g_network_monitor_get_default() from another thread, it’s
possible for network-changed events to be processed after an instance of
GNetworkMonitor has been disposed, causing use-after-free problems.
Fix that by moving some of the initialisation into the GInitable.init()
chain, rather than in a main context idle callback.
This includes a unit test which probabilistically reproduces the bug
(but can’t do so deterministically due to it being a race condition).
Commit amended by Philip Withnall <withnall@endlessm.com> before
pushing.
https://bugzilla.gnome.org/show_bug.cgi?id=793727
Spotted while running valgrind on gsettings-test, as per the previous
commit.
Signed-off-by: Philip Withnall <withnall@endlessm.com>
Reviewed-by: nobody
When using g_settings_bind(), if a range binding triggers a range check
failure, g_settings_binding_property_changed() will return early, but it
won't cleanup properly causing some leaks. The binding will also still
be marked as 'running', which causes an assertion failure when trying to
free it:
"g_settings_binding_free: assertion failed: (!binding->running)"
Signed-off-by: Christophe Fergeau <cfergeau@redhat.com>
https://bugzilla.gnome.org/show_bug.cgi?id=794805
There's a race condition somewhere in GTestDBus that can result in
the next test being started at a time when g_bus_get() would still
return the connection that is in the process of closing. This can
be reproduced reasonably reliably by running the gapplication test
10K times in a loop.
Instead of relying on waiting for the weak reference to be released,
we can force the issue by clearing it.
Bug: https://bugzilla.gnome.org/show_bug.cgi?id=768996
Bug-Debian: https://bugs.debian.org/cgi-bin/bugreport.cgi?bug=894677
This adds a null notification backend implementation for win32, purely
to avoid crashes due to a missing backend when applications use
GNotification. This backend does nothing except print a warning when a
notification is supposed to be emitted.
In future, it can be expanded to use win32 API to present toaster
notifications appropriately.
Signed-off-by: Philip Withnall <withnall@endlessm.com>
https://bugzilla.gnome.org/show_bug.cgi?id=776583
If glib-networking is installed and built with libproxy support, this
test will use it. If a proxy is set in the environment, we might get
correctly told to go through it for certain accesses. However, this isn't
going to work, because the testsuite monkeys with the network monitor to
tell it that all addresses - including the proxy - aren't reachable.
We're trying to check if adding networks to a GNetworkMonitor works in
general. Proxies just get in the way here, so let's use the built in
dummy proxy resolver which just tells us that all URLs are directly
accessible.
https://bugzilla.gnome.org/show_bug.cgi?id=794801
Explain why we say "See new_stateful()" (although it's pretty obvious).
Drop a redundant copy of the argument description in the body text.
Add a # to the GVariant type name so that we can have a nice link.
https://bugzilla.gnome.org/show_bug.cgi?id=795070
If something is nullable, it's always helpful to identify what NULL
means. Also, this is not the parameter for the .activate() vfunc, as we
take that over: rather, it is the parameter for the ::activate signal.
https://bugzilla.gnome.org/show_bug.cgi?id=795070
The `mount_monitor` variable is only set if the boolean
`with_mount_monitor` variable is set to TRUE, but the compiler does not
know that, so it'll warn when calling `g_clear_object()` even if the
clearing operation is gated with the same boolean.
Initializing with NULL does not cost us anything, and eliminates a
conditional branch.
https://bugzilla.gnome.org/show_bug.cgi?id=794732
The g_auto macros are available only with GCC-compatible compilers on
Unix, but having __attribute__((cleanup)) is not part of our toolchain
requirements, so we shouldn't use it — even if we are building on
Unix-compatible systems.
https://bugzilla.gnome.org/show_bug.cgi?id=794732
Some very odd systems have the functions to initialise and destroy a
struct __res_state, but apparently not to do a DNS query using it. Fix
the compilation on those systems.
https://bugzilla.gnome.org/show_bug.cgi?id=794606
GSocketListener can keep internal references to itself for pending
accept() calls, which mean that it can stay alive (and keep listening
on ports) even after a user drops their last reference to it. They need
to call g_socket_listener_close() explicitly to avoid that.
Signed-off-by: Philip Withnall <withnall@endlessm.com>
https://bugzilla.gnome.org/show_bug.cgi?id=794207
There's no need to unconditionally print an error message if xmllint or
json-glib-format are not found when running glib-compile-resources is
called; we only need to warn if they are not available when we need
them. To avoid spamming the build logs, we can also warn once.
https://bugzilla.gnome.org/show_bug.cgi?id=794285
The glib-compile-resources tool has hand-rolled "open a temporary file"
code paths. Since error handling is hard, let's rely on GLib API that is
meant to do that consistently for us.
Get rid of some tabs mixed with spaces while we're at it.
https://bugzilla.gnome.org/show_bug.cgi?id=794284
Similarly to how glib-compile-resources can call xmllint to eliminate
whitespace in XML files to reduce their size inside a GResource, we can
use json-glib-format to achieve the same result.
The mechanism for using json-glib-format is the same, with a separate
environment variable if we want to direct glib-compile-resources to a
version of json-glib-format that is not the one in the PATH.
https://bugzilla.gnome.org/show_bug.cgi?id=794284
Normally, the list of mounts is filtered to exclude mounts in
/run/media/$username where $username is not the current user. However,
root can access all the mounts under /run/media/, regardless of the
username — so there’s no point in filtering out those mounts.
In some cases, filtering them out is harmful. In the case of a system
service which uses GVolumeMonitor, for example, filtering them out means
the service cannot see automounted USB sticks belonging to user
sessions.
Signed-off-by: Philip Withnall <withnall@endlessm.com>
https://bugzilla.gnome.org/show_bug.cgi?id=793994
Tighten up the validation of application IDs so they are always exactly
D-Bus well-known names. This is a slight change to the accepted format,
but since anyone using the API with an application ID which was
previously valid, but which was not a valid D-Bus well-known name, would
have received an error from D-Bus when their application tried to
register on the bus, I think this break is acceptable.
It will affect any applications which have application IDs which are not
valid D-Bus well-known names, and which use the G_APPLICATION_NON_UNIQUE
flag. From a quick search in Debian Codesearch, no C applications use
that flag.
Update the documentation to use the rules from the D-Bus specification,
including the latest advice discouraging use of hyphens:
https://dbus.freedesktop.org/doc/dbus-specification.html#message-protocol-names-bus
Update the tests:
• Add the examples from the documentation to validate them.
• Especially the venerable 7-zip.org example.
• Move a couple of tests from expected-failure to expected-success:
they are valid D-Bus well-known names even if they’re a bit weird.
Signed-off-by: Philip Withnall <withnall@endlessm.com>
https://bugzilla.gnome.org/show_bug.cgi?id=793400
This will make the assertion failure messages a little more useful, and
prevent the assertions being compiled out with G_DISABLE_ASSERT.
Introduces no functional changes.
Signed-off-by: Philip Withnall <withnall@endlessm.com>
https://bugzilla.gnome.org/show_bug.cgi?id=793400
- Stop using a custom thread for listening to kqueue(2) events. Instead
call kevent(2) in non blocking mode in a monitor callback. Under the
hood poll(2) is used to figure out if new events are available.
- Do not use a socketpair with a custom protocol requiring 2 supplementary
context switches per event to commicate between multiple threads. Calling
kevent(2), in non blocking mode, to add/remove events is fine from any
context.
- Add kqueue(2) events without the EV_ONESHOT flag. This removes a race
where some notifications were lost because events had to be re-added for
every new notification.
- Get rid of the global hash table and its associated lock and races. Use
the 'cookie' argument of kevent(2) to pass the associated descriptor when
registering an event.
- Fix _kh_file_appeared_cb() by properly passing a monitor instead of a
source to g_file_monitor_emit_event().
- Properly refcount sources.
- Remove a lot of abstraction making it harder to fix the remaining issues.
https://bugzilla.gnome.org/show_bug.cgi?id=739424
