The actually parsed `@bytes` are not used because the only caller
does not provide an output parameter to request them. So this bug had
no effect in practice.
This allows introspection to properly handle them as GPatternSpec
methods, as per this deprecate g_pattern_match() and
g_pattern_match_string() functions.
Let the compiler figure out the size, rather than hard-coding it the
same as the struct definition.
This makes no functional changes.
Signed-off-by: Philip Withnall <pwithnall@endlessos.org>
`#define G_GNUC_NORETURN __attribute__((__noreturn__))` was previously under
`#if G_GNUC_CHECK_VERSION(2, 4)`.
But `# define G_NORETURN __attribute__ ((__noreturn__))` was under
`#if G_GNUC_CHECK_VERSION(2, 8) || (0x5110 <= __SUNPRO_C)`.
Take the latter for `g_macro__has_attribute(__noreturn__)` fallback. This will
disable `G_GNUC_NORETURN` for GCC 2.4-2.7 though.
Fall back to compiler version checks only when `__has_attribute()` is not
available.
clang-cl doesn't define `__GNU__`, but still accepts attributes. This change
gets rid of a lot of warnings when building GLib with clang-cl. For GCC and
non-cl Clang nothing should change.
glib/gstring.c: In function ‘g_string_replace’:
glib/gstring.c:998:13: warning: comparison of integer expressions of different signedness: ‘gint’ {aka ‘int’} and ‘guint’ {aka ‘unsigned int’}
998 | if (n == limit)
| ^~
Calling g_atomic_ref_count_dec() or g_ref_count_dec() and the reference
count reaches zero results in different side effects depending on
whether the reference count is atomic or not.
The intended side effect when this happens is undefined, i.e. one should
not rely on the reference count actually reaching zero, or staying 1, or
becoming something else, and one should treat the grefcount /
gatomicrefcount to be unusable until reinitialized.
This wasn't documented, so add a paragraph about this.
This change is proposed by Kamil Dudka. It teaches Coverity to assume
that g_critical() will never return, which is desirable for the same
reasons it is for scan-build: once you've triggered undefined behavior,
the game is already lost, and there's limited benefit from trying to
avoid every possible memory leak on such codepaths. Notably, this
affects g_return_if_fail().
Arguably it might be desirable to fix every such issue, but if we're
already not doing so for scan-build, it doesn't make sense to hold
developers working with Coverity to a higher standard. This allows
focusing on more serious issues found by Coverity.
Thanks Kamil!
glib/gdate.c: In function ‘convert_twodigit_year’:
glib/gdate.c:1217:13: warning: comparison of integer expressions of
different signedness: ‘gint’ {aka ‘int’} and ‘guint’ {aka ‘unsigned int’}
1217 | if (y < two)
| ^
When using TAP we want every single line to be one of the following:
- a valid TAP clause
- a comment
- a blank line
Typical explicit test logs are single line comments, but in some cases
we might end up printing debug messages from libraries, and those may
contain multiple lines. When that happens, we break the TAP and fail the
test in conditions entirely outside of our control.
One option to avoid outright failure is to always prepend each line of a
messge with `#`, to ensure that the whole thing is considered a comment.
Resolve "g_date_time_format() does not return UTF-8 if LC_TIME is not UTF8 but other locale settings are UTF-8"
Closes#2055
See merge request GNOME/glib!1777
Use the GCC/Clang macros __has_extension() and __has_feature() on
c_static_assert, which is the documented way for GCC/CLang to check for
_Static_assert if C11 mode is not enabled, as suggested by Aleksandr Mezin.
As a result, add a private macro that is defined to be __has_extension if it
exists, otherwise it is considered to be always false.
It appears that CLang supports _Static_assert() even when not in C11
mode, since at least CLang 3.1, so let's just use that for CLang builds.
Fixes issue #2338.
Functions (_g_get_time_charset and _g_get_ctype_charset) to get LC_TIME and LC_CTYPE charset
by using nl_langinfo with _NL_TIME_CODESET and CODESET).
Another functions (_g_locale_time_to_utf8 and _g_locale_ctype_to_utf8) which uses thel and format
the input string accordingly.
Add new test cases with mixing UTF8 and non UTF8 LC_TIME along with UTF8
and non UTF8 LC_MESSAGES.
Closed#2055
Signed-off-by: Frederic Martinsons <frederic.martinsons@sigfox.com>
If the same `GKeyFile` is reused to load multiple different key files,
any loads after the first which encounter translated keys will crash,
because clearing the data from the first load cleared the cached
language names, but didn’t clear `checked_locales`, so they were never
reloaded.
Signed-off-by: Philip Withnall <pwithnall@endlessos.org>
Fixes: #2361
It should not be possible for `->locales` to be set without
`->checked_locales` being set, so drop the redundant check. This helps
with branch code coverage.
Signed-off-by: Philip Withnall <pwithnall@endlessos.org>
This is basically glnx_steal_fd() from libglnx. We already had two
private implementations of it in GLib.
Signed-off-by: Simon McVittie <smcv@collabora.com>
The (out caller-allocates) and (out callee-allocates) annotations are
meant for structured or pointer types. Plain old data types are just
regular out parameters and don't need the annotation about who
allocates them.
See: https://gitlab.gnome.org/GNOME/gjs/-/issues/386
In the 2.68 cycle we’d added 3 new enumerator elements. Due to the
preceding commit, they can now be annotated with
`GLIB_AVAILABLE_ENUMERATOR_IN_2_68`, which will make it a bit easier for
third party projects to notice when they’re using these symbols without
having bumped their GLib dependency.
Signed-off-by: Philip Withnall <pwithnall@endlessos.org>
Fixes: #2327
