| Commit message (Collapse) | Author | Age | Files | Lines |
|
|
|
| |
Comparing floats directly doesn't always work on all architectures.
|
|
|
|
|
|
|
|
|
| |
There is no way that non-GCC/CLang compilers will pass this test because the
source position will never match the position that is in
tests/scanner/Regress-1.0-expected.gir.
Fix this the fast way: define a macro according to the compiler check and
update the corresponding source position
|
|
|
|
|
|
| |
We may get drive letters and paths either in upper or lower cases
in Windows, which are actually no different. Ignore the cases
in this case.
|
|
|
|
|
|
|
|
|
|
|
|
| |
This avoids compilation erroring out on C4819 (Unicode handling issue in the
Visual Studio compiler), notably when running on Chinese, Japanese and Korean
(CJK) locales.
This also applies -utf-8 into the cflags passed into the various g-ir-scanner
command lines that are used to generate the *.gir files, where -utf-8 is
available, so that we don't get flooded with C4819 warnings during the
build, and therefore avoid potential mishaps, as C4819 is a real warning that
warngs us the code may be incorrectly built.
|
|
|
|
|
|
|
|
| |
Check that g_object_info_get_ref_function_pointer() actually returns something,
which it doesn't if the shared lib isn't found.
In case a shared lib isn't found g-i will emit a warning, so also make sure
we fail on warnings to avoid similar problems in the future.
|
| |
|
|
|
|
|
|
|
| |
Similar to !180 this should prevent devs from not running all tests by
accident.
This also adds some checks for the required doctool dependencies, mako and markdown.
|
|
|
|
|
|
|
|
|
| |
We require cairo only to run all tests and thus default it to false.
This usually results in developers not using it when working on changes and
tests depending on cairo then failing in CI.
This changes it to a feature option that defaults to auto, so that devs that
have cairo headers installed will automatically use it.
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
C99 allows defining an array argument with a fixed size as:
void foo (int arr[static 10])
Compilers conforming to the C99 specification will be able to warn if
the function is called with NULL or with an array smaller than the
specified length, something that does not happen when using pre-C99
declarations like:
void foo (int arr[10])
As the declaration above is identical to:
void foo (int arr[])
Which is, in turn, identical to:
void foo (int *arr)
Fixes: #309
|
|
|
|
|
|
| |
These tests get installed, so config.h may not be available. Do the same
thing that gimarshallingtests.c and regress.c already do, and include
config.h conditionally.
|
|
|
|
|
| |
No need to hardcode things since distutils looks it up.
Similar to !170 but for Windows.
|
|
|
|
|
|
|
|
| |
This exercises the little-used code path where a signal is emitted with
a nullable GError as a parameter. It's modeled after
GstPbutils.Discoverer's 'discovered' signal.
See GNOME/gjs#262.
|
|
|
|
|
| |
We already do the same thing for constants (see _create_const()).
Otherwise macros in .c files which aren't namespaced will trigger a warning.
|
|
|
|
|
| |
This is useful for documentation tools, and other utilities that
rely on full introspection of the C API of a given library.
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
of structures
Prior to this, the only marshalling test cases available for GPtrArrays were
for GPtrArrays of strings.
This commit adds a marshalling test case for GPtrArrays of structures, with the
same objective than the gi_marshalling_tests_array_zero_terminated_return_struct() test case.
This also adds a similar marshalling test case for GArrays of structures, for
completeness.
This is a follow-up to https://gitlab.gnome.org/GNOME/gjs/issues/9 where a
regression has been detected with these types of GPtrArrays.
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
In g_irepository_get_object_gtype_interfaces(), returning the address of
the first GIBaseInfo* does not work reliably, because the GIBaseInfos
are not necessarily stored contiguously. So the second and subsequent
ones might be garbage.
Instead, return the address of the array of GIBaseInfo pointers.
Add a test that verifies the functionality, as well.
This is unfortunately an API and ABI break.
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
Meson unfortunately does not normalize the paths for us, so we couldn't
just rely on it to give us the correct target name without the full
target path when the path separator is not '/' (such as on Visual Studio
builds, where the path separator is '\\' (with escape character).
This means that, that on Visual Studio builds, targetname would be:
D:\\gi.build\\tests\\scanner\\Typedefs-1.0.gir
instead of:
Typedefs-1.0.gir
Since we have the targetbase variable which actually has the correct
info we need, use that to deduce the correct reference .gir file to
compare to.
|
|
|
|
|
|
|
|
|
|
| |
By doing so, we essentially cover the various compiler flags that we
want to use for non-Visual Studio builds to check for warnings that
might cause real concern.
This also skips the checks for the various GCC-isque CFlag checks that
are scattered in the various build files on Visual Studio builds, since
they are essentially meaningless on Visual Studio builds.
|
|
|
|
|
|
|
|
|
| |
os.path.commonpath() raises ValueError if the paths given to it are on different
drives.
Handle that case by giving up and add a test.
Reported here: https://github.com/msys2/MINGW-packages/pull/5258#issuecomment-485230864
|
| |
|
|
|
|
|
|
|
| |
In some cases we don't initialize all the struct fields of Regress
structs. We should either initialize them all, in the case of fields
that were added later; or mark that we don't intend to initialize them
all, by using C99 designated initializers.
|
|
|
|
|
|
|
| |
These files are exported for other projects to compile, so they should
compile with as few compiler warnings as possible. If GJS or PyGObject
has -Wunused-parameter turned on, then there should not be warnings in
these files.
|
|
|
|
| |
Less noisy build
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
On macOS we ideally want to write the final absolute path of the library
into the gir and typelib. Up until now we inferred the install path from
the .la file in case we used libtool and through otool in case we weren't
using libtool hoping that the install_name was matching the install path.
meson currently sets the install_name of libraries to "@rpath/foo.dylib"
and adds rpaths to the executables in the build dir. Only during install does
it change the install_name to the absolute target path in all places.
Since we get called during build time we only get the @rpath from otool,
which then makes things fail at runtime since the executables don't have
the matching rpath set.
To make this somewhat work just fall back to the basename for relative
paths, so we dlopen "foo.dylib" and depend on things being in /usr/local/lib
or DYLD_FALLBACK_LIBRARY_PATH including the lib path (see man dlopen)
|
|
|
|
|
|
|
|
| |
Use call_count to verify the number of calls. This replaces
assert_called_once used previously, which is only available
since Python 3.6.
Fixes issue #274.
|
| |
|
|
|
|
|
|
| |
This reverts commit f606183a010fbec4382acb728882cc0eddbaf7f7.
See https://gitlab.gnome.org/GNOME/gobject-introspection/merge_requests/9#note_409979
|
|
|
|
| |
This makes the scanner consistent with the docs.
|
|
|
|
|
|
|
|
|
|
|
|
| |
They contain things like -fvisibility=hidden which, in case of building with
CFLAGS="-flto -O2" LDFLAGS="-Wl,--as-needed -flto", results in the linker
throwing out unused symbols and not linking the dumper against libregress.
This results in errors like:
Invalid GType function: 'regress_test_enum_get_type'
Failed to find symbol 'regress_test_enum_get_type'
Fix this by only passing the required CFLAGS to the scanner instead.
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
> Certain types like GIConv and GdkAtom are pointers internally but don't
> look like pointers when referenced. They have the form.
>
> typedef struct _X *X;
>
> Parse these as structures/records but mark them in the gir with a 'disguised'
> attribute so that we know that they need special handling.
Additionally, stop relaying on disguised attribute when deciding whether
to render a page. Check number of fields instead, so as to avoid
introducing large regression in the docs.
Fixes #101.
|
|
|
|
|
|
|
|
| |
Instead of allowing each error level to be enabled just enable all of them
through the enable_warnings() method. This matches what the code is currently
doing (minus that one helper script) and simplifies things.
Also remove the error count, it's not used.
|
|
|
|
|
|
|
|
|
|
| |
The scanner matches gtk-doc sections which match the lower case type name
to the type and uses that for the type documentation. The only problem is
it only takes the docs and none of the other annotations like deprecation
info etc.
This changes things to also parse the annotations in that case and adds some
tests while at it.
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
Relatively common in practice are:
* output with transfer full, which is already covered by
`regress_test_array_struct_out`,
* input with transfer none, covered by a new
`regress_test_array_struct_in_none`
Other variants are quite esoteric, but it still might be useful to
include them so that bindings can verify that they are handled
gracefully, e.g., by reporting an error instead of crashing.
Issue #90
|
|
|
|
|
| |
To make it easier to test make sure it always returns absolute paths.
Also fix a typo, oops!
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
Provide partial support for multi-dimensional arrays by representing
them as flattened one dimensional array with size that is equal to
product of sizes in each dimension. Previously only the first dimension
would be actually used.
This should be sufficient to ensure that those fields have layout
compatible with C, without using nested array types that are currently
unsupported by vapigen
Issue #255.
|
|
|
|
|
|
|
|
| |
This also reverts workaround introduced to support this use case in
f77cfc4275b1fba4f9fedea6e40b00e0ebbe142c, since it is no longer
necessary.
Fixes issue #141.
|
|
|
|
|
|
|
|
|
|
| |
Neither `_create_source_type` nor `_create_complete_source_type`
actually support fixed size arrays, so previously generated C types were
incorrect.
Remove C types from array fields instead of producing incorrect ones.
Fixes issue #145.
|
| |
|
| |
|
|
|
|
|
|
|
|
|
|
|
| |
This is necessary to parse types like `unsigned char` or `long double`,
and is already done when parsing `declarations_specifiers`. Examples
that are fixed by this change include:
* `GLib.TestLogMsg.nums` previously parsed as `long` but should be `long
double`.
* `GMime.Encoding.uubuf` previously parsed as `unsigned` but should be
`unsigned char`.
|
| |
|
| |
|
|
|
|
| |
This time pass the libgirepository build dir
|
|
|
|
|
|
|
| |
These functions were removed a long time ago in commit af7be95a. Being
present in the header file means they are present in the generated GIR
file, but trying to resolve them with dlsym will fail and crash the
program.
|
| |
|
|
|
|
|
| |
meson doesn't set PATH for the test dependencies so we have to do it manually.
Also make sure the test depends on the test binary.
|
|
|
|
|
|
|
| |
G_GSIZE_FORMAT should only be used with glib functions and G_STRUCT_OFFSET()
returns glong not gsize.
This fixes various compiler warnings on Windows.
|
|
|
|
| |
I added it to the wrong one in d2bd6390ed8, oops..
|
|
|
|
|
| |
It broke continuous
https://build.gnome.org/continuous/buildmaster/builds/2019/01/01/0/build/log-gobject-introspection.txt
|
| |
|