Emmanuele Bassi authored Feb 17, 2022 and Philip Chimento committed Apr 21, 2024

Members of enumeration types should not have a name that starts with
a number, as that will inevitably break languages that do not allow
identifiers to start with a numerical value.

We cannot skip members, as that would break the introspection ABI of
existing libraries; and we cannot rename members, as that would break
existing language bindings. We can emit a warning, though, and allow
libraries to fix their enumerations.

Fixes: #418

Chun-wei Fan authored Jun 28, 2023 and Philip Chimento committed Apr 21, 2024

Drop some items that actually became obsolete and let people know that for
building introspection for C++ items, it is recommended that the latest Visual
Studio 2019 or later is recommended for the build.

Andy Holmes authored Mar 12, 2023 and Philip Chimento committed Apr 21, 2024

The signals parameters `<dl>` is incorrectly placed inside the previous
`<dl>` for signal flags, resulting in invalid HTML.

Move the signal parameters out of the signal flags list, so that it is a
sibling instead of a child.

Marco Trevisan authored Oct 06, 2022 and Philip Chimento committed Apr 21, 2024

When updating the tests the expected results should be updated, but
doing it manually can be quite annoying.

So add a script to do it with some nicer automation.

Frederic Martinsons authored Jan 29, 2023 and Philip Chimento committed Apr 21, 2024

Fixes: #453

Signed-off-by: Frederic Martinsons <frederic.martinsons@gmail.com>

Andy Holmes authored Sep 19, 2022 and Philip Chimento committed Apr 21, 2024

There have been some API changes with Markdown-3.4.x that have broken
our extensions used in `docwriter.py`. Adjust the API usage so that
they keep working

fixes #439

macOS has to have a different environment variable set,
DYLD_LIBRARY_PATH instead of LD_LIBRARY_PATH.

Some of these file objects were unused to begin with. For the
gimarshallingtests ones, we don't need to use files().

Since Meson 0.60, names for custom targets are optional, and will be
named after the output file if not specified.

Generating these names depended on a broken feature, which recent Meson
versions complain about: using a non-simple object in format().

When generating introspection on aarch64, some system headers might end
up including the ARM Scalable Vector Extension types, and break the
build.

Fixes: #502

The whole docblock needs to be reworded, now that gobject-introspection
is completely decoupled from gtk-doc.

No functional change intended. This will make future changes and review
easier.

As suggested on !456.

Signed-off-by: Simon McVittie <smcv@debian.org>

Signed-off-by: Simon McVittie <smcv@debian.org>

Like we do for boxed types.

We are currently skipping types that derive from G_TYPE_POINTER, mainly
for historical reasonds. GTK4 started using G_TYPE_POINTER types for
opaque structures passed to callbacks and signals by reference to avoid
the cost of a copy/free. This currently has the effect of adding an
untyped data structure to the GIR data, and including the get_type
function in its symbols.

If a G_TYPE_POINTER type has a record or a union associated to it, then
we should pair the GType name and get_type function to the corresponding
structured type in the GIR, instead.

Fixes: #508

A property with only (setter SETTER) can crash here with:

```
Traceback (most recent call last):
  File "/usr/bin/g-ir-scanner", line 104, in <module>
    sys.exit(scanner_main(sys.argv))
             ^^^^^^^^^^^^^^^^^^^^^^
  File "/usr/lib64/gobject-introspection/giscanner/scannermain.py", line 613, in scanner_main
    main.transform()
  File "/usr/lib64/gobject-introspection/giscanner/maintransformer.py", line 134, in transform
    self._pair_property_accessors(node)
  File "/usr/lib64/gobject-introspection/giscanner/maintransformer.py", line 1565, in _pair_property_accessors
    getter = ['get_' + normalized_name]
```

Otherwise, after a `continue` one might still carry the old value.

Fixes the error:

```
../gtk/gtkimage.h:114: Warning: Gtk: Getter method 'gtk_image_get_storage_type' for property 'resource' has a mismatched '(get-property storage-type)' annotation
```

And open the 1.82 development cycle.

Signed-off-by: Philip Withnall <pwithnall@gnome.org>

Signed-off-by: Simon McVittie <smcv@debian.org>

Previously we assumed that these are int-sized, but that isn't necessarily
true on every Unix variant (and in fact dev_t is 64-bit on x86 Linux,
so was always wrong here). Give them the same treatment as time_t.

These types are not included in INTROSPECTABLE_BASIC and therefore not
included in `everything.[ch]`, because the code that generates those
files is not set up to be able to deal with types that only exist on
certain platforms. They *are* included in `gimarshallingtests.[ch]`,
but only within `#ifdef G_OS_UNIX`.

Resolves: https://gitlab.gnome.org/GNOME/gobject-introspection/-/issues/506


Signed-off-by: Simon McVittie <smcv@debian.org>

The closure annotation only takes an argument when used for parameters
of a function. When used on the parameters of a callback type, it's a
unary annotation and takes no arguments.

Make sure we warn whenever the closure argument on a callback type is
wrongly annotated.

Fixes: #449, #450

Annotations on callback can be very confusing due to to many
annotations, particularly where they should be placed. This generally
comes down to using (closure) on the incorrect parameter but it can also
be nice to work on incorrect (destroy) and (scope) usage.

We also need to gracefully handle aliased callback types, and not just
the generic GCallback.

Original patch by: Garrett Regier <garrett.regier@riftio.com>

Original bug: https://bugzilla.gnome.org/show_bug.cgi?id=752058

Fixes: #137

`-i` was previously documented to be an alias for `--library`, but is
actually an alias for `--include`.

`-l` was not previously documented, and is an alias for `--library`.

`g-ir-scanner --help` already shows them correctly.

Signed-off-by: Simon McVittie <smcv@debian.org>

GStreamer is referencing a bunch of new types, resulting in incomplete
and invalid `GstVulkan-1.0.gir` when it tries to reference inexistant
types from the `Vulkan` namespace.

Libraries can't always change enumeration members without breaking API,
so using a warning is too drastic. Ideally, maintainers should enable
strict mode during development, and disable them during regular builds.

Naming your C APIs according to the conventions makes it a lot easier to
produce a correct GIR definition of them. So link prominently to the
GObject documentation which covers those conventions.

Signed-off-by: Philip Withnall <pwithnall@gnome.org>

Jonathan Blandford authored Jul 09, 2024 and Philip Chimento committed Jul 28, 2024

Fixes #495

The gnome:userid for this maintainer does not exist. Remove the entry for clarification and to avoid errors in automated processing of .doap-files.

Chun-wei Fan authored Jun 25, 2024 and Philip Chimento committed Aug 17, 2024

Treat that directive like an inline when running the g-ir-* scripts, so
that the scanner won't complain when encountering this Visual
Studio-specific directive.

This takes the approach of a git submodule without a meson subproject,
simply adjusting the paths to the source files so that they point into the
git submodule. Everything else should be the same. We just ignore the
meson.build file in the submodule.

I considered using a meson subproject which would be the recommended way
to integrate g-i-tests for language bindings, but I believe that approach
is fundamentally incompatible with g-i where we need to use the just-built
g-ir-scanner. g-i-tests uses gnome.generate_gir() in its build file, which
uses meson.override_find_program() for g-ir-scanner. That is incompatible
with using it in the parent project to override g-ir-scanner with the
just-built copy.

Closes: #489

a492069f renamed the
_GI_DISABLE_CAIRO to GI_TEST_DISABLE_CAIRO but forgot
to also change the regress_scanner args we pass on.

Followup to a492069f

When we made the switch to g-i-tests, I took the opportunity to rename
that symbol in the new codebase so that it doesn't start with an
underscore, which is reserved.

Other tests in g-i still use _GI_TEST_EXTERN which is defined in a
separate copy of gitestmacros.h. It's not good to have two copies
floating around, so remove the old copy and use the new symbol. Adjust
include paths to point into g-i-tests where necessary.

Part-of: <https://gitlab.gnome.org/GNOME/gobject-introspection/-/merge_requests/488>

the resulting URL does not exist on g-i forks thus breaking CI
for every MR from a fork.

This reverts commit 98a02780.

It was removed with setuptools 74.0.0. Since we still depend on the
MSVCCompiler class use new_compiler() to get it some other way.

Remove any reference to MSVC9Compiler, which was for Visual Studio 2008
which we no longer support anyway.

Fixes #515

Some Linux distributions routinely build all executables and libraries
with -Wl,--as-needed, either by adding it to LDFLAGS or by making it the
compiler or linker default.

For installed executables and libraries this is a desirable way to
drop unnecessary transitive dependencies, minimizing the number of
rebuilds required if a lower-level library breaks ABI, but for our
dump executable it's counterproductive and can prevent us from finding
the SONAMEs of libraries that were passed as explicit -l arguments,
especially if we use a ldd replacement that is not recursive
(like the one in gobject-introspection#482).

We already had a special case to add -Wl,--no-as-needed to the linker
arguments, but only for "internal" libraries and only when not using
libtool. Extend that to cover "external" libraries and libtool as well,
and filter out any -Wl,--as-needed from LDFLAGS to avoid it from being
defeated.

Resolves: GNOME/gobject-introspection#505
Signed-off-by: Simon McVittie <smcv@debian.org>

We are generating the GIR file for Cairo, which means we can't use it as
a string when defining the validation test() objects.

To avoid the deprecation warning from Meson, we can switch to an array
of dictionary entries containing a name for the GIR file, and the file
itself, and use the former when generating targets.

Getting rid of the deprecation warning also makes the `meson test`
output a lot nicer, since Meson doesn't have to make up a complicated
test name.

Apple expects dylibs to have an ID of the form `@rpath/libyadda.dylib`.
For this convention to work, a RPATH entry pointing to the library
prefix and/or to the relative location of said prefix w.r.t the
dependent executable or library, must be inserted.

This need previously did not arose because gobject-introspection
implicitly relied on the consumers having IDs set to their absolute
paths. This is no longer the case with GStreamer's Cerbero, which now
sanitizes RPATHs, load commands, and dylib IDs at install time.

The simplest and most obvious fix is to ensure that, for each library
load path, the corresponding rpath entry is inserted.

The custom_target() rule is missing the installation step.

Fixes: #516