This allows not using the installed source files, and instead pinning the
tests to a known commit that we test against.

g-i-tests is both a git submodule and a meson subproject.

See https://gitlab.gnome.org/GNOME/gobject-introspection/-/issues/489

This was previously needed for up-to-date g-i-tests. Now that g-i-tests is
a submodule, we don't need to build this copy. The latest stable version
is fine.

Now that g-i-tests is a submodule pinned at a particular commit, we don't
need to skip tests until they appear in a g-i release.

Gary Li authored Aug 04, 2024 and Philip Chimento committed Aug 09, 2024

GJS currently fails to import modules if specifiers contain query parameters.

This is because the query question mark delimiter has been %-encoded when canonicalizing or resolving relative uris . GJS then attempts to load the resource or file with this encoded uri and fails.

Ensure the special "?" delimiter is not mangled and stop using the query component when loading resources or files. However, use the full uri in the module registry to enable cache-busting behaviour.

Fixes https://gitlab.gnome.org/GNOME/gjs/-/issues/618

module: handle import query component

Closes #618

See merge request GNOME/gjs!944

This matches what other environments do and, combined with the recently
merged support for query parameters, allows parameterizing modules.

module: Include full module specifier in import.meta.url

See merge request GNOME/gjs!947

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

doap: Remove invalid maintainer entry

See merge request GNOME/gjs!948

Closes #636

installed-tests: fix libexecdir

Closes #636

See merge request GNOME/gjs!949

Integrate gobject-introspection-tests as submodule

See merge request GNOME/gjs!946

This was originally used since g_type_query fails for dynamic types. Now that https://gitlab.gnome.org/GNOME/glib/-/issues/623 has been resolved, we can use g_type_query directly.

Inheriting from a final class currently logs criticals from g_type_register_static and subsequently crashes GJS due to instance_type == 0. Handle this more gracefully by throwing the user an error when registering the type.

Fixes https://gitlab.gnome.org/GNOME/gjs/-/issues/640

Gary Li authored Aug 24, 2024 and Philip Chimento committed Aug 29, 2024

SpiderMonkey incremental GC may interleave mutator code after JSFINALIZE_GROUP_START. However, GJS keeps track of finalize status through m_in_gc_sweep on prepare and collection end.

This causes callbacks to fail the non-sweeping check if incremental GC is happening. Instead of maintaining a parallel state, use JSAPI's RuntimeHeapIsCollecting to determine if we can safely call back into JSAPI.

Co-authored-by: Evan Welsh <contact@evanwelsh.com>

callbacks: fix sweeping check for incremental GC

See merge request GNOME/gjs!950

type: throw if registering a type with final parent

Closes #640

See merge request GNOME/gjs!951

This pointer is supposed to be freeable by js_free(). According to a
comment in js_free(), that means it is supposed to be allocated in the
malloc arena, even though that is currently not enforced.

Previously, it was allocated in the string buffer arena.

In gjs_byte_array_from_data(), previously we copied the data and then gave
that copy to the array buffer as an external pointer. It's much simpler to
allocate the ArrayBuffer with the right amount of storage and then copy
the data directly into the ArrayBuffer's storage.

Rename the function to gjs_byte_array_from_data_copy() to make it clearer
that it copies the data.

In the case of intern_gdk_atom(), we just explicitly ignore the error with
mozilla::Unused. In the case of converting GdkAtom to a JSString in
gjs_value_from_gi_argument(), we do actually want to throw if we can't
correctly call gdk_atom_name().

The variable installed_tests_execdir already contains the prefix, but the
prefix isn't defined in this substitution anymore, so we ended up with an
empty extra "/" component.

Similar to renaming BaseInfo to HasIntrospectionInfo, this frees up the
name TypeInfo to use for a C++-style introspection info wrapper.

Now that we don't need to keep track of the "is sweeping" status, we can
move that callback into GjsProfiler, and skip it altogether if profiling
is not enabled.

Similar to the finalize status callback, although there still needs to be
some work done in GjsContext; we split the profiler-only parts into a
GjsProfiler method that is only called if profiling is switched on.

Brings in a WarnLib test by Gary for enum members that start with a digit.
We previously didn't do anything with the updated test in Regress, so add
that to testRegress.js along with its neighbour which we also didn't test.

These functions need to throw JS exceptions if they return false or null.

In gjs/profiler.cpp there is a weird loop in this version of IWYU where it
wants a header to be removed if it is present and added if it is absent.
We ran into this due to editing profiler.cpp.

August maintenance

See merge request GNOME/gjs!952