GitLab

SpiderMonkey 78 has removed this feature and recommends using linters.

(For some bizarre reason, some of the output in the LCov coverage data
is influenced by this flag, so we need to change some of the expected
output.)

See: #329

This will be unsupported in SpiderMonkey 78. It seems that in these
particular cases, we can pre-define these properties on classes with
resolve hooks, without using the RESOLVING flag because the resolve hook
doesn't try to access any of these.

See: #329

Rework global code to support multiple global "types". (Part 1)

See merge request !453

CI: Switch ASAN jobs to runners tagged so

See merge request !461

IWYU already does this, but we need to implement it as well for our
logic that decides what files to check.

(Philip: For context, I am committing this in advance of !432
as I go through the commits there one by one. A substantial part of this
file is being rewritten by Marco)

We're quite sure we do, otherwise we'd have already returned false but
g++ isn't smart enough to catch it, when building with release optimizations.

Otherwise, they will build sysprof-capture-4 as a subproject every time
they run, and collect coverage from the subprojects directory, making
the coverage percentage artificially low.

Unreviewed, pushing to fix build.

See merge request !457

In case sysprof-capture-4 is not installed, this should allow building
it from source.

To keep in lockstep with development of sysprof.

This is an ABI break, but (with the previous commit in place) not an API
break for the API which GJS uses.

See: sysprof!30 (comment 857673)Signed-off-by: Philip Withnall <withnall@endlessm.com>

`sysprof_capture_writer_set_flush_delay()` is going away in v4 of
libsysprof-capture, so open-code it in GJS instead.

See: sysprof!30 (comment 857673)Signed-off-by: Philip Withnall <withnall@endlessm.com>

We need sysprof-capture-4 which has not been released yet. So we need to
build it ourselves. Build it only in the debug image, at least until it
is packaged on Fedora.

Oops, forgot the "except: schedules" that's on all the other jobs;
otherwise these will run during the monthly scheduled image build.

Unreviewed; pushing to fix scheduled jobs.

June maintenance

See merge request !459

Apparently llvm-config has been moved into a package that is no longer
included in mozjs68's build dependencies.

Ensure that native boolean type is saved as we expect to be in gboolean
world.

Instead of manually picking the GIArgument union value, make possible to
select the right one using a template type, so that we can avoid
repeating the same code multiple times in the various template
functions.

As per this, utilize to avoid repetition in arg template functions

See merge request !444

g_assertion_message() can return if nonfatal assertions are enabled, but
for the purposes of static analysis, we want to consider it non-
returning.

See: glib!1553

For some reason this one wasn't found by IWYU, but it is required when
analyzing the file by itself.

I mistakenly put it into the postprocessor thinking it was forward-
declared in js/TypeDecls.h, but it isn't.

See merge request !454

Getting the function name for logging a warning/error message using
format_function_name() is expensive and shows up with quite a few
percent in the profiler.

Luckily, we can completely get rid of that overhead by only getting the
function name in case we actually have to warn. To do that, check the
number of arguments ourselves instead of using args.requireAtLeast() and
then move the call to format_function_name() into the if-conditions of
those checks.

Update ESLint CI image.

See merge request !451

- In ESLint v6, dangling commas on functions are now linted.

See merge request !452

Add a simple implementation of GListModel, primarily as a demonstration
of implementing interfaces in genera, but particularly this interface as
it will be used quite a bit in Gtk4.