Add variadic PAS_ASSERT support.

[MenloDorian]

b9a2b29

Add variadic PAS_ASSERT support.
https://bugs.webkit.org/show_bug.cgi?id=241626
rdar://95204263

Reviewed by Keith Miller.

This patch adds support for a variadic PAS_ASSERT, and uses it to gather
more debugging info for asserts in pas_local_view_cache_stop and
pas_segregated_heap_ensure_allocator_index.

From manual local disassembly of pas_local_view_cache_stop, I verified that
this patch only changes assertion fail reporting code in an out of line slow
path.  The change does perturb code size, and any changes in code may also
cause clang to arbitrarily rearrange the order of some control flow diamonds
(e.g. the cases in a switch statement).

If no extra data is provided to the variadic PAS_ASSERT, it will emit exactly
same code as it does previously when PAS_ASSERT was not variadic.  This gives
us confidence that any perturbations in code will only manifest in PAS_ASSERTs
that we choose to add extra info for reporting assertion failures.

Preliminary benchmark results appear to show that performance is neutral.

* Source/bmalloc/libpas/src/libpas/pas_local_view_cache.c:
(pas_local_view_cache_stop):
* Source/bmalloc/libpas/src/libpas/pas_segregated_heap.c:
(pas_segregated_heap_ensure_allocator_index):
* Source/bmalloc/libpas/src/libpas/pas_utils.c:
(pas_crash_with_info_impl1):
(pas_crash_with_info_impl2):
(pas_crash_with_info_impl3):
(pas_crash_with_info_impl4):
(pas_crash_with_info_impl5):
(pas_crash_with_info_impl6):
(pas_report_assertion_failed):
* Source/bmalloc/libpas/src/libpas/pas_utils.h:
(pas_assertion_failed):
(pas_assertion_failed_noreturn_silencer1):
(pas_assertion_failed_noreturn_silencer2):
(pas_assertion_failed_noreturn_silencer3):
(pas_assertion_failed_noreturn_silencer4):
(pas_assertion_failed_noreturn_silencer5):
(pas_assertion_failed_noreturn_silencer6):

Canonical link: https://commits.webkit.org/251567@main
git-svn-id: https://svn.webkit.org/repository/webkit/trunk@295562 268f45cc-cd09-0410-ab3c-d52691b4dbfc

[kmiller68]

r=me with a style comment

[kmiller68]

I think webKit style would put a space after each ,

[MenloDorian]

Thanks. Fixed.

[kmiller68]

For what it's worth when we did this in JSC for our RELEASE_ASSERTS we didn't see a perf impact. Probably due to the crashing side being considered UNLIKELY so the compiler puts that info off in oblivion. The only thing is that some crashes may not be merged but we never saw a perf impact if that happened.

[MenloDorian]

Yeah. It's just that we've been told that libpas appears to be extra sensitive to this for reasons unclear to us. So, we should proceed with caution and let perf measurements do the talking.

[webkit-early-warning-system]

Committed r295562 (251567@main): https://commits.webkit.org/251567@main

Reviewed commits have been landed. Closing PR #1537 and removing active labels.