Bug#1069211: clutter-1.0: build-time tests fail on loong64

[Simon McVittie]

Control: retitle -1 clutter-1.0: build-time tests fail on loong64
Control: tags -1 + moreinfo

On Thu, 18 Apr 2024 at 02:26:08 +0000, wuruilong wrote:
> The clutter software fails to compile on loongarch architecture, 
> please refer to the attached patch to fix it.

This is not a fix, this is a workaround: it's adding loong64 to the list of
architectures where the tests are expected to fail. Why are the tests failing
on loong64?

If our first reaction to tests failing is to disable the tests completely
instead of investigating what is failing and why, then the tests have
no value.

One possible reason is that Mesa's llvmpipe might be broken on loong64,
similar to #1002690 on mips64el, src:mutter bug #1059000 on loong64,
src:gnome-shell bug #1058687 on riscv64, and various src:gtk4 bugs on
all of those architectures.

loong64, mips64el and riscv64 seem to all have similar issues when using
llvmpipe, so a fix for one might well help the others.

Note that unlike mutter and gtk4, clutter-1.0 is dead upstream (#996690),
so if loong64 users do not have a specific need for packages that depend
on clutter-1.0, it might be better for clutter-1.0 and the packages that
depend on it to be permanently missing from new architectures.

The only important package that would be missing if clutter-1.0 is not
present seems to be evolution, although I see that in #1018120 I thought
that disabling one optional feature ("contact maps") would avoid that
dependency.

    smcv