On 2/17/23 01:56, NightStrike via Gcc-patches wrote:
On Thu, Feb 16, 2023 at 3:16 AM Eric Botcazou <botca...@adacore.com> wrote:
This fixes dg.exp/stack-check-2.c, -7, 8, and -16.c, which is great!
Try the attached patch.
Well... that patch just marks all of the tests as unsupported. But
for example, the ones quoted above run, work, and pass. And when they
didn't pass, they highlighted the ICE that you fixed. So running the
test despite the nature of stack protection on Windows still has
value. It is useful to ensure for example that stack protection
continues to work even if options are passed to disable it. But the
tests that remain are looking for rtl patterns that (I assume)
shouldn't exist. So it's perhaps useful to modify the test to say
that on windows, the scan-rtl-dump-times check should have zero hits.
If it found a match, that would be an error.
Those tests may compile, work and pass, but they're really there to
check the probing decisions. Windows has a totally different model and
none of those tests really make sense in that model.
Is there a way to say that the test results should be inverted on a
particular platform (instead of purely unsupported)?
You can express all kinds of things, I'm just not sure it's worth the
effort for these tests.
