Stream: t-compiler

Topic: why are tests ignored


Tom Tromey (Sep 25 2018 at 16:25, on Zulip):

Is there any way to find out why a test is ignored? I've built rustc against 3 versions of llvm, and the debuginfo tests are ignored for the llvm5 and llvm6 builds, but not for the build against src/llvm.

nikomatsakis (Sep 25 2018 at 16:28, on Zulip):

ignored by x.py test?

Tom Tromey (Sep 25 2018 at 16:28, on Zulip):

Yes. Also in both config.toml files I have codegen-tests = true

nikomatsakis (Sep 25 2018 at 16:34, on Zulip):

@Tom Tromey hmm. I mean you can read into the code a bit

Tom Tromey (Sep 25 2018 at 16:34, on Zulip):

eddyb thinks it is because the test infra remembers the results

nikomatsakis (Sep 25 2018 at 16:35, on Zulip):

plausible

nikomatsakis (Sep 25 2018 at 16:35, on Zulip):

you could test by removing the stamp files

Tom Tromey (Sep 25 2018 at 16:37, on Zulip):

that worked, thanks.

nikomatsakis (Sep 25 2018 at 16:38, on Zulip):

we should probably adjust the logic

nikomatsakis (Sep 25 2018 at 16:38, on Zulip):

to include info about the llvm build being used or whatever

nikomatsakis (Sep 25 2018 at 16:38, on Zulip):

the stamp files are great except when they don't consider all the variables

Tom Tromey (Sep 25 2018 at 16:39, on Zulip):

The debuginfo tests in particular are exposed to PATH changes - I have to test this patch against different versions of gdb etc.

Tom Tromey (Sep 25 2018 at 16:40, on Zulip):

Caching here just doesn't seem great to me.

nikomatsakis (Sep 25 2018 at 16:42, on Zulip):

I forget where that logic lives

Tom Tromey (Sep 25 2018 at 17:25, on Zulip):

I found it, but it's difficult to have the stamp hash incorporate the path, because the hash check is done before the test is created and so can only depend on the overall config.

nikomatsakis (Sep 25 2018 at 17:59, on Zulip):

disabling those for the debuginfo tests might be ok

nikomatsakis (Sep 25 2018 at 17:59, on Zulip):

if we can't incorporate all the dependencies

Tom Tromey (Sep 25 2018 at 18:15, on Zulip):

actually I think I'm mistaken and maybe it can be done - going to try

Last update: Nov 16 2019 at 01:20UTC