It seems to me that check-stage1 is broken. Maybe snapshot is bad? Or is it just me?
$ make check-stage1-cfail
run cfail [x86_64-unknown-linux-gnu]: x86_64-unknown-linux-gnu/stage1/bin/compiletest
running 1833 tests
fatal runtime error: Could not unwind stack, error = 5
Illegal instruction
I think it has been like that for a long time now (bors also doesn’t run the tests on stage1)
Stage1 is built with -Z no-landing-pads these days, so anything that requires unwinding (like compiletest) isn’t going to work.
bluss
August 2, 2015, 11:04pm
4
That’s concerning, if we can’t panic, tests are nearly useless for development. Saw this on stage0 earlier today.
bluss
August 3, 2015, 12:46am
6
@eefriedman Do you mean that even when “fixed”, we can’t run the testsuite in stage0 or stage1 and see test failures?
@bluss The current version of PR27417 forces testing to always use a stage2 compiletest binary, if that’s what you’re asking.
bluss
August 3, 2015, 9:59am
8
I don’t think that fixes it?
Scenario:
I bugfix something in libcollections
I run make check-stage0-collectionstest or make check-stage1-collectionstest
Test fails, because I produced a new bug
Can I use the panic message in the test to quickly understand why it failed?
If the test failure can’t produce the assertion error message, then tests are suddenly useless for development, unless I build all the way up to stage2.
Oh, you’re right, make check-stage0-collectionstest won’t work because it directly uses the broken stage0 libtest.
make check-stage1-collectionstest should be fine, though.
bluss
August 3, 2015, 5:34pm
10
Ok stage1 working, that’s nice, but stage0 would be the best.
system
Closed
March 25, 2019, 8:24am
11
This topic was automatically closed 90 days after the last reply. New replies are no longer allowed.