This is an archive of the discontinued LLVM Phabricator instance.

[lit] Output the time elapsed for each test when running lit with -v.
AbandonedPublic

Authored by jlebar on Jan 7 2016, 3:54 PM.

Details

Summary

For example:

PASS: Clang :: Driver/ppc-features.cpp (346 of 348) in 5775ms
PASS: Clang :: Driver/fsanitize.c (347 of 348) in 7346ms
PASS: Clang :: Driver/arm-cortex-cpus.c (348 of 348) in 21752ms

Diff Detail

Event Timeline

jlebar updated this revision to Diff 44285.Jan 7 2016, 3:54 PM
jlebar retitled this revision from to [lit] Output the time elapsed for each test when running lit with -v..
jlebar updated this object.
jlebar added reviewers: jroelofs, cmatthews, MatzeB.
jlebar added a subscriber: llvm-commits.
jroelofs edited edge metadata.Jan 7 2016, 3:57 PM

Why? Isn't -time-tests already way more useful?

Also, this breaks any out-of-tree scripts which read the PASS/FAIL lines... which is something that we do. If you do want to do this, please put it under a flag.

jlebar abandoned this revision.Jan 7 2016, 4:03 PM

Why? Isn't -time-tests already way more useful?

Oh, indeed it is. I somehow missed that when looking through --help.

Sorry for the noise!

Do we provide any guarantee on the output format? Scripts that rely on the output format to never change seems like a wrong design IMO. Such needs should be filled with a file output like JUnit.
(The other part of the comment about --time-tests remains valid, even though it is not a complete overlap)

In D15980#321919, @joker.eph wrote:

Do we provide any guarantee on the output format? Scripts that rely on the output format to never change seems like a wrong design IMO. Such needs should be filled with a file output like JUnit.

I don't think we do, but the current format is extremely close to DejaGnu's... if that has any bearing on it.

(The other part of the comment about --time-tests remains valid, even though it is not a complete overlap)