This is an archive of the discontinued LLVM Phabricator instance.

[NFCI] Rerun update_test_checks.py on all Analysis tests
AbandonedPublic

Authored by jdoerfert on Oct 31 2019, 3:04 PM.

Details

Summary
NOTE: This is, for now, mostly to allow some discussion.

Rerun update_test_checks on all tests that have the 'Assertions have
been autogenerated by' line in llvm/test/Analysis except the ones below
that do have larger diffs for reasons not caused by D68819.

Exceptions:

llvm/test/Analysis/MustExecute/
llvm/test/Analysis/MemoryDependenceAnalysis/invariant.group-bug.ll

Event Timeline

jdoerfert created this revision.Oct 31 2019, 3:04 PM
Herald added a project: Restricted Project. · View Herald TranscriptOct 31 2019, 3:04 PM
Herald added subscribers: bollu, Prazek. · View Herald Transcript
nikic added inline comments.Oct 31 2019, 3:33 PM
llvm/test/Analysis/ConstantFolding/math-1.ll
16

I guess a problem with doing these kinds of mass-regenerations is that it's easy to miss meaningful diffs that hide in between. Like this one -- I'm not sure why (presumably some host dependent float precision issue), but the use of {{.+}} here probably needs to be preserved.

jdoerfert marked an inline comment as done.Oct 31 2019, 3:56 PM
jdoerfert added inline comments.
llvm/test/Analysis/ConstantFolding/math-1.ll
16

I think, at least for this diff, it is not hard to spot/identify any change that does not include define {{[^@]+}}. While that might not be always true, we can probably write decent filters for changes of this kind.

Wrt. this actual change, I see your point. Though, I think that it is more of a problem with our usage of this script (or, arguably, my intended usage).

On the one hand, the files "says" that check lines are auto generated but in reality they are not. If we cannot regenerate them, I'd say the NOTE is relatively useless. Maybe we don't need to rerun on this file ever again, then there is no need to mark the file with the NOTE. If the idea (of the note) is to tell people to run the update script when they change something, then it is unclear why we have manual changes like these that do not survive a rerun.
Maybe we shouldn't have such manual changes, so you see the actual diff. Maybe we should have manual changes that survive the update script.
Or maybe I misunderstand what the script and the NOTE is supposed to do.

I came up with a partial solution for some of the problems here: D69701

Around manually modified parts you can now place markers that will prevent future changes, e.g.:

; UTC_ARGS: --turn off

// lines are copied

; UTC_ARGS: --turn on
jdoerfert abandoned this revision.Nov 4 2019, 4:47 PM

This was just for discussion anyway.