[PATCH] D123897: [X86] Unbreak LIT/FileCheck

[Roman Lebedev via Phabricator via llvm-commits]

lebedev.ri added a comment.

In D123897#3455421 <https://reviews.llvm.org/D123897#3455421>, @MaskRay wrote:

> You need evidence to back up your statement.

Let's unflip the table. Why do you believe the current default is right, for X86 tests?, for FileCheck in general?
If you like having broken default, that's up to you, but i'm confident that that not many agree with you on this.
I'm only fixing the X86 story that affects me.

> How is writing new `llvm/test/CodeGen/X86` tests impossible now?

It is **especially** egregiously fundamentally by-design incompatible with the way
multi-prefix output deduplication works for autogenerated checklines.

It is *normal* to have unused check prefixes afterwards, yet now i'm forced to remove them,
and guess what, one day after some codegen change the prefix would become used,
yet it is not there anymore, and the check lines will go missing.

The last drop in the bucket that prompted me to submit this was that i refreshed X86 test coverage for interleaved load/store patters,
which took me ~30 minutes, and them i spend 20 minutes trying to reconcile with FileCheck on 1/8'th of the new tests.
It's a monumental waste of time, that only hurts, both in short run and long run.

> Are there concrete examples when `--allow-unused-prefixes=false` makes the test less appealing in a way which can't be mitigated?

Again, anything to that effect is fundamentally wrong question, akin telling people to stop using the `update_llc_test_checks.py` family of scripts.

> I think some folks may be interested to improve `update_llc_test_checks.py` to mitigate the effect while retaining the error checking effect. There may also be users who want the new effect even with `update_llc_test_checks.py` generated tests.




Repository:
  rG LLVM Github Monorepo

CHANGES SINCE LAST ACTION
  https://reviews.llvm.org/D123897/new/

https://reviews.llvm.org/D123897