[PATCH] D53066: [Driver] Use forward slashes in most linker arguments

[Martin Storsjö via Phabricator via llvm-commits]

mstorsjo reopened this revision.
mstorsjo added a comment.
This revision is now accepted and ready to land.

Nope, still no really working properly with all the tests out there (I had only run the Driver subdirectory since I'm only testing in a very underpowered VM, and it had skipped the cuda/hip tests which also seemed to have a few issues):
http://lab.llvm.org:8011/builders/llvm-clang-x86_64-expensive-checks-win/builds/13562
http://lab.llvm.org:8011/builders/clang-x64-windows-msvc/builds/886
http://lab.llvm.org:8011/builders/llvm-clang-lld-x86_64-scei-ps4-windows10pro-fast/builds/20994

I could potentially make all the tests in Driver work by continuing doing whack-a-mole for normalizing all paths in places where they are matched in tests, but this also failed some tests in CodeGen/thinlto which I haven't studied closer yet. It also seems to break tests in clang-tidy/clangd. So this doesn't really seem sustainable unless we do it completely thoroughly all over the place, but given cross-project dependencies, it looks like that is a much much larger scope change that I probably don't have time to pusue.

What other options are there? The actual original problem (libtool) I'm trying to solve only requires paths in this form in the output of `clang -v` (for linking). I guess that one could be worked around by scaling the change back to one of the original versions where I only changed a couple paths, and making that change conditional on `-v` being set (plus conditional on target as it is right now). Having that behave differently than `-###` as used in most tests, isn't really nice though, but if we change the output of `-###`, we're back at whack-a-mole.


Repository:
  rL LLVM

https://reviews.llvm.org/D53066