[PATCH] D134337: [clang] [Driver] More flexible rules for loading default configs

Wed Sep 28 04:03:30 PDT 2022

mgorny added inline comments.

================
Comment at: clang/test/Driver/config-file3.c:27

-//--- Invocation qqq-clang-g++ tries to find config file qqq-clang-g++.cfg first.
+//--- Invocation x86_64-unknown-linux-gnu-clang-g++ tries x86_64-unknown-linux-gnu-clang++.cfg first.
 //
----------------
sepavloff wrote:
> mgorny wrote:
> > arichardson wrote:
> > > sepavloff wrote:
> > > > Tests must check the case when target prefix is not a real triple as in the original test (qqq-clang).
> > > It would be quite important for me that this continues to work. I made use of that in the CheriBSD toolchain when creating [[ https://github.com/CTSRD-CHERI/cheribuild/blob/master/pycheribuild/projects/cross/llvm.py#L499 | symlinked binaries to easily build for different ABIs]] such as `cheribsd-riscv64-hybrid-clang++` and `cheribsd-riscv64-purecap-clang-cpp`. It appears this previously only worked if the prefix did not start with a valid triple (which is why I put the OS before the architecture). I think it would also be nice if the whole prefix was checked even if it starts with a valid triple, but this does not need to be changed in this patch (haven't looked at it in detail so this might actually work).
> > If the prefix is not a valid triple, then clang ignores it and uses host triple instead. And now we're back to square one. If I check both variants, it's too complex. If I don't, it's bad too.
> Our customers use this feature. Target prefix may designate, for example, debug build or build with specific framework.
How are we supposed to avoid the "absurd" case where x86_64 configs are loaded for `-m32` invocation then?

CHANGES SINCE LAST ACTION
  https://reviews.llvm.org/D134337/new/

https://reviews.llvm.org/D134337