MaskRay added a comment.

In D110663#3029146 <https://reviews.llvm.org/D110663#3029146>, @phosek wrote:

> In D110663#3029124 <https://reviews.llvm.org/D110663#3029124>, @MaskRay wrote:
>
>> In D110663#3029088 <https://reviews.llvm.org/D110663#3029088>, @phosek wrote:
>>
>>> The reason I removed this behavior in D101194 
>>> <https://reviews.llvm.org/D101194>, aside from extra overhead introduced by 
>>> the extra checks,  is that we've seen cases where people would have both 
>>> paths on their system which lead to difficult to diagnose issues where 
>>> Clang would pick up a library different from the one they expected. I'd 
>>> therefore prefer supporting only a single spelling of the triple that could 
>>> be configured, and making sure that Clang build on Debian uses the right 
>>> configuration.
>>
>> The problem is that when LLVM_DEFAULT_TARGET_TRIPLE is "x86_64-linux-gnu" 
>> (no vendor part), CMake respects it (instead of adding `unknown`) and 
>> installs files to `lib/clang/14.0.0/x86_64-linux-gnu/`.
>> Currently clang driver only inspects the normalized 
>> `lib/clang/14.0.0/x86_64-unknown-linux-gnu/` and cannot find it.
>
> I ran into this issue the past and the solution I considered would be to 
> modify the CMake build of runtimes to query Clang for the triple (using 
> `-dumpmachine`) and use that instead of `LLVM_DEFAULT_TARGET_TRIPLE`. I still 
> think that change would be desirable for consistency and would be happy to 
> implement it.

If the host compiler is GCC (`gcc -dumpmachine` => `x86_64-linux-gnu` on Debian 
amd64), how to query Clang for the triple?

My previous comment has been edited. I believe by making Debian derivatives 
stick with `*-linux-gnu`, we can actually remove more code from the driver.


Repository:
  rG LLVM Github Monorepo

CHANGES SINCE LAST ACTION
  https://reviews.llvm.org/D110663/new/

https://reviews.llvm.org/D110663

_______________________________________________
cfe-commits mailing list
cfe-commits@lists.llvm.org
https://lists.llvm.org/cgi-bin/mailman/listinfo/cfe-commits

Reply via email to