rorth wrote: > > It used to be possible to set CLANG_DEFAULT_LINKER to e.g. gld. This > > doesn't work any longer. > > That said, the patch introduces quite a number of regressions > > There are two, by my count, and they are both "regressions" in the sense that > the tests erroneously pass when `gld` is supplied as `CLANG_DEFAULT_LINKER`, > not as an environment variable, but rather a CMake variable. > > Indeed, if you `cmake ... _DCLANG_DEFAULT_LINKER=gld` and run the supposedly > failing test cases, you'll see this:
The please also try a regression test with `-DCLANG_DEFAULT_LINKER=/usr/gnu/bin/ld`, which unless I'm mistaken worked before, but no longer does. Looking at the test output, there's ``` > # | 6: "gld" "-e" "_start" "-m" "elf_x86_64_sol2" [...] ``` This is wrong: the linker is run using `execve` or `posix_spawn`, both of which expect a **pathname**, not a filename, and don't do a `PATH` search. > That is, a "failing" test case in which clang is clearly doing the correct > thing, i.e. respecting your build-time selection, and unlike before, is > driving it using the correct set of flags... I'm afraid that's not true. https://github.com/llvm/llvm-project/pull/163000 _______________________________________________ cfe-commits mailing list [email protected] https://lists.llvm.org/cgi-bin/mailman/listinfo/cfe-commits
