How is this supposed to work? I've just rebuilt and reinstalled LLVM with LLVM_INSTALL_UTILS=ON, I certainly see FileCheck when grepping /usr/local/lib/cmake/llvm/LLVMConfig.cmake yet I can't build compiler-rt anymore:

In D56976#1365190, @ilya-biryukov wrote:

Was confused about your comment, so it works on other systems because they don't rely on argv[0] to get the path to main executable. Got you.

Added a comment as requested.

In D56976#1365001, @ilya-biryukov wrote:

LGTM to unbreak the tests on *BSD, but could you please leave a comment?
It would probably be best to take argv[0] into account on *BSD systems as well (possibly only if clang could not be found).

At least if that's the behaviour on all other systems.

Added FreeBSD warning comment.

Looks good. Optionally, I'm suggesting one cleanup.

@ruiu, what do you think? The current logic forces some precedence, i.e. if you pass -z nognustack, further -z {no,}execstack are ignored. I suppose we could just force passing -z nognustack as last option clang-wise.

In D47819#1356347, @Lekensteyn wrote:

The direction of this patch looks reasonable to me. Is it worth mentioning the issue (https://github.com/google/sanitizers/issues/974) in the commit message?

Fixed leaving triple unset on invalid --target.

Ping. I'd really like to merge this before the branch.

Split target logic into D56650, switched to using target to determine which paths to apply. While at it, copied the code from clang since it now can match exactly.

Updated the tests to account for most of the known sanitizer types.

Thanks, builds fine now.

Implemented checking the triple against target registry. Also made --version output the detected target.

Actually, I've just researched a bit and default stack permissions depend on arch in glibc, so assuming it's RWX by default is wrong.

@ruiu, what if one of the systems changes defaults (e.g. due to Hardening) and starts defaulting to noexecstack? In that case we'd want `-z execstack' to actually emit PT_GNU_STACK, and I don't think we really are able to 100% detect the default in clang. I'd really see this as a trinary option: emit RW, emit RWX or not emit at all. I don't think it's a good idea to make linker rely on implicit assumptions or hidden guesses that could easily confuse user as to what's happening and why.

Following the changes to D47817, updated this one to disable relevant tests when sunrpc headers are not available.

Please don't risk merging this before the branching.

Thank you.

Need to update this for the new idea of not using libtirpc at all.

I think it'd be better to have a positive version, e.g. something like COMPILER_RT_USE_LIBCXX_SOURCES. Maybe it could default to ${LLVM_TOOL_LIBCXX_BUILD}.

Adjusted to make paths sysroot-relative.

This actually breaks using external lit. It no longer respects -DLLVM_EXTERNAL_LIT passed to CMake, and instead forces the hardcoded libcxx path, which — since the file listed there doesn't exist — causes fallback to also-non-existing llvm-lit to happen.

This commit is causing regression when building with -DLIBUNWIND_ENABLE_CROSS_UNWINDING=ON:

Could you also add a local switch to control using it independently of main build? The sanitized builds are quite time-consuming.

Updating for completeness.

Next version, based on recognizing NetBSD from triple.

Ping.

Ping.

Ping again.

Ok, maybe I'm being silly but since clang driver has to pass --enable-new-dtags for GNU ld compatibility anyway, wouldn't it make sense to keep the default as disabled in order to match GNU ld behavior?

For the record, another option is to actually fix other software not to call LD directly.

We've discussed this a bit and given other changes we need to do, and I see pretty much three options here:

1. We hardcode stuff under defined(__NetBSD__) which kinda solves the problem, except lld won't be very cross-friendly.
2. We try to do conditionals based on triple but this works only when we customize the install to include it in executable name. We probably would still need to default based on defined(__NetBSD__) when triple isn't available via process name.
3. We create *nbsd* emulations for all arches (e.g. amd64/x86 don't have such emulations right now), and use that to switch logic. This would be closer to what FreeBSD does, I think. However, if we do this, then I suppose we should also add similar aliases to GNU ld.

Removed non-x86.

Updated to use clang's libdir logic.

@ygribov, ping.

Thank you for addressing this!