[clang][test] Rebuild lld if the clang driver tests will use it

[ian-twilightcoder]

Cyndy suggested that the reason I was seeing lld-repro.c fail wasn't because lld doesn't support Darwin, but rather I had a swiftlang branch checked out when I built lld (which has Darwin support explicitly disabled), and lld didn't rebuild when I switched to an llvm branch. That appears to be the case, so set lld as a dependency of the clang tests.

[llvmbot]

@llvm/pr-subscribers-clang

Author: Ian Anderson (ian-twilightcoder)

Changes

Cyndy suggested that the reason I was seeing lld-repro.c fail wasn't because lld doesn't support Darwin, but rather I had a swiftlang branch checked out when I build lld (which has Darwin support explicitly disabled), and lld didn't rebuild when I switched to an llvm branch. That appears to be the case, so set lld as a dependency of the clang tests.

---

Full diff: https://github.com/llvm/llvm-project/pull/174512.diff

1 Files Affected:

• (modified) clang/test/CMakeLists.txt (+4)

diff --git a/clang/test/CMakeLists.txt b/clang/test/CMakeLists.txt
index 3faf3d7d5f2aa..bee19fd375d3e 100644
--- a/clang/test/CMakeLists.txt
+++ b/clang/test/CMakeLists.txt
@@ -185,6 +185,10 @@ if( NOT CLANG_BUILT_STANDALONE )
     yaml2obj
     )
 
+  if(TARGET lld)
+    list(APPEND CLANG_TEST_DEPS lld)
+  endif()
+
   if(TARGET llvm-lto)
     list(APPEND CLANG_TEST_DEPS llvm-lto)
   endif()

[cyndyishida]

Sorry, I didn't mean to add more work for you 🙈 This approach looks strictly correct, so LGTM, but it would be nice if the test were layered differently to avoid the dependency on lld for check-clang.

I.e., didn't actually invoke & check the output of lld. It would instead emit whatever reproduction info was necessary (contents of linker-crash*.tar), and that was checked. Then a test in the LLD sub-project verified that output without invoking clang. I don't know enough to say that's practical, though. Please wait a day before merging in case anyone else may have comments.

[perry-ca]

Looks good. Thinking along the lines that Cyndy is, should the support for lld/lto be captured in a resource so we can clean up the REQUIRE:/UNSUPPORTED: lines in the test cases?