[PATCH] [ASan tests] Use clang-cl to build tests on Windows

Timur Iskhodzhanov timurrrr at google.com
Mon May 12 01:45:01 PDT 2014


> 1. Should we add -o to clang-cl?
> MSVC ignores it silently, and CMake always uses it when it does its initial test compile.

I don't see any reasons not to do that.

> 2. Should we even use clang-cl when building compiler-rt?  Do we have lots of CFLAGS in compiler-rt?
> We will either have to expose them as core clang-cl options, or switch to the regular clang driver.
> Do you need this to handle /MT or /MD?

I think we can start using clang-cl as a //secondary// compiler building c-rt so we have extra test coverage, but stick with CL as a //primary// compiler until we declare clang-cl production-quality.

================
Comment at: CMakeLists.txt:65
@@ +64,3 @@
+    set(COMPILER_RT_TEST_COMPILER_EXE "-o")
+    set(COMPILER_RT_TEST_COMPILER_OUT "-o")
+  else()
----------------
Alexey Samsonov wrote:
> _OUT makes no sense to me. You can probably use EXE/OBJ instead.
Ah, good idea -- I forgot about the "OBJ" possibility :)

================
Comment at: cmake/Modules/AddCompilerRT.cmake:140
@@ -140,1 +139,3 @@
+    COMMAND ${COMPILER_RT_TEST_COMPILER} ${TEST_OBJECTS}
+            ${COMPILER_RT_TEST_COMPILER_EXE}"${output_bin}"
             ${TEST_LINK_FLAGS}
----------------
Reid Kleckner wrote:
> Alexey Samsonov wrote:
> > Sanity check: is it ok to not have a space between ${COMPILER_RT_TEST_COMPILER_EXE} and "${output_bin}" ?
> There cannot be a space between -Fo and its argument.  =/
What Reid says.

http://reviews.llvm.org/D3680






More information about the llvm-commits mailing list