[compiler-rt][tests] Fix plugin support detection when no ld.gold

[hubert.reinterpretcast]

Summary

The plugin support detection is broken when ld.bfd supports plugins but there is no ld.gold. The script tries to call GOLD_EXECUTABLE-NOTFOUND and does not handle the resulting exception. This patch handles such cases.

[daltenty]

compiler-rt/test/lit.common.cfg.py

413

Hmm, it would be nice if we avoided making the subprocess call in the first place if CMake found no gold executable. This also doesn't look like the only place to make this mistake either, some other places in the lit config (e.g. is_gold_linker_available() ) seem to be treating it as equal false if it is unset, but that's clearly not what will happen from what we see here.

Perhaps we should consider checking the validity of the value passed through from CMake in the .cfg.in and convert the -NOTFOUNDs to Nones?

[hubert.reinterpretcast]

Addressed by D133358.

compiler-rt/test/lit.common.cfg.py

 
 def is_binutils_lto_supported():
   if not os.path.exists(os.path.join(config.llvm_shlib_dir, 'LLVMgold.so')):
     return False
 
   # We require both ld.bfd and ld.gold exist and support plugins. They are in
   # the same repository 'binutils-gdb' and usually built together.
   for exe in (config.gnu_ld_executable, config.gold_executable):
+    try:
       ld_cmd = subprocess.Popen([exe, '--help'], stdout=subprocess.PIPE, env={'LANG': 'C'})
+    except FileNotFoundError as e:
+      return False
     ld_out = ld_cmd.stdout.read().decode()
     ld_cmd.wait()
     if not '-plugin' in ld_out:
       return False
 
   return True
 
 def is_windows_lto_supported():