Make libFuzzer test more resilient to changes

[george.karpenkov]

Summary

LIT launches executables with absolute, and not relative, path.
strncmp would try to do exact comparison and fail.
Changing to strnstr makes test less fragile.

[kcc]

LGTM

[george.karpenkov]

@kcc my bad, strnstr is actually not available outside of FreeBSD-like systems.
Would it be OK to change to strstr?

[kcc]

In D36242#829800, @george.karpenkov wrote:

@kcc my bad, strnstr is actually not available outside of FreeBSD-like systems.
Would it be OK to change to strstr?

Data is not null-terminated, so you will need to copy it to a null-terminated buffer first.

Why not memmem?

[george.karpenkov]

@kcc that seems to work in both environments, but makes me confused as to why it would act differently from strncmp, since we only enter the branch if length are equal. Yet when running under lit, one version times out, and another finishes in 0.8 seconds.

[kcc]

LGTM

[george.karpenkov]

@kcc Great, memmem is not available on Windows :P

[kcc]

But that won't matter for us soon, right?

[george.karpenkov]

@kcc so should I proceed with disabling tests on Windows? Feels terrible to say that "tests pass" when they don't.

[kcc]

Yes, please make check-fuzzer a no-op on windows

llvm/trunk/lib/Fuzzer/test/InitializeTest.cpp

extern "C" int LLVMFuzzerInitialize(int *argc, char ***argv) {
   assert(*argc > 0);
   argv0 = **argv;
   fprintf(stderr, "LLVMFuzzerInitialize: %s\n", argv0);
   return 0;
 }
 
 extern "C" int LLVMFuzzerTestOneInput(const uint8_t *Data, size_t Size) {
   if (Size == strlen(argv0) &&
-      !strncmp(reinterpret_cast<const char *>(Data), argv0, Size)) {
+      !strnstr(reinterpret_cast<const char *>(Data), argv0, Size)) {
     fprintf(stderr, "BINGO %s\n", argv0);
     exit(1);
   }
   return 0;
 }