Prevent gcc from issuing a warning upon coffnamecpy

[serge-sans-paille]

Summary

This is a follow up to d1262a6e9, more explicit to cope with GCC smartness.

The intent is also more explicit now. Otherwise GCC 9.2 inlines coffnamecpy and trigger a (rightful) warning.

[thakis]

llvm/lib/Object/WindowsResource.cpp

723–724

s/larger/smaller/?

725

Why do we need this if at all? The buffer is initialized to all-zeroes, right? (if not, we'd have to fill all remaining padding with zeroes, not just one byte.)

[serge-sans-paille]

I've manually checked the fact that out memory has been zeroed, and added an assert to make sure it remains so.

[lbenes]

I can confirm that this patch does silence the warning on gcc 9.2.

llvm/lib/Object/WindowsResource.cpp

WindowsResourceCOFFWriter::write(uint32_t TimeDateStamp) {
   writeStringTable();
 
   return std::move(OutputBuffer);
 }
 
 // According to COFF specification, if the Src has a size equal to Dest,
 // it's okay to *not* copy the trailing zero.
 static void coffnamecpy(char (&Dest)[COFF::NameSize], StringRef Src) {
   assert(Src.size() <= COFF::NameSize &&
-         "Src is not larger than COFF::NameSize");
-  strncpy(Dest, Src.data(), (size_t)COFF::NameSize);
+         "Src is larger than COFF::NameSize");
+  assert((Src.size() == COFF::NameSize || Dest[Src.size()] == '\0') &&
+         "Dest not zeroed upon initialization");
+  memcpy(Dest, Src.data(), Src.size());
 }
 
 void WindowsResourceCOFFWriter::writeCOFFHeader(uint32_t TimeDateStamp) {
   // Write the COFF header.
   auto *Header = reinterpret_cast<coff_file_header *>(BufferStart);
   Header->Machine = MachineType;
   Header->NumberOfSections = 2;
   Header->TimeDateStamp = TimeDateStamp;