Fix a bug introduced in r305092 on big-endian systems.

[ecbeckmann]

Summary

We were writing the length of the string based on system-endianness, and
not universally little-endian. This fixes that.

[ruiu]

Yes, write{16,32,64}le is usually much more convenient than casting to ulittle{16,32,64}_t.

[ecbeckmann]

Use writele16 instead of casting.

[ruiu]

llvm/lib/Object/WindowsResource.cpp

694–696

Don't you want to make this a reference? You are making a copy of a vector.

[ecbeckmann]

Good point, I'll change this in a future patch

[ruiu]

You can do it in a single-line follow-up patch. It is actually more preferable than doing it in an unrelated patch.

llvm/lib/Object/WindowsResource.cpp

for (auto DataNodes : DataEntriesTreeOrder) {
     CurrentOffset += sizeof(llvm::object::coff_resource_data_entry);
     CurrentRelativeOffset += sizeof(llvm::object::coff_resource_data_entry);
   }
 }
 
 void WindowsResourceCOFFWriter::writeDirectoryStringTable() {
   // Now write the directory string table for .rsrc$01
   uint32_t TotalStringTableSize = 0;
   for (auto String : StringTable) {
     auto *LengthField =
-        reinterpret_cast<uint16_t *>(BufferStart + CurrentOffset);
+        reinterpret_cast<support::ulittle16_t *>(BufferStart + CurrentOffset);
     uint16_t Length = String.size();
     *LengthField = Length;
     CurrentOffset += sizeof(uint16_t);
     auto *Start = reinterpret_cast<UTF16 *>(BufferStart + CurrentOffset);
     std::copy(String.begin(), String.end(), Start);
     CurrentOffset += Length * sizeof(UTF16);
     TotalStringTableSize += Length * sizeof(UTF16) + sizeof(uint16_t);
   }