[PATCH] D41846: [DWARFv5] MC support for MD5 checksum

[Paul Robinson via Phabricator via llvm-commits]

probinson added inline comments.


================
Comment at: llvm/include/llvm/MC/MCDwarf.h:56
+  /// The MD5 checksum, if there is one.
+  MD5::MD5Result *Checksum = nullptr;
 };
----------------
aprantl wrote:
> Optional<MD5::MD5Result>?
This is a non-owning pointer to data allocated in MCContext.  Does that change your opinion?  (Or Blaikie's?)


================
Comment at: llvm/lib/MC/MCParser/AsmParser.cpp:3361
+      Checksum = fromHex(Checksum);
+      assert(Checksum.size() == 16);
+      CKMem = (MD5::MD5Result *)Ctx.allocate(sizeof(MD5::MD5Result), 1);
----------------
aprantl wrote:
> I think I would feel more comfortable with a report_fatal_error on invalid inputs.
Potentially being user input, this ought to be using the check() mechanism instead of an assertion.  Sorry about the brain-lapse here.


================
Comment at: llvm/test/MC/ELF/debug-md5.s:4
+        .file 1 "dir1/foo"   md5 "00112233445566778899aabbccddeeff"
+        .file 2 "dir2" "bar" md5 "ffeeddccbbaa99887766554433221100"
+        .loc 1 1 0
----------------
aprantl wrote:
> Just out of curiosity: why did you opt for a string literal instead of a hex numeral? Is there some precedent for that?
I believe .cv_file does it that way (the CodeView version of .file).  IIRC the integer-parsing functions don't have a way to return a 128-bit result.  Handled as a string, I can parse it directly into the allocated MD5Result.


https://reviews.llvm.org/D41846