[PATCH] D26016: BitcodeReader: Require clients to read the block info block at most once.

[Jordan Rose via llvm-commits]

jordan_rose added a comment.

That's how I read it too, but that's reasonable behavior if you have several data blobs of the same kind. Your file can be { MAGIC_NUMBER, { BLOCKINFO, DATA, DATA }, { BLOCKINFO, DATA, DATA }, …} instead of { MAGIC_NUMBER, BLOCKINFO, { DATA, DATA }, { DATA, DATA}, …}. Why is the former better for concatenation than the latter? Because you don't need to know the size of the BLOCKINFO.

It's not a //huge// advantage, but someone could certainly be making use of it, and this will break them. (I don't think we're currently using it but I could check.)

Again, if the two BLOCKINFO structures are different then you can't reuse the cursor today, so this is effectively no change there. I guess it helps catch reuse mistakes.


https://reviews.llvm.org/D26016