Memory leak in cmark_consolidate_text_nodes

[elibarzilay]

AFAICT, the call it does a call to cmark_node_set_literal copies the string from buf, but that string is never copied. It's easy to verify the leak with running it in a loop and looking at the memory (which is how I caught it) then adding a free for the detached string and the leak is gone.

[I'd submit it as a PR, but it seems silly to put stuff in an allocated string, then copying it and freeing the first. A more proper solution would be some way to signal a request to not copy the text (which cmark_node_set_literal could get as a new argument which will be passed on to cmark_chunk_set_cstr) but that's an API change, so I'm just reporting it...]

[jgm]

This can also be revealed by

valgrind --leak-check=full --dysmutil=yes \
  build/src/cmark --normalize README.md

I have improved 'make leakcheck' to detect this case.

Could you say more about your "more proper solution"? I
didn't quite understand. I just pushed a fix that avoids
allocating the extra string, but you may have an even better
idea.

[elibarzilay]

Well, what I tried was adding a nocopy argument to cmark_node_set_literal, and that leads to a new argument to cmark_chunk_set_cstr. It's not great, especially since it adds an argument to a public function -- but maybe it makes sense to have it there. (And since I don't know the answer I left it as a comment only.)

I'm not sure about it, but it looks like after your fix it still double-copies the string (allocates the string buffer, then allocates a copy, then frees the first copy).