On Thu, Oct 15, 2020 at 11:41:36AM -0700, Junio C Hamano wrote: > Jeff King <peff@xxxxxxxx> writes: > > >> - * If "ss" is not NULL, compute SHA-1 of the exclude file and fill > >> + * If "oid_stat" is not NULL, compute SHA-1 of the exclude file and fill > > > > Makes sense. This changed as part of 4b33e60201 (dir: convert struct > > sha1_stat to use object_id, 2018-01-28). Perhaps it would likewise make > > sense to stop saying "SHA-1" here, and just say "hash" (or even "object > > id", though TBH I think the fact that the hash is the same as an > > object-id is largely an implementation detail). > > I do not quite get your "though TBH", though. I do agree with you > that it is an implementation detail that an object is named after > the hash of its contents, so saying "compute object name" probably > makes sense in more context than "compute hash" outside the narrow > parts of the code that actually implements how object names are > computed. So I would have expected "because TBH", not "though TBH". Sorry, I was just confused. The implementation detail I meant is that we are using a "struct object_id" in the oid_stat type (and also that "oid_stat" is likewise exposing too much). I thought this was another version of our stat_validity, where we are checking quickly to see if a random file that is not necessarily part of the working tree has been updated. And indeed, we do use it that way for files like .git/info/exclude, where having an "object_id" is really irrelevant. But we also use it for checking the validity of tracked files, where we populate it from an actual blob (see dir.c:do_read_blob). So it is actually reasonable to expose that in the name, and to think about it as an object_id. > Anyway. Nipunn, can you fix both of them in the same commit, as > they are addressing a problem from the same cause (i.e. we are no > longer SHA-1 centric). The v2 that Nipunn sent with "oid" in the comment looks good to me. -Peff
