Nguyen Thai Ngoc Duy <pclouds@xxxxxxxxx> writes: > 2010/9/28 Junio C Hamano <gitster@xxxxxxxxx>: >> Nguyán ThÃi Ngác Duy <pclouds@xxxxxxxxx> writes: >> ... >>> - If it's recursive diff, directories are just an intermediate step. >>> All path must end with a file name. Thus, directories will be >>> unconditionally matched. >> >> Hmm, I am not sure what you mean by this. If the pathspec says a/b*/c, >> you are in "a" and are deciding if you should descend to its subdirectory, >> then you would surely want to be able to say: >> >> (1) Ah, the subdirectory I am looking at is "bar" and it does match "b*". >> It might contain "c"; I should descend into it. >> >> (2) Nope, the subdirectory I am looking at is "frotz" and it can never >> match "b*", so there is no point recursing into it. > > I did not go that far, trying to analyse the pattern. When I wrote > "immediate step" I was thinking of "*foo" pattern, which effectively > means descending to any reachable directories because '*' matches > slashes too. I think that's the worst case. > > Anyway I did not know that I could borrow some optimizations from > pathspec_matches() in builtin/grep.c. As you noticed, over time we have polished our re-implementations of the pathspec logic to avoid needlessly descending into subdirectories, and the one in the grep may be the latest incarnation of it. The logic that uses the path_simplify structure in dir.c may also offer you some inspirations. Note that the worst case will fall out as a natural consequence of doing the fallback in the dumb and simple way, so you do not have to worry too much about it ;-) I would prefer to see the new, consolidated logic to at least know the easy optimizations we already have. Thanks. -- To unsubscribe from this list: send the line "unsubscribe git" in the body of a message to majordomo@xxxxxxxxxxxxxxx More majordomo info at http://vger.kernel.org/majordomo-info.html