On Wed, May 03, 2023 at 12:28:58PM -0400, Jeff King wrote: > I think the main reason we do not have other callers is that urlmatch.c > implements its own percent-encoding code, and that's where we do most of > our URL handling. It does make me wonder if credential-store could > simply switch to using that, but that is probably out of scope for your > series. Poking at it a bit, I think the answer is "not easily". credential-store has broken-out elements of a URL, and wants to turn them back into one. And urlmatch.c's url_normalize() always takes a URL string as input. Even though it finds the broken-out fields as part of its work, the parsing and reconstruction of the URL are inter-mingled in a complex way (i.e., it's not just "parse it, then turn the fields back into a normalized string). I'm sure it could be refactored, but it's probably not worth it (and again, this tangent is somewhat orthogonal to your series, except that it would make these is_rfc3986 functions go away entirely). -Peff
