[ok2c]

No description provided.

[arturobernalg]

@ok2c , I found a couple of regressions compared to master.

For /a/b/.., master returns /, whereas this change returns /a/b/. I reproduced this with HttpCore: a cached authentication entry for /a/b/ causes preemptive Authorization to be sent, while master waits for a challenge. The same issue occurs with encoded dot segments.

Also, foo returns fo, and since / is not in UNRESERVED, the fast path is never taken for normal absolute paths.

I think we should preserve the existing normalization semantics and add regression tests before merging.

[ok2c]

@arturobernalg Excellent findings! Thank you so much!

For /a/b/.., master returns /, whereas this change returns /a/b/

But this is wrong, is it not? It should be /a/ should it not? I kept the existing behavior for now, but I think this has to be fixed

since / is not in UNRESERVED, the fast path is never taken for normal absolute paths.

Corrected. Please do another pass.

[arturobernalg]

@arturobernalg Excellent findings! Thank you so much!

For /a/b/.., master returns /, whereas this change returns /a/b/

But this is wrong, is it not? It should be /a/ should it not? I kept the existing behavior for now, but I think this has to be fixed

since / is not in UNRESERVED, the fast path is never taken for normal absolute paths.

Corrected. Please do another pass.

But this is wrong, is it not? It should be /a/ should it not? I kept the existing behavior for now, but I think this has to be fixed

One remaining difference I noticed: consecutive slashes now bypass normalization. For /a//b, master returns /a/, while the new fast path returns /a//. this change intentional?

[ok2c]

One remaining difference I noticed: consecutive slashes now bypass normalization. For /a//b, master returns /a/, while the new fast path returns /a//. this change intentional?

@arturobernalg No, it is not. I will look into it.

[ok2c]

@arturobernalg Please do another pass. Your feedback was super-valuable!

[arturobernalg]

@ok2c LGTM