Eliminate // in paths

[edk0]

Incoming paths are redirected to remove repeated slashes upon matching. Rules containing repeated slashes collapse them to just one and send a warning.

I haven't included a way to turn this off; do people have reasons to?

closes #1132

[davidism]

The only problematic but somewhat valid case I can think of is /scrape/https://example.com, so it might be good to be a little conservative and only merge slashes that aren't preceded by :.

There should be a merge_slashes=True attribute on Rule and Map, similar to strict_slashes, to control this behavior.

Like strict_slashes, Rule.match should raise an exception that MapAdapater handles.

I was considering only merging slashes in static parts, not variable parts, of matched URLs, but that would get really complicated. I think it's OK to just merge everything.

[davidism]

Also remember to add a changelog entry and a versionchanged note to the docstring.

[edk0]

Like strict_slashes, Rule.match should raise an exception that MapAdapater handles.

I was considering only merging slashes in static parts, not variable parts, of matched URLs, but that would get really complicated. I think it's OK to just merge everything.

Doing the work in Rule.match would mean each rule had to repeat the transformation, which I didn't like (I thought it would be slow, but since measured, and it isn't really), or altering the generated regexp, which might be a pain but would make it fairly easy to do the "only merge in static parts" bit.

Oh, and probably if we want to go /a//b -> /a/b/ it should take only one redirect. Hmm.

[edk0]

So I've done all that (I think). The top commit makes only static parts affected—as you can see it's not a whole lot of extra work, but I'm happy to remove it if you prefer the other thing. (Personally, I like that it eliminates the special case for ://)

One thing testing this threw up is that a <path> can't start with /. With merge_slashes on, any inital slashes are eaten by the merging. Without it I think it will always 404. Anyway, it's trivial to let paths start with a slash, if you want.

[RDProjekt]

One reason for supporting the switch to turn off collapsing slashes is that Werkzeug is used by moto (https://github.com/spulec/moto), a library emulating Amazon AWS infrastructure. Since double slashes are meaningful for Amazon's servers (see e.g. getmoto/moto#1637),
unconditionally collapsing slashes may make the full emulation impossible.

Incidentally, the problem with paths starting with slashes mentioned above is also a problem for moto, so it would be great to have it fixed as well.

[davidism]

@edk0 I've rebased and update the code style for this, but I can't push. Would you enable that option?

[davidism]

@edk0 would you review this and make sure I didn't bork anything during the rebase?

[edk0]

Looks good to me.