GH-126601: pathname2url(): handle NTFS alternate data streams

[barneygale]

Adjust pathname2url() to encode embedded colon characters in Windows paths, rather than bailing out with an OSError.

• Issue: urllib.request.pathname2url() raises on NTFS alternate data stream #126601

[barneygale]

Hint for reviewers: if it helps to grok the changes, have a look at the initial commit in this MR (0ed23a5), which is a smaller patch.

Nevermind - I found a nice middle-ground I think!

[zooba]

LGTM with one trivial change (or a reason we can't make it).

We should probably also .. versionchanged any functions that may be affected by this.

[barneygale]

Thanks Steve!

We should probably also .. versionchanged any functions that may be affected by this.

I'm intending to backport this to 3.13 and 3.12 because it seems like a straightforward bugfix to me. So I'm not sure if .. versionchanged would be appropropriate. What say you?

[zooba]

I haven't tracked down what functions are impacted, but my assumption would be that existing users may notice behaviour changes where the path is an arbitrary input.

I'm most concerned about opening up a potential exploit, especially since alternate streams are not commonly used (and would rarely traverse a URL). Perhaps there's a scenario that's worth the risk to have the change, but I'm not aware of it, so I'm inclined to play it safe on a strict correctness issue that isn't actively hurting anyone.

[barneygale]

Thanks again. I've added a .. versionchanged:: bit to the docs.

pathname2url() isn't used by any other standard library code, so we only need one change to the docs I think.

[zooba]

So the only question now is whether to backport, and all I can say is "I'm not sure" (and so default to saying no). Users running into this issue in real code would be pretty persuasive.

[barneygale]

I'm erring towards not backporting, on reflection. I don't think the current behaviour is great, but it's better to get an exception than an incorrect result. I only logged this issue because I spotted it in the code (and I've fixed similar issues in pathlib reported by Eryk Sun) - I'm not aware of any user running into it.