3.14.0a1 urljoin regression wrt //

[hroncok]

Bug report

Bug description:

The tests of lxml do (simplified):

from urllib.parse import urljoin
urljoin('file:', '/foo/bar')

• On Python 3.13.0 and earlier, the result of the urljoin call is file:///foo/bar.
• On Python 3.14.0a1, the result of the urljoin call is file:/foo/bar.

This change of behavior was introduced in fc897fc #123273 cc @serhiy-storchaka

The NEWS entry in that change does not seem to indicate this change was intentional.

CPython versions tested on:

3.14, CPython main branch

Operating systems tested on:

Linux

[ZeroIntensity]

Duplicate of #125926

[hroncok]

That seemed like a different problem to me, but perhaps the cause is the same, yes.

[serhiy-storchaka]

This is not a duplicate of #125926.

But file:///foo/bar and file:/foo/bar are equivalent URIs. And we can argue that since file: and /foo/bar have undefined authority, the result should have undefined authority as well. This is a result of the algorithm in RFC 3986, Section 5.2.2, although it need an absolute base URI.

[ZeroIntensity]

I marked this as duplicate because it seemed like there was some underlying issue introduced into urljoin. I was assuming they will both be fixed by the same PR, but I guess not, sorry!

[serhiy-storchaka]

I suggest to close this issue as "not a bug".

[hroncok]

In that case, we should at least note the change in What's New in Python 3.14

[serhiy-storchaka]

Maybe. But what should we write? This is not the only change in urljoin(). I'm planning to do more changes (see #69589, #84774).

I hesitated between backporting these changes or not. This is a long standing behavior, so user code can have workarounds or even been depending on it. AFAIK the last large changes were in 3.5, and they were not backported.

If not backport these changes, they should be documented as a new feature. Note that urljoin() still is not completely RFC 3986 conforming -- some differences I left for backward compatibility. But we perhaps should change these parts too.

[asvetlov]

It's a hard choice, but I prefer the conformance to the RFC even if the change breaks backward compatibility.

[notatallshaw]

Is the output of urljoin stable for 3.14 yet or are there more changes planned before beta?

I'd like to add Python 3.14 to pip's CI but any changes to the output will cause pip's tests to fail (pypa/pip#13138).

[serhiy-storchaka]

There may be more changes.