GH-127090: Fix urllib.response.addinfourl.url value for opened file: URIs

[barneygale]

The canonical file: URL (as generated by pathname2url()) is now used as the url attribute of the returned addinfourl object. The addinfourl.url attribute reflects the resolved URL for both file: or http[s]: URLs now.

• Issue: urllib.request.urlopen('file:...').url incorrect value #127090

[serhiy-storchaka]

LGTM in general, but there are not enough tests.

I think that you can add checks for url in test_file_notexists and maybe in test_basic and test_copy in urlretrieve_FileTests.

[serhiy-storchaka]

On other hand, the current code removes port and fragment from url, but the changed code will preserve them. I do not know what is more correct.

For history of the current code see 18e4dd7 and 6057ba1 (bpo-8656/gh-52902).

[barneygale]

Thank you Serhiy, that's very helpful.

I'll mark this PR as a draft for now. Once #127125 and #127194 land, I'll adjust this patch to call pathname2url() to canonicalise the URL, which will strip the port and fragment, restoring previous behaviour. It looks like urlopen() canonicalises HTTP URLs, and I think it makes sense to do so here.

[barneygale]

I think that you can add checks for url in test_file_notexists and maybe in test_basic and test_copy in urlretrieve_FileTests.

I've expanded test_file_notexists, but the urlretrieve() tests aren't relevant here - they use a different codepath for handling local files (with its own set of problems...)

On other hand, the current code removes port and fragment from url, but the changed code will preserve them. I do not know what is more correct.

The existing code raises URLError if a port is given, and that's tested in test_urllib2.HandlerTests.test_file. I've slightly expanded that test to also cover fragment stripping.

[serhiy-storchaka]

LGTM.

[serhiy-storchaka]

Shouldn't it be backported?

[picnixz]

On the issue, Barney decided that it was not worth backporting: #127090 (comment).