Regression in urllib.parse.parse_qsl(None) behavior

[msullivan]

Bug report

Bug description:

#115771 introduced a change in some currently working behavior. Previously:

import urllib.parse
>>> urllib.parse.parse_qsl(None)
[]

but now it raises TypeError: cannot convert 'NoneType' object to bytes.

(parse_qs also broke, since it calls parse_qsl)

I think it's obviously pretty debatable whether the old behavior was desirable but there is popular code in the wild that depends on it: https://github.com/encode/httpx/blob/7df47ce4d93a06f2c3310cd692b4c2336d7663ba/httpx/_urls.py#L431-L433.

(I stumbled across this when one of my colleagues was accidentally using a current dev branch of 3.11 and started getting mysterious errors from inside the bowels of the httpx client.)

CPython versions tested on:

CPython main branch

Operating systems tested on:

Linux

Linked PRs

• gh-116764: Fix regressions in urllib.parse.parse_qsl() #116801
• [3.12] gh-116764: Fix regressions in urllib.parse.parse_qsl() (GH-116801) #116894
• [3.11] gh-116764: Fix regressions in urllib.parse.parse_qsl() (GH-116801) #116895

[msullivan]

@serhiy-storchaka If there's agreement that the old behavior when passing None should be preserved, I can submit a patch.

[serhiy-storchaka]

Thank you for your report @msullivan. This looks as a serious reason to restore support of false values, even if it was not intentional behavior.

I already started to write a PR when I requested for creating this issue. The fix was trivial, but there was substantial lack of tests, and new tests exposed other flaws in this code, so the final code is more complicated.

[serhiy-storchaka]

Opened #116897 for deprecating support of generic false values.