`getpass`'s `echo_char` should not affect keyboard shortcuts

[picnixz]

Bug report

Bug description:

Hitting Ctrl+U should clear the line but this is badly handled when echo_char is given:

>>> getpass.getpass()
Password: 
'bar'
>>> getpass.getpass(echo_char="*")
Password: *******
'foo\x15bar'

See #138514 (comment).

This issue stems from the fact that we switch to non-canonical mode (https://www.man7.org/linux/man-pages/man3/termios.3.html). The rationale is that in canonical mode, we only have a line-by-line feedback (so we would see the characters only when we hit ENTER, which is not really useful...)

CPython versions tested on:

CPython main branch

Operating systems tested on:

Linux

Linked PRs

• gh-138577: Mention limitations of getpass.getpass(echo_char=...) #138677
• [3.14] gh-138577: Mention Unix-specific limitations of getpass.getpass(echo_char=...) (GH-138677) #138696
• gh-138577: Fix keyboard shortcuts in getpass with echo_char #141597

[yagggi]

Hi @picnixz , this seems to be a trival fix, would you mind if I give it a try?

[picnixz]

I'm not sure this is a trivial fix actually as it's about dealing with readline capabilities. It's not just Ctrl+U that is badly handled, it's all keyboard shortcuts so this may be really tricky.

[yagggi]

Oh I see, yeah then it's tricky. Since ICANON is disabled, the best I can think of is simulating erasing logic from tty, but I'm not sure if it's worthy or not.

I'm not going to try to fix this, hope there will be a better solution.

[picnixz]

Since 3.14 is almost released, I think we should just mention this. I don't have enough time to investigate this and I don't know if we can fix it robustly.
We support erasing a single character but we do not support erasing everything that was written via Ctrl+U. If you want, you can write a documentation for that though.

@yagggi

[yagggi]

Sure! I'm working on it.

[picnixz]

I'm actually wondering whether it was the correct way to do it. Shipping a feature that is not entirely complete makes me feel bad and this limitation may be annoying. Maybe we should instead put this feature in 3.15.

@donbarbos How does it work on a Windows machine? do you have access to one? for instance what if I hit Ctrl+Backspace? does it work correctly? if not, we should perhaps revisit the entire feature and use a REPL-like approach if possible. Ok on Windows for 3.13.2, the normal behavior doesn't support those keyboard shortcuts (I hit Ctrl+Backspace) so maybe it's fine to have inconsistencies "everywhere"...

>>> getpass.getpass()
Password:
'abc\x7fdef'

Summary:

• On Unix, keyboard shortcuts work with the default behavior (old). On Windows, those shortcuts don't work.
• On Unix, keyboard shortcuts no longer work with echo_char. On Windows, I think they will still fail.
• Should we ship 3.14 with something that has a weird behavior on Unix? for now, I asked for documentation but maybe it's just better to avoid it entirely on Unix until we can make it work consistently (whether we use echo_char or not).

cc @hugovk