Fix rendering when prompt_suffix is empty

[gene1wood]

This changes the implementation of the fixes in #1836 and #2093 to address the problem that they created when click.prompt or click.confirm has a prompt_suffix value which is empty.

This still fixes the problems reported in #665 and #2092 but does so without modifying the prompt and adding a trailing space.

Previously, the problem was solved by trimming all trailing space characters from the prompt and then adding a single trailing space to every prompt. This was only visible if the last character in the prompt wasn't a space character. In those cases a prompt like "This is my prompt" would render as "This is my prompt ". Additionally, in the previous solution, if the err=True argument was passed, that final trailing space character that was added would be sent to STDOUT while the rest of the prompt would go to STDERR. This also wasn't very apparent since it was a space character.

This different implementation uses the last character of the prompt to work around #665 and #2092 instead of adding a space character. This fixes most of the potentially confusing behavior. It does retain one confusing behavior which is described above, where if you use err=True the last character of the prompt is sent to STDOUT instead of STDERR. Previously, this last character sent to STDOUT would always be a space character. Now if the last character of the prompt is not a space character, that final character will go to STDOUT.

For example if you call

click.prompt("bar", prompt_suffix="", err=True)

The ba will go to STDERR and the r will go to STDOUT. Previously bar would go to STDERR and an added would go to STDOUT.

This odd behavior, which is already present but just applies to a space character, is worth accepting so that the prompt_suffix correctly renders the suffix as it did in version 7.1.2 and prior.

This also adds unit tests for cases where prompt_suffix is empty.

This fixes #3019

[kdeldycke]

I agree with you that having the last character of the prompt being sent to <stdout> instead of <stderr> is confusing.

But your PR has the merit of:

• making the behavior more deterministic
• not hiding the edge-case visually by resorting to spaces
• adding unit-tests on cases that were not covered before

For me this is a case of "worse is better", in which we gradually tends to the ideal behavior. The ideal behavior being a refactor of the prompt() / confirm() methods.

I am in favor of merging it in stable for a Click 8.3.1 release.

But before that, could you:

• add a .. versionchanged:: 8.3.1 item in the docstring of each function to explain the change in behavior,
• add a changelog entry to CHANGES.rst,
• and change the branch target from main to stable please?

[gene1wood]

@kdeldycke Sure thing

I've

• Re-added the note to CHANGES.rst
• Changed the PR to be based off of stable
• Updated docstrings for confirm and prompt
  • Note : I tried to find a sweet spot between too verbose and too succinct with the docstring additions. If you'd like to see more or less detail let me know and I can tweak them.