bug#63539: 29.0.90; TRAMP fails to detect shell prompts containing ]

[Michael Albinus <michael.albinus@gmx.de>]

Spencer Baugh <sbaugh@janestreet.com> writes:

Hi Spencer,

>>> I'm guessing the reason is because this version of the regexp doesn't
>>> support multiple escape codes, or blanks in between/at the end of the
>>> escape codes?
>>
>> Blanks as parts of escape codes are unlikely.
>
> In the middle, perhaps, but after the escape codes seems quite likely to
> me.  At the very least, my prompt (which worked on Emacs 29 and no
> longer works with your change) has blanks after the escape codes.
>
> For reference, my now-broken prompt is:
> PS1='\[\033[01;32m\]\u@\[\e[${hostnamecolor}m\]\h\[\033[01;34m\] \w \$\[\033[00m\] '

Thanks. But nobody forbids that you use "_" (underscore) instead of
blanks in your prompt, for example. Not very likely, I know. But I want
a solution now and forever.

>> Multiple escape codes are not regarded indeed, my patch did expect that
>> there is not more as one escape code at the end of the shell prompt.
>>
>> Your example is an indication of escape codes, mixed inside the shell
>> prompt. My first attempt (removing all escape codes from the buffer
>> region) did handle that. But as said, there were other regressions with
>> that attemot.
>
> What's wrong with my diff, though?  It produces behavior which is much
> closer to how TRAMP behaved before your change, while still fixing my
> issue.

I've pushed another change to master which goes back to my first
approach: Remove escape sequences in the buffer, before you look for the
prompt.

The difference is now, that this happens *only* in the initial
connection phase with a remote host, and not later. After that phase we
have a PS1 setting applied by Tramp, which is escape sequences free.

This approach should avoid the problems I've seen when I've applied a
similar change some days ago. I still need to run all regression test
cases, but a check for crucial tests tell me that we could be optimistic
now :-)

Could you pls check this change?

>> Hmm. Could you please send me a tramp-verbose 10 trace of your failing test?
>
> I'm not sure how useful this is, but:

I meant the Tramp debug buffer, producesd by setting tramp-verbose to
10. But since you have shown me your PS1 setting above, this isn't
needed anymore, thanks.

Best regards, Michael.