`cdh` causes the `fish_prompt` event to be sent twice

[lrworth]

lukeworth@Lukes-MBP /tmp> function show_prompt_details --on-event fish_prompt
      echo fish_prompt received!
  end
fish_prompt received!
lukeworth@Lukes-MBP /tmp> cd a
fish_prompt received!
lukeworth@Lukes-MBP /t/a> prevd
fish_prompt received!
lukeworth@Lukes-MBP /tmp> cdh
 a  1)  /tmp/a
fish_prompt received!
fish_prompt received!
Select directory by letter or number:

This is particularly annoying because I was using the fish_prompt event to display a wrappable environment status (c.f. #904 (comment)).

As a workaround I am using the fish_postexec event. (Um, if anyone's using vim and this starts breaking plugins, try :set shell=sh to force POSIX shell instead of fish. TIL)

Environment

[I] ➜ echo $version
3.1.2
[I] ➜ uname -a
Darwin Lukes-MBP 19.4.0 Darwin Kernel Version 19.4.0: Wed Mar  4 22:28:40 PST 2020; root:xnu-6153.101.6~15/RELEASE_X86_64 x86_64 i386 MacBookPro11,5 Darwin
[I] ➜ echo $TERM
xterm-256color

[zanchey]

This is caused by read. It doesn't seem to me that that should fire the fish_prompt event.

[mqudsi]

I just checked and the only non-prompt related usage of fish_prompt (i.e. would be relevant to read) shipping with fish itself is enabling bracketed paste, which begs the question: is bracketed paste intentionally enabled for interactive reads executed by a command?

• If yes, then we would stop firing fish_prompt in read and use a new event handler fish_read with an accompanying function to enable bracketed paste when it's fired.
• If no, then we would just stop firing fish_prompt in read altogether.

[faho]

is bracketed paste intentionally enabled for interactive reads executed by a command?

Yes. With what bracketed paste does, it should be enabled wherever possible.

Since read understands bracketed paste (because it runs the full bindings), it should be enabled there.

If yes, then we would stop firing fish_prompt in read and use a new event handler fish_read

That would work, yes.

[mqudsi]

Thanks, @faho, and I agree. Just pushed the commit and included a note in the changelog in case anyone was relying on the old behavior.