Proposal: re-direct stderr to file for Nushell, at least

[Gastove]

Hello!

So: nushell has the idiosyncratic property of providing things like deprecation warnings... which you cannot silence. This means that every time they change their API -- which is not super infrequent, given the youth of the project -- the shell keeps working but exec-path-from-shell breaks. This is because, currently, stdout and stderr are both captured by call-process in the same buffer, which winds up looking like:

❯ nu -i -l -c "[ $env.\"LSP_USE_PLISTS\"?, $env.\"GPG_TTY\"? $env.\"PATH\"?] | to json"
Warning: nu::parser::deprecated

  ⚠ Flag deprecated.
    ╭─[/nix/store/xp056rsbl9jn312zdrlv1b1dbybfbh27-atuin-nushell-config.nu:18:19]
 17 │ let _atuin_pre_execution = {||
 18 │     if ($nu | get -i history-enabled) == false {
    ·                   ─┬
    ·                    ╰── get --ignore-errors was deprecated in 0.106.0 and will be removed in a future release.
 19 │         return
    ╰────
  help: This flag has been renamed to `--optional (-o)` to better reflect its behavior.

[
  "true",
  "/dev/pts/1",
  [
    "/home/rdonaldson/.config/carapace/bin",
    "/you/get/the/idea/this/part/is/valid/json"
  ]
]

It's "working", it's just not working, y'know?

For Nushell, at least, I'd like to suggest some version of the following (which I'm happy to PR):

• Re-direct stderr to a log file at a configurable path.
• Switch to a different Emacs process creation function.

Personally, #1 sounds just great to me -- this could be behind a defcustom or similar if it's not a behavior that should be activated all the time. I'm also game to fix this in my own configs if this isn't a behavior you want in this library.

Thanks so much for the excellent project -- this and use-package have made my life vastly simpler.

[purcell]

Emacs processes can also be created such that their error output goes to a separate buffer, which would be perfect in this case. IIRC you pass a list of (STDOUT STDERR) to the call-process function. Would be happy to have a patch for this if you want to give it a go. One would probably specify the buffer *exec-path-from-shell: nushell stderr* and kill it if empty after running the process.

[Gastove]

Hello! Sorry for the delay, life is a lot.

Okay, so: I finally got around to looking at this, and unfortunately, call-process can only use a real file for stderr redirection. Here's the relevant chunk of docstring:

Third argument DESTINATION specifies how to handle program’s output.
("Output" here means both standard output and standard error
output.)
If DESTINATION is a buffer or the name of a buffer, or t (which stands for
the current buffer), it means insert output in that buffer before point.
If DESTINATION is nil, it means discard output; 0 means discard
 and don’t wait for the program to terminate.
If DESTINATION is ‘(:file FILE)’, where FILE is a file name string,
 it means that output should be written to that file (if the file
 already exists it is overwritten).
DESTINATION can also have the form (REAL-BUFFER STDERR-FILE); in that case,
 REAL-BUFFER says what to do with standard output, as above,
 while STDERR-FILE says what to do with standard error in the child.
 STDERR-FILE may be nil (discard standard error output),
 t (mix it with ordinary output), or a file name string.

Currently, I see two options that seem sensible (to me):

Direct stderr to a temp file, then either:

1. On non-zero exit code, either visit the file, or include the file name/path in the error message.
2. On non-zero exit code, read the temp file contents into an *exec-path-from-shell: nushell stderr* buffer.

Option 1 is marginally easier -- or at least, is quite short; option 2 requires just a little more code but preserves the "errors go in buffers with sensible names" pattern very tidily.

Any preferences?

[purcell]

Yeah, I think you'd want option 2. What I did in envrc.el was send stderr to a temp file then insert it into an output log buffer that contains the stderr of many invocations. In reformatter.el, which I think is more similar to what we want for exec-path-from-shell, I have a stably-named error buffer and always insert the stderr file into it: if the command actually failed, I then use display-buffer, which the user can customize independently.

So I'd suggest:

• Delete *exec-path-from-shell: nushell errors* buffer if it exists
• Run nushell command with stderr going to a tempfile
• Insert the contents of that file into *exec-path-from-shell: nushell errors* regardless of nushell command exit code
• On failure exit code, use display-buffer to show that buffer

[Gastove]

This sounds entirely sensible. I'll get something pulled together and tested.

Thank you!