Zsh Mailing List Archive
Messages sorted by: Reverse Date, Date, Thread, Author

Re: feature request

Op 16-01-20 om 23:10 schreef Yefim Vedernikoff:
What would it take to add a new environment variable like $OUTPUT to keep
track of the last output?

The standard 'tee' utility does something like what you want: it copies its standard input both to a file and to standard output. So if you want to save the output of a command, you pipe it into 'tee' like this:

    somecommand | tee OUTPUT

That's not a variable, but maybe it's close enough for you. Instead of $OUTPUT, you can use $(< OUTPUT).

What it would take to implement your feature request to make this permanent and transparent? Zsh would have to capture standard output (and probably standard error) itself, so it can transparently act like 'tee' except for duplicating output to a variable instead of a file.

One way to deal with commands producing huge or infinite output would be to cap the size and delete old lines if the threshold is exceeded -- like terminal scrollback buffers do.

So it's possible, but also non-trivial, and it would have a serious drawback: since this requires capturing standard output, the commands you run would not consider themselves connected to a terminal, so e.g. 'ls' would not produce output in columns and colours by default. And full-screen editors might have no idea what to do.

To experiment with this side effect, start a new zsh process (to avoid screwing up your current session) and then try some process substitutions:

    exec > >(tee stdout.txt) 2> >(tee stderr.txt)

to log standard output and standard error separately, or

    exec > >(tee output.txt) 2>&1

to combine them. As expected, 'ls' output becomes very boring. As for editors, oddly enough, emacs seems to completely ignore that it's not on a terminal, but vim acts like it's on a 1970s teletype, and joe becomes very unhappy.

- M.

modernish -- harness the shell

Messages sorted by: Reverse Date, Date, Thread, Author