psub for long-running commands - fish vs zsh/bash

[czak]

I'm trying to use psub to feed a long-running process' output into another.

To simplify:

# fish
$ cat (begin; echo First; sleep 5; echo Second; end | psub)
# nothing appears for 5 seconds ...
First
Second
$

With zsh/bash, the First line of output is immediately available to cat.
And this is the behavior I'm hoping to achieve:

# zsh
$ cat <(echo First; sleep 5; echo Second)
First
# ^ this appears immediately, then waits 5 seconds...
Second

Is it supported in fish? Am I missing something?

I tried both psub and psub -F.
I'm using fish 2.7.1.

[faho]

psub has some issues. In this case you're hitting #1040 - it fully writes the file before returning.

The question I have here is why you're using it. Usually the easiest way to get a command to accept data from another is to simply pipe it. So instead of command1 (command2 | psub), you'd do command2 | command1 or perhaps command2 | command1 -.

If there really is no way to get command1 to read from stdin, you can use mkfifo:

mkfifo /tmp/fifo
command2 >/tmp/fifo &
command1 /tmp/fifo

which is basically what psub is supposed to do in the background.

[czak]

Thanks for the reply @faho!

My use case is this: I'm implementing a command1 (from your examples).
It's an interactive terminal program which needs input from both:

• a continuous stream of text (ideally, as stdin coming from command2's stdout)
• keyboard

A real-world use case would look like this:

$ rails server | my_interactive_app

The idea is that rails server keeps producing output, and my_interactive_app receives it at stdin, while continuously accepting keyboard input from the user. It's a bit tricky to get piped stdin and keyboard working at the same time, but I managed to get past that.

But then I ran into another challenge, simplified to:

$ begin; sleep 10; echo "Some output"; sleep 10; end | true

This blocks the entire pipeline for 10 seconds, even though true exits immediately.
In my case (rails server | my_interactive_app), it's very possible that rails server will not be producing any output for periods of time, and I want to be able to quit my_interactive_app at any time. However, quitting returns to console only when rails server tries to produce some output (i.e. not immediately).

Note: I recognize I don't know much about pipes and might be missing something trivial.

Process substitution (on bash/zsh for now) is free of this "lock". my_interactive_app <(rails server) allows me to quit my_interactive_app at any time. However, this keeps rails server running in the background, which is not ideal either.

tl;dr

I'm looking for a clean way to:

• pipe command2 | command1
• allow continuous stream of output from command2
• allow command1 to close the pipe (or terminate command2?) when it decides to

[czak]

Ah the power of rubberducking :)

I think what I really need is this syntax instead:

$ my_interactive_app rails server

I can have my_interactive_app spawn and control the rails server process and its output. It's more work for my_interactive_app, but the syntax is even cleaner.