Why is Fish's psub so simple and effective compared to other shells?

[da99]

The psub function works great for my needs. The implementation is simple and quite elegant.

Fish's psub essence from: https://github.com/fish-shell/fish-shell/blob/master/share/functions/psub.fish#L49

cat >/tmp/pipe.fish &
echo /tmp/pipe.fish
#     diff   (echo abc | psub )  (echo def | psub)

However, a non-Fish equivalent would not work. This would block in other POSIX-y shells:

diff   $(echo abc | fishy.psub.sh )  $(echo def | fish.psub.sh)
# blocks until you write to the named pipe

How does Fish get away with implementing process substitution like that?

[zanchey]

it is a mystery

I suspect it would be helpful if we could see the implementation of fish.psub.sh, as the FIFO implementation in fish doesn't actually a) work or b) get effectively used (see #1040).

[da99]

In fishy.psub.sh, it would be a direct translation of Fish's psub function:

  my_pipe=path/to/temp/file
  mkfifo $my_pipe
  cat > $my_pipe &
  echo $my_pipe

But, in a non-Fish shell, this would just block until output was sent to the named pipe from somewhere else. In Fish, "it just works". (But, as @zanchey pointed out, it's not perfect.)

[da99]

"it is a mystery" is good enough for me.

In case anyone is curious how Fish's process substitution can be implemented in other shells, there is this discussion (specific to mksh): http://unix.stackexchange.com/a/273227/42107

It's a lot trickier compared to Fish. One more way Fish stands out from the crowd.

[zanchey]

FYI, the mkfifo stuff in fish does not work as it blocks with large enough outputs - psub always uses a temporary file.

[da99]

@zanchey Thanks for correcting me. Apologies for making you repeat yourself.

When do the temp. files psub creates get erased from the system? Reboot?

[zanchey]

After the pipe is closed - see the function $funcname --on-job-exit caller part of the psub function.

[da99]

Is this a unique feature to Fish? (on-job-exit)
I never encountered anything like that on other shells.