writing about things, sometimes.

Multiplexing data in sh(1)

Written by Omar Polo on 17 October 2019.

In my previous article I used a bash(1) construct to send the same data to two different places.

The fact that I was using a bash-specific (more or less) construct really bothered me this time. So I thought if there is a way in sh(1) to take a piece of data from a pipeline and send it to two different places.

The canonical way to do this is tee(1), and it's what I've used it the linked article:

something | tee >(cmd) | something-else

(the >(cmd) construct is bash, and zsh I think, specific: it executes cmd in a subshell and execute tee with something like /dev/fd/63 -- at least on linux and OpenBSD.)

Now I'll try to describe three different approaches to emulate the same behavior of >(...) in pure sh(1), trying to outline both the advantages and the cons.

temporary files

That is:

something | tee a-file | something-else
cmd < a-file

# and then, probably
rm a-file

(or even better with mktemp(1))

This is the most simple approach I can think of. The obvious disadvantage is the need to create and delete files quickly. This shouldn't be a huge problem in most cases, but for script do this constantly and (possibly) quickly it can be. Plus, at least for me, it doesn't seem very elegant, but it's subjective.

fifo to the rescue

An almost similar approach is to use fifo. A fifo, aka a named pipe, is a special file that you can write to and read from, in a FIFO fashion.

mkfifo a-fifo
something | tee a-fifo | something-else
while read line < a-fifo; do echo $line; done | cmd

This is especially useful if you need to have a long-running script that needs to duplicate a stream from one source. You create the fifos at the startup and then you're done, instead of constantly creating temporary files over and over again. An annoying thing is that you need to wrap the read from the fifo in some sort of loop: AFAIK to read the next item in a fifo you need to close and re-open it.

An unexpected sed

Let's add another requisite: other than dumbly copy the stream, suppose that you want also to apply some sort of transformation to one branch of the duplication, like filter out something.

Well, you can adapt the last two examples by adding a | sed '...'. Or you can use the w command of sed.

I'll be a bit more specific: let's say that you need to dispatch something taken from a single stream and feed n subprocess. It's exactly what I've tried to achieve in my last post: I've used >(cmd) to dispatch the unbound statistics to different fifo, and then I had different processes pulling out those stats from the fifo to draw graphs.

The code was like

unbound-control stats			\
| tee >(grep something > a-fifo)	\
| tee >(grep something-else > b-fifo)	\
| ...

Well, that can be rewritten as

unbound-control stats			\
| sed 's/something/&/w a-fifo'		\
| sed 's/something-else/&/w b-fifo	\
| ...

by leveraging some sed behaviors:

  1. every line read is printed to stdout (eventually transformed)
  2. after the s command you can use w to write the transformed lines to a local file

I've got this illumination by reading Sed - An introduction and Tutorial by Bruce Barnett.

I tried to use sed 'g/something/w a-fifo' but sed complained about "extra characters after command", so I assume that the no-op s/something/&/ is needed.

Edit: I tried with g/... because I thought that sed had a g command that behaves like the vi(1) or sam(1) g command. sed has a g command, but does something different. The correct form would be sed '/something/w a-fifo' because sed, like awk(1), has the ability to perform commands only on lines that match a regular expression.

I find this last technique pretty, even prettier than the >(grep ... >) idiom I used previously since it avoids the subshell to do the filtering and because it does not make assumption on the shell used -- it only requires a POSIX shell.

I hope you found this interesting. At least to me it really is.