Re: Multiple concurrent coprocesses

[Zachary Santer]

> Something more than the trivial cases I've thrown at it. I'm interested in
> whether or not the code has deadlock problems in the presence of
pipelines.

I've considered using sed with the --unbuffered flag in a coprocess, so I
could use a single sed invocation to handle arbitrary lines I feed it, as I
find them. In this case, the parent process would be echoing a line into
the coprocess and then immediately reading sed's output. Without sed's
--unbuffered flag, I can see how this could lead to a deadlock, as sed's
waiting for the parent process to fill its buffer while the parent process
is waiting for sed's response to the one line. If you're expecting one line
back for each line you feed into a coprocess containing a pipeline, you'd
need every command in the pipeline to have a similar option or just work in
a minimally-buffered manner out of the box. I don't know if anonymous pipes
themselves always wait to fill a buffer before sending their contents on.
Am I getting your meaning, though? I don't see how this would be more of a
problem with multiple coprocesses than it is with one.

I feel it's up to the Bash programmer to recognize when something like
the above could happen, and find an alternative solution if it does. It
wouldn't be the fault of Bash itself if some commands buffer input or
output no matter what you do.

In the case of my own script, there are things for the parent process and
both coprocesses to do at the same time; and buffering by the pipes
connecting the coprocesses to the parent process has the potential to
increase parallelism. All of this is still the case with the setup
currently in use, with FIFOs and automatic file descriptors, and I've not
run into any issues.

Zack

On Mon, Mar 22, 2021 at 9:33 AM Chet Ramey <chet.ramey@case.edu> wrote:

> On 3/21/21 4:38 PM, Zachary Santer wrote:
> > Okay. I've at least seen job control report when each coprocess I
> created
> > on the command line has ended, instead of waiting until they're all
> dead.
> > And the coproc name going through name expansion is nice. Unless and
> until
> > I'm allowed to bring my script home, I won't be able to show off what
> it's
> > able to do, taking advantage of this Bash installation, though.
> >
> > I'm curious what satisfactory testing would look like.
>
> Something more than the trivial cases I've thrown at it. I'm interested in
> whether or not the code has deadlock problems in the presence of pipelines.
>
> --
> ``The lyf so short, the craft so long to lerne.'' - Chaucer
>                  ``Ars longa, vita brevis'' - Hippocrates
> Chet Ramey, UTech, CWRU    chet@case.edu    http://tiswww.cwru.edu/~chet/
>