Re: tail: --pid option does not work when input is a FIFO

[Pádraig Brady]

On 17/10/2024 22:38, Bernhard Voelker wrote:
> Hi *,
>
> I noticed that 'tail --pid' does not work when the file to follow is a FIFO.
> Shouldn't --pid behave the same regardless whether it's a pipe or not?
>
> # Good case.
> $ rm f; touch f  # tail to follow regular file.
> $ sleep 5 & timeout 10 tail -f --pid=$! f; echo $?
> [3] 1030528
> [3]+  Done                    sleep 5
> 0
>
> # Bad case.
> $ rm f; mkfifo f  # tail to follow a FIFO file.
> $ sleep 5 & timeout 10 tail -f --pid=$! f; echo $?
> [3] 1030559
> [3]+  Done                    sleep 5
> 124
>
> Looking at strace output, one can see that kill() is only invoked once in the 
> FIFO case.
>
> FWIW: it doesn't make a difference whether one uses -f or -F, whether the
> monitored file is written to by another process or stays empty, or whether one
> tries with the  --sleep-interval option.

In my testing the initial open() of the fifo blocks,
before the -f loops are processed.
If the monitored file was written by another process,
this _does_ unblock things for me, and --pid is processed.

The attached patch avoids the blocking behavior with fifos,
and thus honors the --pid checking in parallel (and passes all tests).

However I've only thought about this for a couple of minutes,
so need to consider all the implications.

cheers,
Pádraig

tail-fifo-block.patch
Description: Text Data