--- Begin Message ---
|
Subject: |
undocumented change in tail in 8.28 (pipe output monitoring) |
|
Date: |
Fri, 20 Apr 2018 16:56:23 +0200 |
Hi all,
during debugging of a background process issue where I have tail -f
feeding into a pipe, I was surprised by a change in tail introduced in
release 8.28 of GNU coreutils: The monitoring of the output pipe and
silent exit of tail e.g. in the case of
tail -f /dev/null | true
. This differs from earlier versions and also from BSD, for example. I
see some subtle portability bugs here. One one platform, killing the
end of a pipe fed by tail will cleanly end the tail process, too,
without tail attempting to write new data and failing _then_. On the
other platform, the pipe stays there until new data arrives to cause
tail to get the error trying to write to the pipe.
Regardless of what behaviour is sensible, it causes scripts that
take GNU tail's behaviour for granted to have different behaviour with
other versions of this very traditional UNIX tool.
So at the very least, this change should be documented as a possible
implementation difference in the manual, and for sure with the version
that changed behaviour, shouldn't it?
Regards,
Thomas Orgis
--
Dr. Thomas Orgis
Universität Hamburg
RRZ / Basis-Infrastruktur / HPC
Schlüterstr. 70
20146 Hamburg
Tel.: 040/42838 8826
Fax: 040/428 38 6270
smime.p7s
Description: S/MIME cryptographic signature
--- End Message ---
--- Begin Message ---
|
Subject: |
Re: bug#31225: undocumented change in tail in 8.28 (pipe output monitoring) |
|
Date: |
Fri, 20 Apr 2018 17:56:10 -0700 |
|
User-agent: |
Mozilla/5.0 (X11; Linux x86_64; rv:45.0) Gecko/20100101 Thunderbird/45.8.0 |
On 20/04/18 07:56, Dr. Thomas Orgis wrote:
> Hi all,
>
> during debugging of a background process issue where I have tail -f
> feeding into a pipe, I was surprised by a change in tail introduced in
> release 8.28 of GNU coreutils: The monitoring of the output pipe and
> silent exit of tail e.g. in the case of
>
> tail -f /dev/null | true
>
> . This differs from earlier versions and also from BSD, for example. I
> see some subtle portability bugs here. One one platform, killing the
> end of a pipe fed by tail will cleanly end the tail process, too,
> without tail attempting to write new data and failing _then_. On the
> other platform, the pipe stays there until new data arrives to cause
> tail to get the error trying to write to the pipe.
>
> Regardless of what behaviour is sensible, it causes scripts that
> take GNU tail's behaviour for granted to have different behaviour with
> other versions of this very traditional UNIX tool.
>
> So at the very least, this change should be documented as a possible
> implementation difference in the manual, and for sure with the version
> that changed behaviour, shouldn't it?
>
>
> Regards,
>
> Thomas Orgis
>
Well we did document it in NEWS as seen in:
https://git.sv.gnu.org/cgit/coreutils.git/commit/?id=v8.27-42-gce0415f
I still regard tail being more responsive to closed output being an
improvement, and the more general solution, however it was a bit terse
and broached as only an improvement.
I pushed the following change to expand on the info:
https://git.sv.gnu.org/cgit/coreutils.git/commit/?id=22424dd
Marking this as done.
thanks,
Pádraig
--- End Message ---