Re: master aeff424: Mark both ends of self-pipe a nonblocking.

Re: master aeff424: Mark both ends of self-pipe a nonblocking.

[Robert Pluim]

>>>>> On Sat, 23 Jan 2021 11:27:43 -0500 (EST), p.stephani2@gmail.com (Philipp Stephani) said:

    Philipp> branch: master
    Philipp> commit aeff424c555da7e80775482db84eecef10286fc3
    Philipp> Author: Philipp Stephani <phst@google.com>
    Philipp> Commit: Philipp Stephani <phst@google.com>

    Philipp>     Mark both ends of self-pipe a nonblocking.
    
    Philipp>     While no deadlocks caused by the blocking write end have been reported
    Philipp>     yet, marking both ends nonblocking is consistent and also recommended
    Philipp>     in the GNU/Linux manpage of 'select'.
    
    Philipp>     * src/process.c (child_signal_init): Mark write end of self-pipe as
    Philipp>     nonblocking.

Iʼve not verified that itʼs due to this commit, but on macOS Iʼm
now getting:

src/emacs: reading from child signal FD: Resource temporarily unavailable

so Iʼm guessing you need to handle EAGAIN (Iʼve added some debug to
see what weʼre actually getting back from emacs_read())

Robert

Re: master aeff424: Mark both ends of self-pipe a nonblocking.

[Philipp]

> Am 26.01.2021 um 16:50 schrieb Robert Pluim <rpluim@gmail.com>:
> 
>>>>>> On Sat, 23 Jan 2021 11:27:43 -0500 (EST), p.stephani2@gmail.com (Philipp Stephani) said:
> 
>    Philipp> branch: master
>    Philipp> commit aeff424c555da7e80775482db84eecef10286fc3
>    Philipp> Author: Philipp Stephani <phst@google.com>
>    Philipp> Commit: Philipp Stephani <phst@google.com>
> 
>    Philipp>     Mark both ends of self-pipe a nonblocking.
> 
>    Philipp>     While no deadlocks caused by the blocking write end have been reported
>    Philipp>     yet, marking both ends nonblocking is consistent and also recommended
>    Philipp>     in the GNU/Linux manpage of 'select'.
> 
>    Philipp>     * src/process.c (child_signal_init): Mark write end of self-pipe as
>    Philipp>     nonblocking.
> 
> Iʼve not verified that itʼs due to this commit, but on macOS Iʼm
> now getting:
> 
> src/emacs: reading from child signal FD: Resource temporarily unavailable
> 
> so Iʼm guessing you need to handle EAGAIN (Iʼve added some debug to
> see what weʼre actually getting back from emacs_read())

Maybe the right approach is to just ignore (expected) errors from read and write on this pipe.  Writing is only used to notify pselect, and attempting to read the dummy byte twice shouldn’t hurt.