bug#15228: [PATCH] Close output port of I/O pipes

[Andy Wingo]

[bouncing this back to debbugs]

On Sat 25 Jun 2016 17:49, Andy Wingo <address@hidden> writes:

> On Sat 25 Jun 2016 15:51, Josep Portella Florit <address@hidden> writes:
>
>>> I dunno how much we should push this "processes are a single port"
>>> abstraction.  In many ways for OPEN_BOTH pipes it's easier to deal with
>>> an input and an output port and a PID instead of the pipe abstraction.
>>> WDYT?  We could just expose `open-process' from (ice-9 popen) to start
>>> with.  It would be good to allow other fd's or ports to map to the child
>>> as well, e.g. stderr or any particular port; but I don't know what
>>> interface we should expose.
>>
>> Since patching was inconvenient for me, I eventually used:
>>
>> (use-modules ((ice-9 popen) #:select (open-process)))
>>
>> Which works even though `open-process` is not exported.
>
> Note that this behavior of #:select is a bug.  We won't remove it in
> stable-2.0 but we have removed it in master.
>
>> For me, exporting `open-process` and documenting it would be enough.
>
> Fine with me, many people have asked for this at this point.  I guess
> that's the next step for this bug.
>
>> I also like the Racket interface to processes:
>> <https://docs.racket-lang.org/reference/subprocess.html>
>> (I've mostly used the `process` procedure.)
>
> Duly noted!  The more we steal from Racket, the better Guile will be :)
>
> Andy