Ludovic Courtès <ludo@gnu.org> writes:

>>>> +++ b/guix/store.scm
>>>> @@ -460,7 +460,9 @@ (define (open-unix-domain-socket file)
>>>>  '&store-connection-error' upon error."
>>>>    (let ((s (with-fluids ((%default-port-encoding #f))
>>>>               ;; This trick allows use of the `scm_c_read' optimization.
>>>> -             (socket PF_UNIX (logior SOCK_STREAM SOCK_CLOEXEC) 0)))
>>>> +             (socket PF_UNIX
>>>> +                     (logior SOCK_STREAM SOCK_CLOEXEC SOCK_NONBLOCK)
>>>> +                     0)))
>>>
>>> We cannot do this here because callers have to be prepared to deal with
>>> non-blocking sockets, and that’s not the case in Guix itself.
>>
>> I can see potential problems for programs outside of Guix which use
>> suspendable ports, but given Guix doesn't use suspendable ports, this
>> won't change behaviour, right?
>>
>> Obviously Guile will be working a bit differently, using poll when it
>> needs to wait for I/O, but at the scheme level within Guix, things
>> should be no different.
>
> Hmm yes, I think you’re right.
>
> One issue is if we hand over the file descriptor to something that’s not
> Guile.  Off the top of my head, this happens with inferiors and in the
> ‘build’ procedure of ‘build-self.scm’ (well, the process that receives
> that file descriptor is Guile, but if it’s older than 3.0 (?), then it
> may behave differently.)
>
> So it should be safe indeed, but adds a bit of overhead (hopping via
> ‘current-{read,write}-waiter’) and needs good testing.
>
> Laziness gives an incentive for the status quo, but I’m not opposed to
> the change if we get more confidence (test suite passing, tests with
> inferiors and ‘time-machine’, and some more auditing.)

Maybe we can just move the with-store/non-blocking in to Guix, as that
will solve the immediate issue.

I've sent a new patch for that.