[bug#32121] [PATCH 1/5] base: Compile CHECKOUT in the fiber.

["Clément Lassieur" <clement@lassieur.org>]

Ludovic Courtès <ludo@gnu.org> writes:

> Clément Lassieur <clement@lassieur.org> skribis:
>
>> Ludovic Courtès <ludo@gnu.org> writes:
>>
>>> Morning!
>>>
>>> Clément Lassieur <clement@lassieur.org> skribis:
>>>
>>>> Because it may take time and thus prevent PROCESS-SPECS to run every INTERVAL
>>>> seconds.
>>>>
>>>> * src/cuirass/base.scm (process-specs): move the COMPILE invocation inside
>>>> SPAWN-FIBER's thunk.  Add log message.
>>>
>>> [...]
>>>
>>>> -               (when compile?
>>>> -                 (non-blocking (compile checkout)))
>>>> -
>>>>                 (spawn-fiber
>>>>                  (lambda ()
>>>> +                  (when compile?
>>>> +                    (log-message "compiling '~a' with commit ~s" name commit)
>>>> +                    (non-blocking (compile checkout)))
>>>
>>> I think this doesn’t bring anything compared to the existing
>>> ‘non-blocking’ call.
>>> The ‘non-blocking’ procedure evaluates its argument in a separate
>>> thread; the calling fiber then “waits” for a message from that thread,
>>> which it gets when the computation is over.  The ‘get-message’ is
>>> non-blocking though: the calling fiber is simply unscheduled until the
>>> message has arrived.
>>>
>>> Does that make sense?
>>
>> Well, no :-)
>>
>> My understanding is that non-blocking is, actually... blocking, because
>> get-message is blocking.  (It doesn't block the scheduler because it's
>> in another thread, but that's not the problem here.)
>>
>> What I wanted to fix here is the fact that if the build takes one hour,
>> we will block for one hour in the COMPILE call, and process-spec won't
>> return for one hour.  If it doesn't return for one hour, that means we
>> can't evaluate anything else for all that time.
>
> Oh, I see.  However we have to wait for compilation to complete before
> continuing anyway, no?

Yes, for continuing that specific evaluation.  But other evaluations
would happen in the meantime.

>> With my change, the one-hour call will be in the fiber, which means that
>> process-spec can return, and other evaluations can be processed.
>>
>> But this is untested (because compilation doesn't work IIRC), so I can't
>> be sure.
>
> Yeah, what about this plan: let’s forget about this patch, and let’s
> remove support for compilation altogether in a future patch.
>
> WDYT?

Agreed!