Em Tue, 09 Sep 2025 12:56:17 -0600
Jonathan Corbet <corbet@xxxxxxx> escreveu:
> Mauro Carvalho Chehab <mchehab+huawei@xxxxxxxxxx> writes:
>
> > Basically, what happens is that the number of jobs can be on
> > different places:
>
> There is a lot of complexity there, and spread out between __init__(),
> run_sphinx(), and handle_pdf(). Is there any way to create a single
> figure_out_how_many_damn_jobs() and coalesce that logic there? That
> would help make that part of the system a bit more comprehensible.
I'll try to better organize it, but run_sphinx() does something
different than handle_pdf():
- run_sphinx: claims all tokens;
- handle_pdf: use future.concurrent and handle parallelism inside it.
Perhaps I can move the future.concurrent parallelism to jobserver library
to simplify the code a little bit while offering an interface somewhat similar
to run_sphinx logic. Let's see if I can find a way to do it while keeping
the code generic (*).
Will take a look on it probably on Thursday of Friday.
(*) I did one similar attempt at devel time adding a subprocess call
wrapper there, but didn't like much the solution, but this was
before the need to use futures.concurrent.
> That said, I've been unable to make this change break in my testing. I
> guess I'm not seeing a lot of impediments to applying the next version
> at this point.
Great! I'll probably be respinning the next (hopefully final) version
by the end of this week, if I don't get sidetracked with other things.
Thanks,
Mauro
