[Buildroot] [PATCH RFC] core: enable per-package log files

[Anisse Astier <anisse@astier.eu>]

On Wed, Oct 18, 2017 at 08:57:48AM +0200, Thomas Petazzoni wrote:
> Hello,
> 
> On Wed, 18 Oct 2017 01:11:54 +0200, Arnout Vandecappelle wrote:
> 
> >  I mean that maybe the user doesn't want to gather the output with -Orecurse or
> > -Otarget but really wants -Onone. If you hardcode -Orecurse in the Makefile,
> > it's impossible for a user who does want to see output immediately.
> 
> Yes, hence my idea of BR2_PARALLEL_BUILD to preserve the existing
> behavior when it is disabled. However, the .config is not included in
> the top-level make invocation, only in the sub-make.
> 
> So, here is my plan:
> 
>  * Introduce BR2_PARALLEL_BUILD
> 
>  * Make the sub-make invocation mandatory. Right now it's only if the
>    umask is not correct or if O is not a canonical path. Let's just
>    always recurse into a sub-make, it costs essentially nothing.
> 
>  * Do a grep ^BR2_PARALLEL_BUILD=y in the config file in the top-level
>    make to decide whether we pass -Orecurse to the sub-make invocation.
>    If .config doesn't exist, then we don't pass -Orecurse because it
>    means we're about to configure Buildroot and we anyway don't care
>    about -Orecurse.
> 
> Thoughts?
> 
> I'm not sure how to handle the MESSAGE macro. One suggestion you made
> was to output directly to the tty. But a big drawback of that is that
> if you redirect the build output (make 2>&1 | tee logfile) then the
> messages displayed by the MESSAGE macro will no longer be stored in the
> logfile, which is super annoying: I always use them to navigate in the
> logfile to the appropriate place. So I believe directly writing to the
> tty in the MESSAGE macro is not a good option.

I think Arnout meant that the MESSAGE macro could write to both the tty
(if available) and stdout, hence getting the best of both worlds.

Regards,

Anisse