On Wed, Sep 12, 2012 at 10:00 AM, Alex Schuster <wo...@wonkology.org> wrote:

> Michael Mol writes:
>
>  On Wed, Sep 12, 2012 at 9:33 AM, Neil Bothwick <n...@digimed.co.uk
>> <mailto:n...@digimed.co.uk>> wrote:
>>
>
>      Instead we get, try USE="-*" :P
>>
>> "Try MAKEOPTS='-j1'"
>>
>
> Which in fact often helps... especially for me, I am using MAKEOPTS="-j
> --load=4", and I often experience build problems that are not reproducible
> with a fixed number of jobs, regardless how large.
>

This can be done on a per-package basis, yet it's inevitably implied I
should do it in /etc/make.conf instead.

And, really, it's not hard (for me, at least) to repro and/or track down
the source of parallel-induced build failures; they tend to have a pretty
clear signature in the build log. If it's a parallel-induced build failure,
I can look at the log and tell you which target should have had a different
target as a dependency, but didn't. Last time it happened to me (months
ago!), everything I needed was in the last twenty lines of build output.

I only wish someone could pay me to do this stuff full-time, because I
enjoy it. :)


>
>  "Turn off distcc"
>>
>
> "revdep-rebuild"
>
> And "emerge -e world && perl-cleaner --all && python-updater &&
> lafilefixer --justfixit".
>

Indeed. And in theory, portage 2.2 (yeah, yeah) should make most of that
unnecessary. And I think the lafilefixer portion is now a default-enabled
feature in portage. Not absolutely sure, though.

(And I think you meant perl-cleaner --reallyall :P )

You should take a look at my gentoo install script and see how many of
those tidy-up-and-rebuild commands it runs...and then a pair of "emerge -e
@world" at the end to ensure any build-time dependent creeping changes work
their way through the entire system. (And that's probably not enough, if
you're looking for an absolute setup...it'd probably be necessary to
compare checksums on files between runs and keep building until the setup
converges...but that'd require being able to ignore anything that includes
build-time timestamps.)

-- 
:wq

Reply via email to