[Libreoffice] [Libreoffice-commits] .: solenv/gbuild

[Norbert Thiebaud]

On Wed, Nov 30, 2011 at 3:45 PM, Francois Tigeot <ftigeot at wolfpond.org> wrote:
> On Wed, Nov 30, 2011 at 05:21:39PM +0100, Bjoern Michaelsen wrote:
>> On Wed, Nov 30, 2011 at 04:34:23PM +0100, Lubos Lunak wrote:
>> > On Wednesday 30 of November 2011, Bjoern Michaelsen wrote:
>> > >
>> > > New commits:
>> > > commit 1f538764acdd67c3f294cd03da006ace27f736ef
>> > > Author: Bjoern Michaelsen <bjoern.michaelsen at canonical.com>
>> > > Date:   Wed Nov 30 14:02:08 2011 +0100
>> > >
>> > >     make build.pl verbose on non-silent builds
>>
>> > Moreover I question the usefullness of having verbose build by default.
>>
>> Please do not go down that road. The default behaviour is the one of GNU make.
>> Its well documented and if you want make to default to make -s you are free to
>> make yourself an alias in bash for it. In addition verbose builds are a good
>> default esp. for builders/bots and newcomers.
>
> My builds kept failing today; without this commit I wouldn't have been able
> to find the exact sequence of commands leading to the crashs and the root bug
> would not have been fixed.
>

But that also mean 5 time bigger compressed log file to upload in case
of tinderbox failure...
for me, with 2 box doing 30-40 build a piece, each a day, that means
close to an extra 100 MB of upload on a bad day...

anyway... I fixed tinbuild2 to do make -s by default...
will push that soon

> None of the verbose build options documented in the wiki were working. Having
> a clean non-verbose terminal output may be a pretty thing but if it leads to
> lost hours of work, it's definitively _not_ an improvement.

It is not just a 'pretty' thing. it is also a chance to see warnings...
with a clean output they stick out clearly... otherwise you'd be hard
press to see them in that soup of 'command lines' echo.
It is also a performance issue. first all that scrolling in a terminal
consume quite a bit of CPU... when you have a nicely tune build that
is CPU-bound, any cpu overhead translate in elapse.

Linux full 'verbose build' (all tests after a make clean and a 100% ccache hot)

real	15m40.877s
user	37m0.570s
sys	10m9.090s

Linux full make -s

real	14m48.117
user	37m8.437
sys	9m52.095

norbert