parallel builds revisited

[youshi10 at u.washington.edu]

On Fri, 13 Apr 2007, David Nečas (Yeti) wrote:

> On Fri, Apr 13, 2007 at 05:10:47PM +0200, Pav Lucistnik wrote:
>> Peter Pentchev píše v pá 13. 04. 2007 v 18:06 +0300:
>>> >
>>> > I was thinking about having it embedded in every port's Makefile
>>> > directly, instead. Something like
>>> >
>>> > USE_MAKE_JOBS=	2
>>>
>>> IMHO, hardcoding the number of jobs in the port's Makefile would not
>>> be the best approach.  I think a port should only flag whether it
>>> supports parallel building at all or not - and leave the number of jobs
>>> to either the ports framework or the administrator's choice.
>>
>> That was just an example. You can do
>>
>> USE_MAKE_JOBS=	yes
>>
>> for autoscaling perfectly well. For details, see the patch I linked.
>
> The patch gives no reason for such hardcoding, it just
> implements it.  How many ports exist that can fail with N+1
> jobs yet cannot break with N jobs (for N > 1)?
>
> Yeti
>
> --
> http://gwyddion.net/

My opinion is that there should be a threshold value empirically derived by the developer / retrieved by bug reports, as well as a knob, to specify the maximum number of parallel jobs to be used for a particular port, that way you don't get people accidentally specifying, say 10 jobs when it can only handle 2-3.

Doing that should decrease the amount of time people have to spend fishing through bug reports for minute information, and decrease the failure encountered by end users going all out trying to run as many jobs as possible on a given port.

-Garrett