ports/135262 (MAKE_JOBS): this was done wrongly

Peter Much pmc at citylink.dinoex.sub.org
Tue Aug 11 08:43:44 UTC 2009


Well, to put it short: the number of simultaneously running tasks 
is now the SQUARE of installed CPUs. :/

Shouldn't do any harm as long as you don't run out of paging
space...

The more severe issue is, parallel build does not always work. :(

Lets have a look at it: 
That build command offers TWO options "-P". One of them decides how
many "Modules" are built in parallel, and the other decides how many
Processes are run simultaneously (seemingly *within* one "Module's" 
build).

With 3.0.1 I noticed that when parallelizing Processes, the build
would fail, but when parallelizing Modules, it did work. (I didn't
investigate further.)

With the Makefile as it is now, the build does fail again. This time
i saved the logfile, and it shows:

   >dmake:  Error: -- `../unxfbsdi.pro/slb/security.lib' not found, and 
	can't be made

And then, a couple of lines *LATER*(!): 

   >echo unxfbsdi.pro/slo/permissions.o unxfbsdi.pro/slo/access_controller.o 
	unxfbsdi.pro/slo/file_policy.o | xargs -n1 > 
	../../unxfbsdi.pro/slb/security.lib
   >making .dpslo
   >Making: ../../unxfbsdi.pro/slb/security.lib

This is NOT reproducible and did NOT happen at the next run.
It is quite likely a timing issue, dependent on machine load or
whatever.
(At that next time the sequence of actions in the log was appropriate.)

So I cannot say if parallelizing Modules would be more safe than
parallelizing processes, or vice versa.
I suggest people should experiment with this. 
Fact is that the parallelization is not very failsafe.


More information about the freebsd-openoffice mailing list