From: Vladimir Prus (ghost_at_[hidden])
Date: 2003-02-05 02:36:54
Jürgen Hunold wrote:
>>Oh... that's not nice. How many targets are there in your project?
>>Secondly, could you run bjam with "-d+10" switch, redirect the output
>>to file and send that file? (Just to make sure: "-d+10" should be the
>>first thing on Jam's command line).
> Please find the output attached.
Thanks a lot!
I notice two things:
1. Our profiling code is not prepared to such large running times,
and reports negative values of top-level callers
2. The feature.attributes rule takes 1706 060 000 time units, which
is 28 mins. It is quite in accordance with other measurements.
Okay, performance tuning bubbles up to the top of my Boost.Build priority
queue. I'll keep you informed of the progress.
> bjam says it found 40000 targets and updates 4005 targets (including
> dirs) when building from scratch. These are two projects sharing some
> basic (qt-related) libraries, build 10 different applications. Quite a
> stress test ;-) Building takes approx. 3 hours, 1 hour bjam and 2 hours
> to compile and link...
Oh... that's pretty big project. Thanks for providing such a test.
>>That's bad, but it's also interesting because it's not the place
>>where I'd expect performance problems. Looks like we're just using
>>low-level structures (list of string), which cannot be optimized. I
>>have a plan of improving that, but before starting, I'd like to look
>>at your profiling data.
> Yes, always wait for profiling data...
I believe there's enough of profiling now. I'm about to turn long list
of properties into class, which would give vast possibilities for
> Well, it produces the names of the two targets/dirs to build.
> On top-level Jamfile I declare build-project dir1.
> There I descend with build-project dir2.
> And then same message again.
>>This basically means that requirement for the project are
>>link-incompatible with the build request. Are you sure there's no
>>"<variant>debug" in project's requirements?
> yes. I've reduced all Jamfile to the minimun, I hope.
I'll take a look.
Boost-Build list run by bdawes at acm.org, david.abrahams at rcn.com, gregod at cs.rpi.edu, cpdaniel at pacbell.net, john at johnmaddock.co.uk