On 17.12.2013 22:06, Bill Deegan wrote:
Dirk,
If the memory patch uses __slots, then won't that likely break some
user logic? If so then we should push out 2.3.1 without it with a
notice in the release notes indicating what such change may break?
I've now created pull request #97, containing the patch for reducing
memory. In a test project with 8000 C files it boils down the
516.8/534.5 MBytes (clean build/update) to 434.9/452.2 MBytes. So we
save about 16%...without using slots and no changes to the already
existing interfaces.
Please do a review and judge whether the commit is good enough to
go...and don't hesitate to reject it, if your tummy tells you to do so. ;)
For the next version after the current one, I already have prepared
another patch on top. It would switch the Node classes and their build
infos to using slots. This could get us down to about 300MB in total
(-40%), if it's okay for you that we'd lose some flexibility in the
framework. Like for the current pull request (#97), trying to squeeze
out the last bits introduces more and more side-effects for us
developers. Variables, like lists and dicts, may not always be present
or need special handling in derived classes...
So, if you'd like to introduce slots to the core sources, I suggest to
warn the users in 2.3.1 about the upcoming changes. After the 2.3.1
release we could apply the "slots" patch immediately, which would give
us enough time to test it, while working towards a 2.4 version...
Regards,
Dirk
_______________________________________________
Scons-dev mailing list
[email protected]
http://two.pairlist.net/mailman/listinfo/scons-dev
