Hello Marcus,
Marcus Lindblom wrote:
> Carsten Neumann wrote:
>> I forgot to mention in the commit message that this makes the
>> SharePtrGraphOp fairly slow (at also seems to share a *lot* more [1]),
>> do we want to remove it from the default graph ops ?
>>
> I think it could be useful in those cases where you have loops.
>
> Perhaps softening the perf-hit by:
> * Making the tracking optional?
hm, maybe, but since I think the default should be enabled tracking, I
wonder if it is worth the effort, I mean who is going to make use of
that option?
> * using a hash/unordered set instead of a the log n tree-set? (Or a big
> bitset of all fcids?)
I tried that using gcc 4.1.2 <tr1/unordered_set>, but it did not make a
significant difference :(
> * replacing count() == 0 with find() == end() ? The latter ought to do
> less work (if it matters)?
I looked at the sources for gnu libstdc++ and there set::count(x) is
find(x) == end() ? 0 : 1, hopefully other implementations do st. similar.
Cheers,
Carsten
------------------------------------------------------------------------------
This SF.net email is sponsored by:
SourcForge Community
SourceForge wants to tell your story.
http://p.sf.net/sfu/sf-spreadtheword
_______________________________________________
Opensg-core mailing list
[email protected]
https://lists.sourceforge.net/lists/listinfo/opensg-core
