This is the mail archive of the libstdc++@gcc.gnu.org mailing list for the libstdc++ project.


Index Nav: [Date Index] [Subject Index] [Author Index] [Thread Index]
Message Nav: [Date Prev] [Date Next] [Thread Prev] [Thread Next]
Other format: [Raw text]

Re: Search algorithms in __gnu_cxx::


Hey guys. I was wondering where all these messages were coming from:
any chance we could start a new thread? In a threaded mail client, this
is showing up in an Sept thread... and it's now November... ;)

> Sure, in *practice* in this specific case it's fine, because we just 
> branched. But, as a general rule, you should *always* patch mainline: if 
> you are encountering special problems, you should try to explain in 
> detail to the people reading the mailing lists or eventually to the 
> overseers.

Yes.... 

For the time being, while gcc-4.3 stage 1 gets going, I think it's ok
to relax a bit and allow people to work in gcc-4_2-branch.

As time goes on, this will become more of a pain. However, for the
moment, it's easier for everybody if the maintainers are flexible about
this if mainline is proving to be too unruly.

> > Here are the numbers. Attached is the file used.
> >
> > +-------------------+---------+
> > | algo              | time    |
> > +-------------------+---------+
> > | Boyer Moore       | 9.475s  |
> > | std::search       | 12.283s |
> > | strstr            | 13.184s |
> > | std::string::find | 20.166s |
> > +-------------------+---------+
> 
> Excellent. Well, the comparison would be easier if you had used the 
> performance testcase that *yourself* added some time ago, but anyway, we 
> can take care of that double check, no problem. Otherwise, I'm unsure 
> that we want to add the new trait to the general ext/traits file, and I 
> still believe a separate file would be better (because of the already 
> mentioned namespace issue). Otherwise, I think we are quickly converging 
> to something we can commit soon, good.

I think all non-TR1 traits should be added to ext/type_traits, FWIW.

Dhruv, I'm encouraged. This looks to be evolving well!

As Paolo noted, you can run the normal tests with

"make check"

in the libstdc++ build directory. Your posted testresults seem fine,
but if you are in doubt you can double check against some known good
results. For most platforms, results are here:

http://gcc.gnu.org/ml/gcc-testresults/current

And for you:
http://gcc.gnu.org/ml/gcc-testresults/2006-11/msg00661.html

In addition, you can run the performance testsuite yourself, by typing

"make check-peformance"

in the libstdc++ build directory.

best,
-benjamin


Index Nav: [Date Index] [Subject Index] [Author Index] [Thread Index]
Message Nav: [Date Prev] [Date Next] [Thread Prev] [Thread Next]