Re: [Boost-bugs] [Boost C++ Libraries] #6324: Flyweight performance improvement

Subject: Re: [Boost-bugs] [Boost C++ Libraries] #6324: Flyweight performance improvement
From: Boost C++ Libraries (noreply_at_[hidden])
Date: 2012-02-01 06:23:31

#6324: Flyweight performance improvement
-------------------------------------------------------+--------------------
  Reporter:  Koh Ohnishi <k_onishi@…>                  |       Owner:  joaquin                                           
      Type:  Bugs                                      |      Status:  new                                               
 Milestone:  To Be Determined                          |   Component:  multi_index                                       
   Version:  Boost 1.48.0                              |    Severity:  Problem                                           
Resolution:                                            |    Keywords:  flyweight, multi_index, hashed_unique, performance
-------------------------------------------------------+--------------------

Comment (by k_onishi@…):

 Thanks for your comment.

 I understood the design policy of Boost.Flyweight.
 But I think you can avoid exception by catch clause in erase method of the
 factory.
 Once you catch the exception, then erasing by iterator.

 I had adapted set_factory and improve performance by using erase(key) in
 the erase of the factory before posting this ticket. To solve whole
 problem completely, I modified multi_index/hashed_index.hpp as well as
 using set_factory of flyweight.

 I think these problems always stem from iterator manipulation of
 multi_index (and unordered). Does anyone have any idea of work around in
 multi_index.hashed_index performance issues?

 My problem is, or I am worried about, the bad reputation of boost library.
 Some commercial projects are using boost::flyweight and
 boost::multi_index::hashed_unique expecting O(1) inserting, erasing, and
 finding entry, but we had faced on the performance problem of erasing both
 flyweight and multi_index at late test phase.

 Users cannot imagine erasing of hashed container is time-consuming.
 Documentation (tell us which method has performance issue) or work around
 or customization (like set_factory) may help my problems.

 Thanks,

-- 
Ticket URL: <https://svn.boost.org/trac/boost/ticket/6324#comment:2>
Boost C++ Libraries <http://www.boost.org/>
Boost provides free peer-reviewed portable C++ source libraries.