Re: [boost] Re: [dynamic_bitset] CW8 regression, a plea for help!

From: Rene Rivera (grafik.list_at_[hidden])
Date: 2004-08-06 00:08:52

Gennadiy Rozental wrote:

>>I'm asking for help in tracking this down. I'm not familiar at all with
>>the dynamic_bitset code so some direction as to likely causes of the
>>problems, or where to start looking are much appreciated. For now I'll
>>just be single stepping through the code, oh fun :-(
> 
> Just small note: why do you use BOOST_CHECK instead of BOOST_CHECK_EQUAL?
> It may clarify some of the errors

I think Gennaro uses it because it's simpler, i.e. he's using 
test/minimal.hpp.

Changing it doesn't help anyway... I've narrowed the error down to a 
curious situation:

When the block type is "unsigned long long" only this sequence passes:

     Bitset b(lhs);
     BOOST_CHECK(b.size() == 0);
     b.resize(2);
     b.reset(1);
     b.set(0);
     BOOST_CHECK(b.size() == 2);
     BOOST_CHECK(b.test(0) == true);
     BOOST_CHECK(b.test(1) == false);

But this one fails:

     Bitset b(lhs);
     BOOST_CHECK(b.size() == 0);
     b.resize(2);
     b.set(0);
     b.reset(1); //NOTE, the different order!
     BOOST_CHECK(b.size() == 2);
     BOOST_CHECK(b.test(0) == true); //FAILS!
     BOOST_CHECK(b.test(1) == false);

WHich means the reset() somehow generates bad code when optimized. I've 
tried manipulating the reset() in various ways to see if anything makes 
a difference, to the point of complete manual inlining, and nothing 
makes a difference.

Clues anyone??

PS. For those that don't know, I didn't until I looked, dyn_bitset 
stores the bits in a std::vector<block_type>, and does the usual bit 
masking manipulations.

-- 
-- Grafik - Don't Assume Anything
-- Redshift Software, Inc. - http://redshift-software.com
-- rrivera/acm.org - grafik/redshift-software.com - 102708583/icq