Re: [boost] Trouble with statistic_tests.cpp and gcc-3.2/3.3

From: Paolo Carlini (pcarlini_at_[hidden])
Date: 2002-10-31 08:44:08

Gabriel Dos Reis wrote:

>My point is that numeric_limits<> is correctly set for integer types;
>so if you're seeing the same problem there, then it ought to be
>something different.
>
I see.
Therefore, in other terms, Nathan's analysis (see audit trail) could be 
*incorrect*.

>As for floating point type, GCC didn't provide fulll support until
>3.3.  So, I'm having hard time to see where you're driving at.
>Before relying on the value of numeric_limits<double>::infinity, you
>ought to check for the value of numeric_limits<double>::has_infinity.
>
Well, before these messages of yours, we agreed (see audit trail) that 
the problem had
to do with numeric_limits<double>::infinity. Indeed, gcc-3.3 (improved 
in this area)
performed much better and also 2.95.x, which has a work around, did so: 
this seemed
to confirm Nathan's analysis.

Now, what should we do? Do you believe boost/random or its configury 
machinery are
to blame or libstdc++-v3 as shipped with gcc-3.2 and gcc-3.3?

Just in case, I'm CC-ing Nathan (Nathan, is all about libstdc++/6718 
which you triaged).

Ciao, Paolo.