mostly complete numeric_limits<>

[Oleg Krivosheev]

Hi,

On Tue, 17 Feb 1998, Peeter Joot wrote:

> Hello,
> 
> I noticed that a limits implementation was posted to this list.  About a month
> back I did one up as well, but never posted it because some things weren't 
> implemented (and some stuff is likely incorrect).  
> 
> Some of the fields/members are probably not correct, but I have made my best 
> guesses.  Of particular concern are the NaNs: 
> 
> I really don't know what the difference between a signalling NaN and a 
> quiet NaN is.  I took a guess that a signalling nan has all the mantissa 
> bits set, and the quiet one has at least the high mantissa bit clear.  This 
> is quite possibly totally incorrect.
> 
> The float denorm_min stuff may be wrong.  I followed the example in the C++
> draft where numeric_limits<float>::has_denorm == false.
> 
> I also took the easy way out for double and long double infinity and NaNs, 
> where I have just cast the float versions.  This is not likely the best 
> way of doing things.

i was trying to make a point that numeric_limits contains
vital information about target architecture - therefore it
should be automatically generated after configure run.

We have to decide where num_limits generator will live
and what part of full header should be generated.

Also should generator itself be small C code or shell
script will be enough.

Without that the providing pieces of specialization for
char, short etc is, IMHO, a bit pointless.


regards

OK