Re: TR1 Math Special Functions

[user at domain dot invalid]

All,

When I looked at the words in my reply "I was waiting for a bootstrap 
regtest on powerpc maxosx but I'm sure it will work." I knew I would 
regret it ;-).

I also knew from the git-go that there would be different tolerances and 
sensitivities for different platforms.  I knew that we would be going 
back and forth a bit looking for a common set of test tolerances and so on.

I tested i686-cygwin and got the same error as i686-linux.  I was able 
to make all but the laguerre function failures go away with a modest and 
acceptable relaxation of the test tolerances.  The Laguerre function 
fails may be a tiny bit deeper and have something to do with comparisons 
involving unsigned ints and _Tpa vs _Tp.  I may have a fix but I am away 
from all of my usual test and development platforms.

In order to do this work, I wrote myself some tools that compare these 
functions against GSL.  I use gsl-1.9 which just came out a week or two 
ago.  One tool essentially writes all the test cases with tolerances 
derived from gsl diffs.  The other one writes out some test diffs to a 
bunch of files.  It might be a good idea to send these auxilliary things 
out so people can compile and run them.  Then we could either put the 
results up on the GCC wiki or send them to the list.  Then I could look 
through them and find at a good set of test tolerances that would be 
"fair but challenging".

Once I am back together I'll send these to the list.  They may be a 
little fugly but I think they'll help.

Ed