Re: fntests results in octave 3.1.51 on mingw No.3

[David Bateman]

Tatsuro MATSUOKA wrote:
> Hello 
> 
> This mail reports the test of data.cc on octave-3.1.51 on mingw.
> 
> Integrated test scripts:
> 
>   C:\usr\tatsu\mingwhome\octaves\octave-3.1.51\src\mappers.cc  PASS  227/230  
> FAIL 3
> 
>>>>>> processing C:\usr\tatsu\mingwhome\octaves\octave-3.1.51\src\mappers.cc
> 1. ***** assert(isna (single(NA)));
> !!!!! test failed
> assert (isna (single (NA))) failed  
> 
> 
> 
> 2.***** assert(isna(single([NaN -Inf -1 0 1 Inf NA])), [false, false, false, 
> false, false, false,
> true]);
> !!!!! test failed
> assert (isna (single ([NaN, -Inf, -1, 0, 1, Inf, NA])),[false, false, false, 
> false, false, false,
> true]) expected
>    0   0   0   0   0   0   1
> but got
>    0   0   0   0   0   0   0
> values do not match  
> 
> 3.***** assert(log (single([1, e, e^2])), single([0, 1, 2]), sqrt (eps));
> !!!!! test failed
> assert (log (single ([1, e, e ^ 2])),single ([0, 1, 2]),sqrt (eps)) expected
>    0   1   2
> but got
>    0.00000   1.00000   2.00000
> maximum absolute error 5.96046e-08 exceeds tolerance 1.49012e-08
> 
> For test 1 and 2
> 
> mingw octave 3.1.51
> octave.exe:4> single (NA)
> ans = NaN
> 
> Msvc 0ctave -3.0.1
> octave-3.0.1.exe:1> single (NA)
> ans = NA
> 
> mingw octave-3.0.1
> octave.exe:1> single (NA)
> ans = NA
> 
> Thus the erros are specific to octave-3.1.51.
> But I not not the reason why.
> 
> 
> For test 3
> 
> octave.exe:2> format long
> octave.exe:3> log (single ([1, e, e ^ 2])),single ([0, 1, 2]),sqrt (eps)
> ans =
> 
>    0.000000000000000   0.999999940395355   2.000000000000000
> 
> ans =
> 
>    0   1   2
> 
> ans =  1.49011611938477e-08
> 
> Is this tolerance not allowable?
> 
> 
> Regards
> 
> Tatsuro
> 
> --------------------------------------
> For All Sports Lovers! SPORTS OHEN PROJECT 2008
> http://pr.mail.yahoo.co.jp/yells/
> 


The patch to fix this was applied in 3.1.51+ and is

http://velveeta.che.wisc.edu/cgi-bin/hgwebdir.cgi/octave/rev/139f47cf17ab

The failing tests were a marker to point out that the single/double
precision handling of NA values was not correct.

D.