Re: log2 replacement

[Jaroslav Hajek]

On Tue, Apr 29, 2008 at 8:19 PM, John W. Eaton <address@hidden> wrote:
> On 25-Apr-2008, Jaroslav Hajek wrote:
>
>  | hello,
>  |
>  | please consider the attached changeset.
>  | The current log2 behaviour with two outputs (`[f, e] = log2(x)') is
>  | inconsistent with Matlab in a number of cases, (x = Inf, x = NaN) or
>  | just plain wrong (x = realmax). Also, on most of the systems the
>  | 2-output version can be mapped directly to frexp (a C89 function),
>  | which is probably what Matlab also does. I have implemented the
>  | single-output log2 as a mapper for octave_value (aka log, log10 etc),
>  | and the two-output version as an overload in liboctave/lo-mappers.cc
>  | with a replacement if frexp is not detected. The DEFUN is in
>  | DLD-FUNCTIONS rather than mappers.cc directly, because it is not a
>  | simple mapper and thus did not seem to fit "aesthetically" to
>  | mappers.cc (but that can be easily changed).
>  |
>  | Note that the 2-output version is also extended beyond matlab
>  | behaviour for complex input - while matlab ignores the imaginary part
>  | (with a warning), this new version gives the result with f complex,
>  | satisfying 1/2 <= abs(f) < 1, e integer - because it makes good sense
>  | also for complex numbers.
>
>  With this patch, I see the following error with make check:
>
>   src/DLD-FUNCTIONS/hex2num.cc ........................... PASS    2/2
>   src/DLD-FUNCTIONS/log2.cc .............................. PASS    0/2    
> FAIL 2
>   src/DLD-FUNCTIONS/lookup.cc ............................ PASS   16/16
>   src/DLD-FUNCTIONS/lsode.cc ............................. PASS    5/5
>   src/DLD-FUNCTIONS/luinc.cc ............................. PASS    2/2
>   src/DLD-FUNCTIONS/matrix_type.cc ....................... PASS   51/51
>   src/DLD-FUNCTIONS/max.cc ............................... PASS   12/12
>   src/DLD-FUNCTIONS/qr.cc ................................ PASS   21/21
>   src/DLD-FUNCTIONS/quad.cc .............................. PASS    6/6
>   src/DLD-FUNCTIONS/rand.cc .............................. PASS   57/57
>   src/DLD-FUNCTIONS/regexp.cc ............................ PASS   81/81
>   src/DLD-FUNCTIONS/time.cc .............................. PASS   13/13
>   src/DLD-FUNCTIONS/tsearch.cc ........................... PASS    6/6
>   src/data.cc ............................................ PASS   98/98
>   src/mappers.cc ......................................... PASS    7/7
>   src/ov-fcn-handle.cc ...................................panic: Segmentation 
> fault -- stopping myself...
>  make[2]: *** [check] Segmentation fault
>  make[2]: Leaving directory `/scratch/jwe/build/octave-tmp/test'
>  make[1]: *** [check] Error 2
>  make[1]: Leaving directory `/scratch/jwe/build/octave-tmp'
>  make: *** [check] Error 2
>
>  Does it work properly for you?
>
>  jwe
>

OK, this one seems to work. There was a type in the test code. Dunno
what caused the fcn-handle failure; I've decided to put log2 into
src/data.cc anyway (some similar math functions are also there, like
atan2, hypot etc.), and it disappeared.

regards,
-- 
RNDr. Jaroslav Hajek
computing expert
Aeronautical Research and Test Institute (VZLU)
Prague, Czech Republic
url: www.highegg.matfyz.cz

log2.diff
Description: Binary data