[GRASS-dev] Re: [GRASS-SVN] r34353 - in grass/trunk/imagery: i.fft i.ifft

[Glynn Clements]

Hamish wrote:

> Markus wrote:
> ....
> > > New Revision: 34353
> ....
> > backport this?

No.

> also r34350 and r34351 look very bugfix-like.
> 
> those too?

Maybe and yes.

r34350 adds a new color table transformation, G_abs_log_colors(),
which is like G_log_colors() but works with signed data (it maps x to
log(fabs(x)+1), with the +1 to prevent the range [-1,+1] from
dominating the colour table).

The change to the G_make_grey_scale_fp_colors() is purely cosmetic. It
uses DCELL for the min/max arguments for consistency with the other
G_make_*_fp_colors() functions, but DCELL is always equivalent to
double.

r34351 is an important bug fix. In the FFTW 3.x case, the order of the
rows and cols was the wrong way around.

This wasn't immediately obvious because i.fft forced the dimensions to
be powers of two (the original fft() implementation required this), so
a typical test case would result in either a square array or at worst
a 2:1 ratio.

[Think about what will happen if you read a 256x256 image as if it
were 512x128: it will be wrong, but it won't look like noise in the
way that a highly irregular ratio would.]

When I removed the power-of-two restriction, the bug became clear.

-- 
Glynn Clements <glynn at gclements.plus.com>