Re: [Gegl-developer] gegl:convert-space

[Øyvind Kolås <pippin gimp org>]

On Sun, Nov 29, 2020 at 2:43 PM JonnyRobbie via gegl-developer-list
<gegl-developer-list gnome org> wrote:
> I still feel like I haven't gotten the hang on it.

Your expectations of gegl:convert-space 's behavior is mostly right,
it was the code that wasn't ready but a rather good
draft had been written and presumed to work when it compiled without
manual testing. Please pull and try again, for
a bonus you can also pull in babl, for fixes to CMYK profile handling,
with both babl and GEGL from master gegl:convert-space
with an ICC profile is now sufficient for converting an RGB to that
CMYK profile, or similarly using a given RGB profile
for CMYK inputs.

> First of all - if I don't specify bitdepth and fp in the gegl:tiff-save, gegl seems to convert the image to 
> 32b float.

This is expected behavior, the specified output formats for
gegl:convert-space was "RGBA float", and now is "RGBA float",
"YA float" or "CMYKA float" depending on the type of color space. A
GEGL graph doesn't have a global
concept of u8 vs u16 vs half vs float, so just like gaussian-blur we
need to increase precision from - for example u8 - to
avoid losing some fidelity that is lost in any u8 <-> u8 color
conversion. There are a few ops, like gegl:crop, that do
not tend to "upgrade" the precision like this. In GIMP results gets
merged back down to the documents precision when
changes are stored in layers.

gegl:convert space is also not clipping the resulting colors to
0.0-1.0 range following it with an gegl:rgb-clip would do so though.

/pippin