The previous commit was not taking into account usages of
gimp_cairo_surface_create_buffer(), other than the text layer rendering
usage, as it was focused on fixing #14970.
In particular, in other pieces of code, we were using a temporary
GeglBuffer as a way to edit the original Cairo surface but the syncing
back was not happening anymore. So previous commit was breaking all
thumbnails of layers or images, as well as PDF export for all raster
layers.
After some thought, I decided to implement a new function
gimp_cairo_surface_get_buffer() which would do what
gimp_cairo_surface_create_buffer() was doing, but adding a boolean flag
to declare whether or not we were expecting to sync back into the Cairo
surface's data. After all, when we don't need to sync back, better avoid
the useless work of copying possibly huge data.
I also added back the possibility of returning a linear-memory backed
buffer, after some discussion with pippin. Because for very small
surfaces, and if we are indeed looking to sync back into the source
Cairo surface, the advantages of using a linear buffer may still be
worth it, even though it's not true for the general case (where tiled
buffers will be much more efficient). I used 4096 (64^2 because our
current tile dimensions are 64 right now in GIMP), though this number is
kinda random and doesn't matter too much (it doesn't have to be the
square of the tile's width or height). I don't put any info about this
in the new function's documentation because developers should not have
any expectations on the type of buffer they get (only whether or not it
will sync back to the source).
As a consequence of all this, gimp_cairo_surface_create_buffer() has
been deprecated as of GIMP 3.2, though it will still work the same. The
new function will be recommended as a replacement.
Creating a gegl buffer from some data (e.g. cairo surface) through gegl_buffer_linear_new_from_data
causes crashes if the cairo surface is too big, because the gegl buffer will have only one huge tile
so tile_size will overflow if surface size > INT_MAX. Using the appropriate api, gegl will calculate
tile size and create the needed tiles automatically.
Our build files were relying 'sysroot' to find gexiv2.h but this is
not possible with Apple Clang om which sysroot points to macOS SDK.
So, exotic environments like Homebrew were failing. Let's fix this.
As a consequence, here are needed new functions:
* libgimp:
- gimp_param_spec_resource_defaults_to_context()
- gimp_param_spec_resource_none_allowed()
* libgimpbase:
- gimp_param_spec_file_get_action()
- gimp_param_spec_file_none_allowed()
- gimp_param_spec_unit_percent_allowed()
- gimp_param_spec_unit_pixel_allowed()
The code to retrieve the alpha-variant of the indexed palette was wrong,
and therefore babl was creating a brand-new palette fornat. Not sure if
the palette was empty, or was random or anything. It's even weird it
even produced somewhat resembling colors.
This was bothering me that we were keeping these macros marked as legacy
in our public headers. Now they aren't so we can delete the whole file
when we'll finally get rid of all usage later, without breaking API.
This completes the GimpRGB API removal
project for the color space invasion.
Note that the GIMP_RGB_LUMINANCE macro is
temporarily moved to gimpcolor. It does not
require GimpRGB but was included in gimprgb.h.
We have a bunch of special-casing format passing through the PDB, but
either we were only passing the encoding, or else we were reconstructing
the full format through private intermediate functions. In the
space-invasion world, this is not right. Let's have a proper "format"
type for PDB which does all the relevant data-passing for us, once and
for all!
Note that I am creating a wrapper boxed type GimpBablFormat whose only
goal is to have recognizable GValue since Babl types don't have GType-s.
Moreover I'm not using the GeglParamSpecFormat either, because it just
uses pointers which again are a bit annoying in our various PDB code.
Having a simple boxed arg is better.
GimpRGB replaced with gdouble arrays.
Note that some temporary intermediate
GimpRGBs objects were added, which will
be removed when map-object and
gimpoperationgradient are fully converted
in a separate commit.
..with gimp_bilinear_rgb ().
This function takes in a double array of raw pixels,
a boolean to determine if the pixels have an alpha channel,
and a reference to return the final pixel to.
This abstract spec type is basically a GParamSpecObject with a default
value. It will be used by various object spec with default values, such
as GimpParamSpecColor, GimpParamSpecUnit and all GimpParamSpecResource
subtypes. Also it has a duplicate() class method so that every spec type
can implement the proper way to duplicate itself.
This fixes the fact that in gimp_config_param_spec_duplicate(), all
unknown object spec types (because they are implemented in libgimp,
which is invisible to libgimpconfig) are just copied as
GParamSpecObject, hence losing default values and other parameters.
As a second enhancement, it also makes it easier to detect the object
spec types for which we have default value support in
gimp_config_reset_properties().
As a side fix, gimp_param_spec_color() now just always duplicates the
passed default color, making it hence much easier to avoid bugs when
reusing a GeglColor.
This patch removes the two instances of
gimp_rgb_luminance_uchar () and one
instance of gimp_rgb_luminance () from
the codebase.
* plug-ins/common/checkerboard.c:
Use gegl_color_get_pixel () to get the
luminance value at the same time we're
getting the RGBA value from the GeglColor.
* plug-ins/gradient-flare/gradient-flare.c:
Replace with GIMP_RGB_LUMINANCE ()
macro, which does not use GimpRGB.
gimp_cairo_set_source_rgb () and
gimp_cairo_set_source_rgba () are no longer
used in the code base, so let's remove them
as part of the move to GeglColor.
This reverts commit f7e7f396aa.
See issue #8269.
For GIMP 2.10, I'll just revert this commit so that we get back to pre-2.10.32
situation. ~~For 3.0, we'll try to do something better.~~
-> Let's delay improving the situation even for GIMP 3.0, in which case
it's better to at least keep the same render as for the 2.10 series and
avoid confusion. I'll take more time after release to study better the
situation and see what we should actually do.
(cherry picked from commit 2fd738bb3e)
- gimp_rgb_list_names() ported as gimp_color_list_names() using NUL-terminated
GimpColorArray.
- GimpColorHexEntry uses the new function (and is therefore now GimpRGB free!).
- gimp_rgb_parse_name() deleted as it's unused since my previous commit.
With all this, gimprgb-parse.c is now deleted from the repository! \o/
… test-color-parser.c file.
The file libgimpcolor/test-color-parser.c was compiled but never actually called
by the build. Now that we have a nice infrastructure to test libgimp API, I am
moving it there with the new format. Doing this also allowed me to discover some
bugs in CSS parsing, as well as discover Python binding was failing here (cf.
the few previous commits).
Only one test is disabled so far, the one where 4 digits are used per channel in
hexadecimal notation: "#64649595eded". This format simply doesn't appear
anywhere in the spec, and also the result values in the samples listing don't
even fit. So far, I'm just unsure what to do with it, if we want to keep this
support (of some kind of higher precision hex notation, but not specified, so is
it even used by anyone?) or not.
All the other tests just work in both C and Python!
New functions with the same name as these functions are created, except without
the length argument (i.e. it's equivalent to calling these with -1).
The reason for this is that using strings with a length variant which may be
negative to switch to NUL-terminated strings are not bindable. At least in our
case, when testing in Python, the input string ended up as corrupted garbage and
GObject Introspection docs warns about such interfaces:
> In particular, avoid functions taking a const char * with a signed length that
> can be set to a negative value to let the function compute the string length
> in bytes. These functions are hard to bind, and require manual overrides.
(see: https://gi.readthedocs.io/en/latest/writingbindableapis.html#strings)
So instead, I create a simple version which runs on NUL-terminated strings only
and which is bound, whereas unbinding the generic length-version (making it
C-only, or maybe usable in some other bindings which ignore the (skip)
annotation; apparently some do this).
- "transparent" is now recognized. It was forgotten (probably because on the
GimpRGB interface, we separated the API in a _rgb_ and a _rgba_ variant).
- rgba() and hsla() formats are now fixed (implementation was there but the
function names were not recognized.
- Adding some comment about limitations of the hexadecimal notation (we don't
support the alpha channel which is now in the CSS specs, while we also support
some non-specified variant with every channel on 3 or 4 digits) for future
work.
This is meant to obsolete GeglParamColor with at least an additional argument
has_alpha which we need in GIMP. It allows to advertize when a parameter wants
an opaque color, which in particular means we know when displaying a GUI to pick
colors with alpha or not.
There are no plug-ins which uses GimpRGB for procedure argument, nor is there
any base PDB procedure. We don't pass this type anymore through from/to
core/plug-ins. So let's clean the whole code out as a next step to get rid of
GimpRGB from our codebase!
With "HSVA" and a few others, the function was not returning the right value (it
could not see it was encoding alpha channel.
I had fixed it on babl directly with commit a28309c yet I had forgotten to push
it! Let's just add some #ifdef and a comment.
This is used for the gimp_color_is_perceptually_identical() function,
because the Euclidean distance in LCH is extremely limited, if not wrong
in many cases. Indeed LCH is not perfectly perceptually uniform, and for
this exact reason, the CIE defined the specific Delta E algorithms.
Later versions are also based on LCH values, so my intuition to use it
for distance was on a good start, yet these algorithms were refined a
few times to allow for corrections in perceptual uniformity
imperfections.
This was in particular needed to verify if a color is out of a CMYK
space gamut. The idea is to compare the distance of the RGB (or other)
and the CMYK version, since we cannot just check if the CMYK color is
out of the [0; 1] range (it never is). Instead if both colors are
perceptually identical, then we consider that the RGB color was inside
the CMYK space gamut.
The naive algorithm was giving any (or nearly) color as out-of-gamut.
Now using CIEDE2000, I get a much nicer results.
- New libgimpcolor functions: gimp_color_parse_hex() and gimp_color_parse_name().
- GimpColorHexEntry is now space-invaded. Though recognized color names
and hexadecimal syntax are sRGB only (because CSS and SVG
specifications explicitly say that this syntax is for sRGB values), it
is possible to enter non-sRGB values with
gimp_color_hex_entry_set_color().
- GimpColorSelection is now space-invaded.
