In GimpTileBackendPlugin, avoid storing read/written tiles in the
libgimp tile-cache, since caching is already done by GEGL.
(cherry picked from commit 47dcd4b93a)
In GimpTileBackendPlugin, change the default tile multiplier,
specifying the ratio between the backend tile-size, and GIMP's
tile-size, from 2 to 1. Since we're reading/writing each GIMP tile
using a separate command anyway, using a large multiplier doesn't
provide any benefits, while it does have drawbacks. In particular,
it reduces the chance that a write operation will affect an entire
tile, which allows us to avoid reading the tile data from GIMP.
(cherry picked from commit a5e2945b68)
Add an internal _gimp_tile_ref_noinit() function, which increases
the ref-count of a tile *without* initializing its data (in
particular, without reading its data from GIMP, or zeroing it.)
Use this function, instead of gimp_tile_ref(), when storing a tile
in GimpTileBackendPlugin, to avoid unnecessarily reading the tile
data from GIMP.
(cherry picked from commit 5ffdb9aa41)
In GimpTileBackendPlugin, return NULL when fetching z>0 tiles,
instead of simply ignoring the z coordinate, so that the mipmapped
tile is rendered locally. Likewise, avoid storing z>0 tiles.
Note that this is suboptimal, since all the necessary level-0 tiles
need to be sent to the buffer as a result. Ideally, we should
extend the wire protocol to handle mipmapped tiles.
(cherry picked from commit d0ae39f017)
... and G_TYPE_INSTANCE_GET_PRIVATE()
g_type_class_add_private() and G_TYPE_INSTANCE_GET_PRIVATE() were
deprecated in GLib 2.58. Instead, use
G_DEFINE_[ABSTRACT_]TYPE_WITH_PRIVATE(), and
G_ADD_PRIVATE[_DYNAMIC](), and the implictly-defined
foo_get_instance_private() functions, all of which are available in
the GLib versions we depend on.
This commit only covers types registered using one of the
G_DEFINE_FOO() macros (i.e., most types), but not types with a
custom registration function, of which we still have a few -- GLib
currently only provides a (non-deprecated) public API for adding a
private struct using the G_DEFINE_FOO() macros.
Note that this commit was 99% auto-generated (because I'm not
*that* crazy :), so if there are any style mismatches... we'll have
to live with them for now.
In the command handler of GimpTileBackendPlugin, forward unhandled
commands to gegl_tile_backend_command(), instead of asserting that
they're within range (which has already been disabled by commit
bc3b076caf). See GEGL commit
30047e65723ebb44fcde9c6b5f60ceecb43b0895.
(cherry picked from commit 668fee966a)
In gimp_tile_backend_plugin_command(), disable the range check for
the input tile command. This check prevents us from adding new
tile commands to GEGL without breaking the ABI; yet, the next GEGL
release will add a new command. We're going to have to decide what
to do about this, but for now, let's just disable the check, so
that at least GIMP 2.10.6 is compatible with newer versions of
GEGL, no matter how we end up handling this.
... with several threads
Commit d8ae581703 didn't go far
enough in protecting GimpTileBackendPlugin against race conditions.
The underlying GimpTile cache is shared across all drawables, so we
must use a common lock for all instances of GimpTileBackendPlugin,
instead of one per instance.
Do just that -- replace the per-instance mutex of
GimpTileBackendPlugin with a global one. This makes
GimpTileBackendPlugin instances thread-safe w.r.t. themselves, and
w.r.t other GimpTileBackendPlugin instances. However, we don't aim
to make GimpTileBackendPlugin thread-safe w.r.t. other libgimp
functions at this point, since the original API has never been
thread-safe.
GIMP was doing evil hacks lying to GEGL about it's pixels being in a linear
color space when they are not. This causes incorrect rendering, makes gaussian
blur misbehave etc.
The legacy projection modes should be implemented using the same 2.2 gamma
formats that are correct to specify for sRGB data. (for proper color
management in higher bitdepths; icc backend babl formats should be used.)
For the old image modes correct babl formats are:
R'G'B'A u8 - 8 bit RGB with 2.2 gamma (sRGB) with linear alpha component
R'G'B' u8 - 8 bit RGB with 2.2 gamma (sRGB)
Y'A u8 - 8 bit Grayscale with 2.2 gamma with linear alpha component
Y' u8 - 8 bit Grayscale with 2.2 gamma
Y u8 - 8 bit linear data, used for masks/channels
A u8 - 8 bit linear alpha
-----------------------------------------------
RGBA float - 32bit floating point linear light RGB
RaGaBaA float - 32bit floating point linear light RGB, premultiplied alpha
to be used for processing that needs to scale by the alpha,
(blurs, resampling etc)
R'G'B'A float - 32bit floating point sRGB with gamma, to be used where
the result depends on being closer to perceptual when
processing, can be used a cheaper alternative to CIE Lab
based modes.
-----------------------------------------------
The legacy layer modes should use the formats with gamma 2.2 only for loading
and rendering legacy XCF files correctly, in the brave new world compositing
should most likely be done in linear light with "RGBA float" and even better
"RaGaBaA float" like GEGL does for porter duff and other compositing modes.
The ability to chose the legacy layer modes should probably be hidden from the
user unless an old .xcf has been opened.
