Only check the _cplusplus version when it is defined.

If we don't cancel it before unreffing it, its last-unref might happen
on the worker thread, which can lead to a deadlock.

Workaround for <glib#1941>.
Signed-off-by: Simon McVittie <smcv@debian.org>
Resolves: #226
Bug-Debian: https://bugs.debian.org/cgi-bin/bugreport.cgi?bug=954197

gegl:bloom adds a glow around bright areas of the image.

In gegl:rgb-clip, remove the restriction that low-limit <= 0 and
high-limit >= 1, and allow arbitrary limits.  However, keep the UI
ranges as-is.

... instead of int.  The gegl:median-blur "radius" property, which
"grow-radius" is redirected to, is an int, but this was a mistake,
and it's not easy to fix rectroactively, because it's part of the
ABI.  Let's at least future-proof gegl:dropshadow.

Fallout from autotools->meson port, where we lost ability to use more
than 2GB of swap on 32 bit windows.

This might not be neccesary in the future, if meson starts defining this
for windows itself https://github.com/mesonbuild/meson/pull/1455

Fixes issue #231

In gegl:dropshadow, set the abyss policy of the gegl:median-blur
node to NONE, so that it may expand the input.

Adds the grow-shape and grow-radius options to gegl:dropshadow, allowing
the shadow to be expanded to a different thickness than the source image
before the blur is applied.  Passing a negative value to grow-radius
will contract the shadow instead of expanding it.

Using a negative radius with gegl:median-blur will now calculate the
op using the same radius, but with opposite value and alpha percentiles
(i.e. each subtracted from 100%).  This makes the op a bit easier to use
for region growing/shrinking purposes.

GLib 2.64 enforces the rules that were previously only documented for
property and parameter names, resulting in this property not being
looked up successfully, leaving uninitialized junk in the max_diff
variable.
Signed-off-by: Simon McVittie <smcv@debian.org>
Resolves: #230
Bug-Debian: https://bugs.debian.org/cgi-bin/bugreport.cgi?bug=954196

In addition to last commit, also free the block when failing to set
the file offset while writing a tile, in which case nothing is
written, to avoid reading back junk data, causing a segfault with
RLE compression.

In the swap backend, when writing a tile to disk fails (usually,
when running out of space), free the corresponding block, so that
attempting to later read it back returns a NULL tile.  Previously,
we'd just read back junk, which would then segfault when attempting
to RLE-decompress the data.

Signed-off-by: Jan Vesely <jano.vesely@gmail.com>

Use ccache in babl builds.
Signed-off-by: Jan Vesely <jano.vesely@gmail.com>

Signed-off-by: Jan Vesely <jano.vesely@gmail.com>

Do not duplicate full config line.
Signed-off-by: Jan Vesely <jano.vesely@gmail.com>

Signed-off-by: Jan Vesely <jano.vesely@gmail.com>

Expose needed structure, fix typos.
Add gegl_get_timestamp implementation
Signed-off-by: Jan Vesely <jano.vesely@gmail.com>

... which checks if a pair of buffers share the same underlying
tile storage.  This can be used, e.g., for certain optimizations,
as will be done in GIMP.

To make it clearer what is going on when this is failing, making bug
like in issue #225 triage easier.