GitLab

Caleb Xu authored Dec 22, 2021 and Øyvind "pippin" Kolås committed Jan 16, 2022

Follow-up to 392e60be.

In the above commit, the opencl_dep dependency was declared but
this dependency was never added to the gegl_lib target, which
needs it. As such, the addition was a no-op.

This commit adds the opencl_dep dependency to the gegl_lib target
and keeps only the opencl_headers sources, since the gegl_sources
are already included in the gegl_lib target.

This is for scaling for display and generation of mipmaps.

The loader fallback in particular is "gegl:magick-load" which depends on
a runtime tool (`convert` from Image Magick) so gegl_has_operation()
would verify that the operation is actually available at runtime (since
previous commit). If unavailable, let's just raise a warning and not
call the fallback, which is better than calling it and raising weird
unavoidable errors.

Similar to how we hide gegl:introspect, gegl:magick-load should not be
made available when Image-Magick's convert CLI tool is not present.

This reverses cf7e328c, ctx.h is no longer part of the gegl/
directory subtree but part of operations/

Jehan authored Sep 05, 2021 and Øyvind "pippin" Kolås committed Dec 18, 2021

div() expects integer arguments, and therefore any denominator less than
1.0 gets rounded to 0, hence doing a division by 0.

Instead remainder() works on double values, and anyway we only needed
the remainder.

See gimp#6888.
Thanks to Massimo for reporting the source of the issue.

Fixing issue #229.

Simon McVittie authored Sep 09, 2021 and Øyvind "pippin" Kolås committed Dec 16, 2021

The on-disk format for GEGL buffers does not appear to have been
designed to be endianness-independent (all headers and image data are
in native endianness), so we cannot load a little-endian reference image
on a big-endian architecture.

Resolves: #289

Signed-off-by: Simon McVittie <smcv@debian.org>

Lukas Oberhuber authored Oct 30, 2021 and Øyvind "pippin" Kolås committed Dec 16, 2021

According to the meson docs, dependencies between generated headers the
ultimate users, need to be declared, as ninja will run
builds out of order.

This attempts to fix this, but without being absolutely sure the
dependency is correctly recorded. In any case, the build is working.

This changes the scope of the visibility of the symbols, moving them
closer to where they actually are used adressing issue #292.

This fixes issue #298 by avoiding the shell parsing being invoked at
all, this less brittle than any forms of escaping characters, while
retaining the ability to address all existing files.