Previous OpenBlas patch fixed the crash with Sophos (see reports #3633
and #4246) but created a huge slowdown of startup because of a timeout
change and most likely OpenBLAS being loaded at startup during the
various verifications.
A new patch has been merged upstream to lower this timeout to something
more reasonable. Reporters confirmed GIMP now runs fine (neither crashes
nor very long startups).
See: https://github.com/xianyi/OpenBLAS/pull/2339
GIMP will now process the remote gimp_versions json file to look if one
is using the last version of GIMP. This initial code doesn't act up yet
on this information. This will come in further commits.
Here are the characteristics:
- Since this requires internet access, a new checkbox is available in
the Preferences dialog, allowing to disable version checks. Note that
it is enabled by default as it is an important security feature, but
it has to be deactivatable.
- The remote access is done as an async operation because we don't want
it to block the startup in any way (for whatever reason). Also it
doesn't output errors if it fails to not be a bother (you don't
technically need internet access for an image program).
- We don't check at every startup. At each successful check, we save a
timestamp to prevent too frequent useless checks (I set it the timer
to a week or more for now).
The point will be for a packager to create a unique build ID to identify
the build or provenance. I also add a revision number so that we can
identify 2 builds from the same version/commit, same maker and platform.
It will also be used later to check for new versions (see "phone home"
feature #2584).
Separating autotools and meson commits for easy backport.
Some issues may happen only for 32-bit builds. So it may be worth make a
Windows 32-bit build too.
Issue #2794 triggered this reasonning, though in this issue's case,
simply the build would not have discovered the bug since it was only a
build warning + run-time bug. Still long-term if we manage to get a
no-warning build, it could become relevant to discover more errors at
build time. So preparing the ground work here.
The GNU/Linux builds should start as soon as the Linux dependencies are
built. There is no need to wait for the Windows dependencies (and
reciprocally of course).
This should make for much faster CI total duration (with current
configuration).
Note: this "needs" keyword is quite a recent feature since gitlab 12.2,
3 months ago: https://docs.gitlab.com/ee/ci/yaml/#needs
Anyway meson is based itself on python3 so it has to be present. Just
using `python` may be any python (2 included).
Of course, it would be ok most of the time, but with the Fedora 31 CI,
apparently just `python` is not found.
8b5060349a fixes the issue of version headers
not being available when building out of VCS tarball (without .git directory)
but we were getting `unknown (unsupported)` version still. Turns out the version
silently fell back to `unknown` when git was not available.
Let's warn the user when this is the case.
Reported by Jan Tojnar as a comment to commit e4c7fc23 that builds from
simple archives of the repo contents (without the .git tree) are
currently broken. Well this is normal, as we only support builds from
release versions or from the repository where we can extract a git hash.
Any other kind of nightly build can be from any commit out of thousands
and is maintenance hell.
To be nice though, let's unbreak such builds, but they will be clearly
marked as unsupported (warning at configure time + the extract commit
hash will now be labelled "unknown (unsupported)", which will be a
visible string in About and on unstable version canvases).
Basically do so at your own risk.
Also generate INSTALL all the time (not sure why it was only generated
in non-git case).
Actual patch contributor wants confidentiality to avoid leaking
proprietary information or whatever (I am not sure either what to be
scared of as it's all good and harmless to me, but let's respect the
request). See also #4246 for more details.
The way we use CC_VERSION macro is to give information on the compiler
used during build. This information may be useful when debugging in
particular. So we can't just use `cc.version()` which only gives a
version number, not even the name of the compiler.
Restore the logics of autotools where we were using the result of `cc
-v` (for gcc and clang) and testing various CLI options for other
compilers.
Also this test may fail on meson because of various bugs, which I now
reported and provided patch for (hopefully soon merged). In particular,
when using ccache, the command run fails (also I have to change newlines
in C-style `\n` myself as meson's set_quoted() creates broken header
when newlines are present).
If it fails, let's at least store the compiler name + its version, still
more useful than version only.
My previous test (commit 41285813a5) was a bit misinformed. So it turns
out bug #4185 is for all platforms and the broken libheif versions are
1.5.0 and 1.5.1 only.
So my new test (platform independent) is: prefer libheif versions with
profile support, except 1.5.x; then prefer lower versions without
profile support; and only as last resort accept 1.5.x versions (but
output a warning).
Same as what I did for the configure script. Note that I break commits
in 2 to make the autotools commits easily backportable to gimp-2-10,
i.e. without any meson files.
Known bug in libmypaint dependency. It has been fixed in libmypaint
1.4.0, which we cannot hard require unfortunately (Debian testing still
at 1.3.0).
Still let's make add a warning so that packagers are aware of the issue
and update when possible.
It was typoed as HAVE_LIBHEIF_4_1_0 so profile support was never
working for HEIF format.
Also add warnings and better output, similar to configure script one.
Replace the "Heif >= 1.4.0" line in the summary output by a comment in
the "Heif" line explaining this is about profile support.
Also add a >= 1.6.0 test and output a warning for Windows and macOS (cf.
bug #4185).
I am not touching the translation, only the formatting for the Keywords
field in desktop file. As requested by a translation comment, all
translations of this field need to end with semicolon too (without, we
get validation warnings).
... or copied RGB channel
In gimp_drawable_merge_filter(), make sure the drawable's source
node is constructed before applying the operation. The
construction of the source node connects the drawable's filter
stack to the udnerlying source node (usually, the buffer-source
node), which we rely on when calling
gimp_gegl_apply_cached_operation(), since we pass
connect_src_buffer == FALSE. Otherwise, the operation is applied
to an empty input, instead of the drawable content.
