Reduce number of undo steps created by Spryrogimp previews.
First, by grouping them into a single undo.
The implementation adds a function for clearing the
incremental preview drawing that is being executed as an idle task,
and closing an undo group in case the preview left it open (which
would usually be the case). A second way to reduce undos is
by generating a new preview only upon mouse button release of
scales in the UI, and not while dragging the scale.
AVIF import and export (need libheif >= 1.8.0)
High bit (10/12) import and export
Auto pre-select suggested output bit depth in save_dialog
NCLX color profile import (link with lcms)
Metadata import (link with gexiv2)
Plugin will not return GIMP_PDB_SUCCESS in case of failure to import file.
Our currently registered JPEG magics was using possible fields at the
start of the file (JFIF, Exif) in 7th byte. A JPEG provided in #5584 was
starting with yet another field (apparently some Photoshop-specific
field "Photoshop 3.0.8BIM") thus magics detection failed (and since
extension was wrong, the file was sent to another plug-in, hence loading
failed).
Looking further, it looks like the 3 first bytes are actually always
common to all JPEG and are the right magics/signature for a JPEG. These
3 bytes don't seem to be used by other formats as well according to my
research, which means we should not have false positives either. So
let's switch to this magics 0XFFD8FF at byte 0.
Starting from psp file version 4 the specification recommends to always use the
block/chunk length to determine the next part of the image. This way it is
possible to skip parts you don't know or don't care about or additions in
newer versions.
This change makes sure to always do this which fixes reading several images
which crashed the plug-in before.
Also only try to read layer data if it is a raster layer.
According to the PSP specifications "Each scanline in the image data is stored on a 4 byte boundary. "
Based on all actual images I've seen this is incorrect. Possibly it is a leftover from an older specification.
Fatal errors should not be done with g_message(), but by setting the
returned GError appropriately. The main GError must therefore be passed
through functions.
Some non-fatal warnings may still be outputted with g_message() if we
want people to be aware of them (for instance if we managed to load some
data yet there might have been data loss or errors).
Finally when we are confident that we recovered from the format error,
it is still nice to output some error on standard error (not be totally
silent about it), but probably not need to bother people with popups.
See #5357.
Also clean up C++-style comments into C-style, to follow our coding
style.
Note: I only cleaned ddsread as a continuation of commit 81a3370e1d. I
expect ddswrite to have similar issues, but I did not look at it.
File paths can contain confidential information such as usernames
and directory structures, making the previous behaviour a potential
privacy and security risk.
In file-tiff, add an option to crop the layers to the image bounds
when exporting individual layers (using GIMP_EXPORT_NEEDS_CROP
added in the previous commit), since TIFF has no concept of global
image bounds otherwise. Cropping is enabled by default.
When exporting a TIFF file without merging the image, make sure non
of the exported layers has a negative offset, by offsetting all the
layers as necessary. TIFF doesn't support negative page offsets,
giving an error in libtiff.
When loading a TIFF file with signed-int samples, convert the
samples to unsigned-int by offsetting them to the unsigned range,
instead of misinterpreting them as unsigned values.
Implement the GUI using prop widgets and remove the glade file.
Change loading to attach the right parasite so the save procedure
finds the fake "last values".
... since GEGLification of the twain plug-in (2.10.14 and later
versions)
In the data-transfer functions, allocate a temporary buffer for the
converted data on each call using the current chunk size, instead
of reusing the buffer allocated on the first chunk. This allows
for different chunks sizes across calls.
Just "Console" is enough in the context of menus with submenu hierarchy,
but when accessing directly the feature (with Action Search for
instance), a more accurate name is nicer. It avoids to have to check
what is what in the tooltip text.
Color picking on a single layer still works as it used to. On multiple
layer, it will now pick on the composited color, similarly to sample
merged if only selected layers were made visible.
The PDB/libgimp function gimp_image_pick_color() is also updated to work
on multiple drawables too, giving the same ability to plug-ins (the only
call to this function in core plug-ins have been updated).
When several layers are selected, select their render, similar to how
"edit-copy-visible" would have copied an image with only these layers
made visible.
Also apply the same logics to PDB function gimp_edit_copy() which can
now be used on several drawables at once.
This commit just changes our saving API (i.e. the GimpSaveProcedure
class) to take an array of drawables as argument instead of a single
drawable.
It actually doesn't matter much for exporting as the whole API seems
more or less bogus there and all formats plug-ins mostly care only
whether they will merge/flatten all visible layers (the selected ones
don't really matter) or if the format supports layers of some sort. It
may be worth later strengthening a bit this whole logics, and maybe
allow partial exports for instance.
As for saving, it was not even looking at the passed GimpDrawable either
and was simply re-querying the active layer anyway.
Note that I don't implement the multi-selection saving in XCF yet in
this commit. I only updated the API. The reason is that the current
commit won't be backportable to gimp-2-10 because it is an API break. On
the other hand, the code to save multi-selection can still be backported
even though the save() API will only pass a single drawable (as I said
anyway, this argument was mostly bogus until now, hence it doesn't
matter much for 2.10 logics).
This gives a big cleanup in the meson.build files of the plug-ins.
It's also quite a bit more maintainable, since anything that changes in
libgimp's dependencies, linkage, ... doesn't have to be copy-pasted into
each plug-in.
Add visual tab to spyrogimp plugin for a more intuitive, visual
way of specifying the spyrograph pattern.
In addition, fix using the selection as the fixed gear, and add
option to save the pattern to a path.
The GIR parser is giving warnings because both e.g. a signal, a
corresponding vfunc and a method emitting it are named
"channel_changed", which can and will give issues in some bindings.
The easiest option is to follow the general convention of starting the
signal emitters with `emit_`, which also makes clear the intention of
the method.
Change while() to go from 0..360 instead of 0..361 so the code passes
automated tests. The code never ran into this situation but the change
is a small improvement anyway.
For some reason we reverse the channel order when loading/saving a
PSD file. This doesn't seem to be necessary, and leads to a
reversed channel order across GIMP/Photoshop. Simply use the
normal channel order instead.
1 > warning: "_WIN32_WINNT" redefined
2 > pointer targets in passing argument 2 of 'send' differ in signedness
3 > passing argument 4 of 'setsockopt' from incompatible pointer type
For the signedness/type issues, I just casted to (void *) which was the
expected type for these parameter on the Linux API anyway. As for
Windows API (which was expecting char* for these various API), the
compiler just does the cast itself from void* without complaining
anymore.
On Windows, accept() wants an int for addrlen but on Linux, it wants a
socklen_t which is an unsigned int. So we can't just switch to gint as
proposed in !232 (if we do so, the signedness warning now happens on the
Linux build instead of the Windows one).
Fortunately it looks like socklen_t is actually typedef-ed to int in
Windows headers. So let's just use this type, which is much more proper
anyway as this variable is only used in functions which want this type
on Linux.
Fixes:
> warning: pointer targets in passing argument 3 of 'accept' differ in signedness
I am comparing absolute value of 2 double computation results. Might as
well just use fabs() instead of truncating both values to int.
Fixes the following:
> warning: using integer absolute value function ‘abs’ when argument is
> of floating point type ‘double’ [-Wabsolute-value]
Fixes the following bug:
> warning: absolute value function ‘abs’ given an argument of type
> ‘glong’ {aka ‘long int’} but has parameter of type ‘int’ which may
> cause truncation of value [-Wabsolute-value]
In file-psd, add support for exporting high bit-depth images. This
is currently limited to 16-bit images, since 32-bit images seem to
have a different structure (our loading code can successfully load
32-bit images exported by the plug-in, but not actual 32-bit PSD
files saved in Photoshop.) Higher bit-depth images are saved as
16-bit for now.
Note also that when saving a linear image with a built-in linear
profile the result is wrong (the image is exported with a linear-
TRC profile, but the data is perceptual), but this is a general
problem we have to fix, not restricted to the PSD plug-in.
Let's go with the logics "bottom layer is first", which is also the
logics in animated formats. Hence changing the layer order on loading
multi-page PDF.
I don't reproduce these build warnings in my build but frogonia reported
them:
> warning: 'strcpy' accessing 1 byte at offsets 0 and [0, 2147483645] may overlap 1 byte at offset 0 [-Wrestrict]
And indeed the man of strcpy() clearly states that "The strings may not
overlap" (which is clearly not guaranted in current code as we don't
know the size of what's after the substring).
strncpy() might have done the trick as it doesn't say anything about
possible overlap or not, but let's just use memmove() instead where it
is clearly written that memory areas may overlap (note that bcopy() as
proposed by ankh would have also done the trick, but it is marked as
deprecated).
… defaults to FALSE for PNG, TIFF.
The logics is that in most case, when alpha is 0, you don't expect color
to be retrieved and some people were bothered by this when erasing data
with eraser tool or selection deletion.
For the WIP image as well as for XCF format, we definitely want to keep
color data as there are workflows where you would want to erase/unerase
pixels subtly (selection also are not binary and should not be expected
to erase color data on layers with alpha channel). So we want to keep
the non-destructive behavior there.
Yet for export formats, going for "deleting what looks deleted" is a
valid approach, especially as it has been argued that experienced users
would know how to read and change the format options (and these
experienced users are likely the ones who may have valid reasons to want
pixels with full transparency to keep color values).
PNG and TIFF were the only 2 formats where I see such options, so
changing the default for these 2 plug-ins.
* Don't generate our own marshallers if they are available in GLib
already
* Don't set the c_marshaller parameter in `g_signal_new()` if it's a
default marshaller provided by GLib. See commit message of commit
39e4aa3c57 on why this is the case.
And by "supporting CR3", I mean sending the file over to raw developer
software which are supposed to support it! At the very least RawTherapee
support CR3 (since recent version 5.8 apparently), not sure about
darktable (but if not yet, I assume it won't be long either).
In the screenshot plug-in, don't clean the image, since it's not
backed by any persistent source, and disable undo while modifying
it, so that the initial edit history is clean.
gimp_int_radio_group_new() was still complaining about the scope of
radio_button_callback(). Make it (scope notified) because it needs to
stay alive after the function returns and may be called multiple times.
Also adding a GDestroyNotify to free the callback data once the widget
is destroyed (additionally it will also serve as a notifier for bindings
to properly free the callback closure itself, not only it's data).
With this last one done, GObject Introspection generation now happens
without any warning output.
… twice.
It should not be freed by the caller since it is annotated as (transfer
none). It gets freed when calling gimp_procedure_config_end_export().
Thanks to Massimo for noticing.
When the "Omit hidden layers and layers with zero opacity" option is
set, this property was only checked on non-group layers. So if we had
non-hidden layers inside hidden layer groups, they ended up exported,
which is not the expected behavior.
See https://gitlab.gnome.org/GNOME/gimp/issues/4425#note_675350
Jehan