En route to on-canvas gradient editing, add support for persistent
handle selection to GimpToolLine (a handle being either an endpoint
or a slider). Handles are selected through clicking, however,
unlike before, the selection persists after the mouse is released.
A new "selection" property specifies the currently-selected handle
(who knows, maybe in the future we'll add multi-selection), and a
new "selection-changed" signal is emitted when the selection changes.
The visual feedback has been changed to better suit the new behavior,
and the behaviors yet to be added: The selected handle is marked
using highlighting; the highlighting doesn't change while hovering
over other handles. Only the hit-test circle is used as hover
indication, however, we use a fixed-size circle, and only show the
circle for the currently hovered-over handle -- no more trippy
expanding circles :)
A few minor changes along the way:
- The selected handle is now the (first) one that's closest to the
cursor, instead of the first one to pass hit-testing.
- We don't move the selectd handle upon button-press, only upon
motion, so that handles can be selected without moving them.
- Show a MOVE cursor modifier when hovering over a handle.
otherwise, implicit transitive linking will pull in the installed
libs, not the ones from the source tree, and the build can fail when
any of the libs' APIs changes.
Also remove some useless #includes.
Tweak the layer group saving code so that the saved PSDs match
Photoshop-produced PSDs a bit more closely. For the most part, it
doesn't seem to matter much, but it does somewhat improve
compatibility with other programs that read PSDs.
... and ignore language setting (e.g. en_US).
The problem came from the fact that these settings names are class
properties of GimpPaintOptions/GimpContext which is first instanciated
when the Gimp object is created. This unfortunately happened before
language_init() since we needed these objects when loading gimprc
(making inversion of calls rather complicated).
Therefore they were localized with the system language, not the
configured language.
The solution was to create a very simple object GimpLangRC which
implements the GimpConfig interface, for sole purpose to read the
language from `gimprc` in a first pass. gimp_load_config() will still
happen later as a second pass to properly load the rest of the
configuration.
I had an error in the previous commit (2 args in 1). Also adding access
so that the file `bookmarks` is visible from the contained GIMP
(otherwise bookmarked folders are lost in flatpak and that's bad
experience).
For some reason, the "cleanup" tag won't delete the files produced by
the BaseApp inside the main GIMP manifest (I still need them for build,
but want to delete them for runtime).
This is a workaround to delete them with a few command lines for now.
See: https://github.com/flatpak/flatpak/issues/1082
... off-by one size in special cases
The last commit wasn't drastic enough. We changed SIGNED_ROUND()
to use RINT(), which, in turn, may use rint(). However, rint()
effectively breaks ties to even, so that we get stuff like
'rint (1.5) - rint (0.5) == 2.0 - 0.0 == 2.0'. This can't be
good--it's entirely possible that we're bitten by this in other
cases without noticing.
Avoid rint() entirely in RINT(), and always use 'floor (x) + 0.5'
instead, which always breaks ties up. Hopefully, this doesn't
break anything else...
... off-by one size in special cases
SIGNED_ROUND(), which is used by GimpToolRectangle, among other
things, used to round negative values which lie exactly between
two integers, i.e., -foo.5, down. This could lead to the rectangle
being one pixel bigger than expected, in either dimension, when one
of its edges had a negative coordinate, and the opposite edge had a
positive coordinate.
Fix SIGNED_ROUND() to always round such values up, regardless of
sign.
Keeping all dependencies inside the main manifest is very annoying
because flatpak-builder will check them every time the package is
rebuilt. Worse, sometimes the cache won't be hit (even though it should
have), resulting into a rebuild of many dependencies. I create a BaseApp
build which is the recommended process (and not creating our own runtime
based on GNOME one's, as I first thought) which won't need to be built
as often as the main manifests. The other advantage is obviously that
this BaseApp can be shared between the dev and nightly (and likely even
the stable later) builds. I will only keep differences inside the main
manifests (for instance lcms2 which requires a higher version on master
than on the GNOME runtime and the last dev release).
I also move webkitgtk as the first dependencies since it takes too long
and flatpak uses a sequential dependency graph (so any change to a
previously listed dependency, even when actually unrelated, was
triggering a rebuild of webkitgtk!).
Only remaining issue is that I don't manage yet to run the cleanup of
the BaseApp at the end of the main manifests (for files needed for
building, but not at runtime).
So I discover today that there is an option --ccache to request
flatpak-builder to compile using ccache, which is obviously a great idea
when rebuilding the same deps too often. Update the howto with the info.
While I am at it, let's spread the improvement to options_box which was
also a weak pointer with g_object_add_weak_pointer(). Let's make it
rather a GWeakRef for the same reason as I did options_gui.
Other than multi-threading (which here is not the problem), using
GWeakRef has the other advantage that it makes the type of pointer
obvious, hence avoiding the kind of errors as fixed in commit 12df796.
One can't just change the pointer value directly, and has to use
g_weak_ref_set(), so such problem won't happen again.
Repalce the two separate size entries, used for the position and
size properties of GimpRectangleOptions, with a single size entry
with two fields, so that they accept ratio expressions. Note that
this doesn't change the UI.
When a size entry has exactly two fields, enable ratio expressions
in eevl. Set the reference value to the value of the field that is
not currently being evaluated, and invert the ratio when evaluating
the second field.
Ratio expressions have the form 'x : y' (the ':' operator has the
highest precedence for a binary operator, and is left-associative).
Given a reference value 'a', the expression evaluates to
'a * (x / y)'.
Ratio expressions can be controlled by the caller by:
- Enabling or disabling them: They're meant to be used when the
eevl servers two paired entries, and can be disabled otherwise.
- Setting the reference value: That's normally the value of the
"other" entry of the pair--the one not currently being
evaluated.
- Inverting the ratios: Normally, one entry refers to the
antecedent term of the ratio, and the other entry refers to the
consequent term of the ratio. When evaluating the latter one,
the ratio should be inverted.
Pass the evaluation options to gimp_eevl_evaluate() using a single
parameter of type GimpEevlOptions, instead of using individual
parameters for each option. Add a GIMP_EEVL_OPTIONS_INIT macro,
used to initialize a GimpEevlOptions struct to the default set of
options. This would allow us to add evaluation options more
easily.
When loading tile data, avoid copying the data into the GEGL
buffer when the tile is empty (i.e., all its bytes are 0), so that
GEGL doesn't allocate memory for it unnecessarily.
Better factorize by reusing code rather than recreating a combo box
which does basically the same thing. I only added a boolean parameter to
retrieve only the sublist of manual language.
It also takes advantage of the self-translated language names from
initialization.
In gimp_canvas_sample_point_get_extents(), use the drawn number's
actual size instead of some random constant that was good enough for
my own font size.
Without a domain error, glib outputs on console:
> g_error_new_valist: runtime check failed: (domain != 0)
Let's just create a domain error for the file-pdf-load plugin.
