2001-04-20 Jon Trowbridge <trow@ximian.com>
* gal/e-text/e-completion-test.c: Limit total matches, for better
performance on slow machines. It is supposed to be a test, so
correctness of the completion operations isn't really a
priority...
* gal/e-text/e-completion-view.c (e_completion_view_construct):
Set GTK_CAN_FOCUS flag.
* gal/e-text/e-entry.c (e_entry_show_popup): Evil! Evil! Unclean!
Unclean! Manually check if the pointer is in the area where the
popup is going to appear, and if it is, warp the pointer out of
the way. After days of fucking around, this horrible hack is the
only way that I've been able to figure out to keep the focus from
being taken away from the entry and ending up somewhere strange
when the popup pops up. (The main problem is with the case of
focus-follows-cursor --- click-to-focus works fine. Sawfish
idiocincracies may also be causing problems, but I don't want to
unjustly accuse the WM of anything, as tempting and appealing as
that can be.)
(key_press_cb): Proxy for forwarding the popup's key press events
to the entry.
(key_release_cb): Proxy for forwarding the popup's key release
events to the entry.
These proxies should be enough to take care of my focus problems.
Unfortunately, they aren't, and the pointer-warping-focus-horror
is required for reasons that I don't fully understand.
* gal/e-text/e-text.c (_get_xy_from_position): Made
_get_xy_from_position return a boolean. It returns TRUE if the
computation was successful
(and if valid data is now in *xp and *yp), FALSE otherwise. Make
sure that text->lines is not NULL, and return FALSE if it is.
(_get_position): Test that _get_xy_from_position returns TRUE
before using the values in x and y.
(_get_position): Test that _get_xy_from_position returns TRUE
before using the values in x and y.
Garbage values being returned in passed-in pointers created a race
condition where you could hang an EText if you deleted the entire
contents of the buffer really quickly.
svn path=/trunk/; revision=9468
2001-03-06 Jon Trowbridge <trow@gnu.org>
* gal/e-text/e-entry.c (e_entry_text_keypress): Added. This is
part of a change of sematics that makes things much nicer for the
user --- auto-activate the completion popup only on keypresses
than also change the entry.
(e_entry_proxy_changed): Record that the underlying EText has
changed, and set up a timeout to clear the changed_since_keypress
flag in a very short amount of time.
(e_entry_init): Connect to the EText's "keypress" signal".
(e_entry_destroy): If our completion_delay_tag timeout is still
floating around out there when we are winding things down, remove
it.
* gal/e-text/e-text.c (e_text_class_init): Added keypress signal.
(e_text_text_model_reposition): Removed some old cruft.
(e_text_event): Emit our keypress signal after handling events of
type GDK_KEY_PRESS.
svn path=/trunk/; revision=8566
2001-03-05 Jon Trowbridge <trow@gnu.org>
* configure.in: Set version number to 0.5.99.5
* gal/e-text/e-entry.c (e_entry_class_init): Add a "popup" signal.
If you have trepidations about this, see the technical argument
below.
(e_entry_init): Connect to the EText's popup signal.
(e_entry_proxy_popup): Proxy emitter for popup signals from the
EText.
* gal/e-text/e-text.c (e_text_class_init): Added a "popup" signal.
(e_text_event): Emit the "popup" signal on right-clicks. Now you
could ask yourself: "should there be a signal named 'popup' in
EText that gets emitted on right-clicks?" And this is a
reasonable question, since... well, this has a funny feeling to
it. But the whole point of GNOME, or at least one of the original
points of GNOME, was to impose policy in a reasonable way when it
made sense in order to simplify the user's experience. Convention
dictates that popup menus are tied to right-clicks --- so rather
than setting up some elaborate forwarding of button-press signals,
why not just impose a little policy and set up a signal that is
closely tied to a familiar set of semantics? Maybe it isn't the
best thing to do from a aesthetics-of-the-API point of view, but I
doubt anyone could condemn it as being anything more than mostly
harmless.
svn path=/trunk/; revision=8565
2001-03-01 Jon Trowbridge <trow@ximian.com>
* configure.in: Boost version number to 0.5.99.3.
* gal/e-text/e-entry.c (e_entry_show_popup): Grab pointer when the
popup is visible, and then hide the popup if any button press
events occur outside of the popup. This lets up avoid most of the
worst "floating popup" cases that would occur if windows are
moved, desktops changed, etc. with the mouse. (Doing things like
changing desktop w/ keybindings can still cause a "floating
popup", but that is also true of Gtk's own combo box.) Change
popup positioning to slightly offset it from the entry, rather
than just plopping it down directly below.
(button_press_cb): Determine if a button press occured outside of
the popup when the pointer was grabbed, and unbrowse accordingly.
* gal/e-text/e-completion-view.c
(e_completion_view_key_press_handler): Improve keystroke handling.
Allow Tabs to pass through (after hiding the pop-up) in order to
allow focus change requests to work properly.
(e_completion_view_construct): Disable horizontal scrollbars.
* gal/e-text/e-completion-test.c (main): Reworked to use signals
instead of explicit callbacks.
* gal/e-text/e-completion.h:
* gal/e-text/e-completion.c: Fix the awkward mix of signals and
explicitly-specified callbacks by taking out the explicit
callbacks. This approach is more gtk-ish, after all.
svn path=/trunk/; revision=8458
2001-02-19 Jon Trowbridge <trow@ximian.com>
* gal/e-text/e-entry.c (e_entry_enable_completion_full): Added.
Allows you to attach an ECompletion to an EEntry, and have that
ECompletion be used for (obviously enough) completions.
* gal/e-text/e-completion-view.h, gal/e-text/e-completion-view.c:
Added. ECompletionView is a widget for displaying the results of
a completion request in a format that is appropriate for a
drop-down window.
* gal/e-text/e-completion.h, gal/e-text/e-completion.c: Added.
ECompletion is a "pure virtual base class" for completion-type
operations. It is implemented so that completions can be either
synchronous or asynchronous.
* gal/e-text/e-text.c: Lots of changes to accomodate the
ETextModel changes. First of all, we render embedded text objects
as being underlined. We also cause the model to emit the
appropriate object activation signal when an embedded object is
double-clicked. Also, all of the code that moves the cursor in
response to user input has been removed. Instead, the EText now
listens for "reposition" events from the underlying model, and
bases all cursor motions on those.
(get_bounds_item_relative): Fixed bug in the handling of
differently-anchored text. Being differently-anchored is not a
crime or a perversion --- it is an alternative lifestyle that we
have to respect.
* gal/e-text/e-text-model-uri.h, gal/e-text/e-text-model-uri.c: A
sample ETextModel that converts URIs into embedded objects that
get opened in the browser when you double-click them.
* gal/e-text/e-text-model-repos.h,
gal/e-text/e-text-model-repos.c: Added. A group of simple
structures & functions for handling various cursor movement rules.
These are the sorts of things that are passed as arguments to
ETextModel "reposition" event handlers.
* gal/e-text/e-text-model.h, gal/e-text/e-text-model.c: Privitized
the ETextModel struct and "methodized" all of the operations, so
that derived classes can do arbitrarily respond to get/set
requests in arbitrarily strange ways. Also added the concept of
declaring regions of the text as "embedded text objects".
Finally, caused operations that change the text to emit a
"reposition" signal that passes information that can be used by a
view (like an EText) to move the cursor or selection in an
intelligent way in response to those changes. This means that you
can now open two ETexts that look at the same ETextModel, and have
the cursor in one do the right thing when you edit the other. (As
opposed to producing a lot of potential segfaults, as it was
before.)
svn path=/trunk/; revision=8280
