mirror of
https://gitlab.gnome.org/GNOME/gtk.git
synced 2024-11-18 17:30:10 +00:00
211 lines
7.2 KiB
Plaintext
211 lines
7.2 KiB
Plaintext
|
This document describes some of the internals of the DND handling
|
||
|
code.
|
||
|
|
||
|
Organization
|
||
|
============
|
||
|
|
||
|
The DND code is split between a lowlevel part - gdkdnd.c and a
|
||
|
highlevel part - gtkdnd.c. To put it simply, gdkdnd.c contain the
|
||
|
portions of DND code that are easiest to do in raw X, while gtkdnd.c
|
||
|
contains the portions of DND that are easiest to do with an event loop
|
||
|
and high level selection handling.
|
||
|
|
||
|
Except for a few details of selection handling, most of the
|
||
|
dependencies on the DND protocol are confined to gdkdnd.c.
|
||
|
There are two or three supported protocols - Motif DND,
|
||
|
Xdnd and a pseudo-protocol ROOTWIN, which is used for drops
|
||
|
on root windows that aren't really accepting drops.
|
||
|
gdkdnd.c divides into 4 pieces:
|
||
|
|
||
|
1) Utility functions (finding client windows)
|
||
|
2) Motif specific code (the biggest chunk)
|
||
|
3) Xdnd specific code
|
||
|
4) The public interfaces
|
||
|
|
||
|
The code in gtkdnd.c roughly consists of three parts
|
||
|
|
||
|
1) General utility functions
|
||
|
2) Destination side code
|
||
|
3) Source side code.
|
||
|
|
||
|
Both on the source and dest side, there is some division
|
||
|
between the low level layers and the default handlers,
|
||
|
though they are rather mixed in many cases.
|
||
|
|
||
|
Structures and Memory Management
|
||
|
================================
|
||
|
|
||
|
Information about source sites and drop sites is stored
|
||
|
in the structures GtkSourceSite and GtkDestSite.
|
||
|
|
||
|
Information about in-progress drags and drops is stored
|
||
|
in the structures GtkSourceInfo and GtkDestInfo.
|
||
|
|
||
|
The GtkSourceInfo structure is created when the drag
|
||
|
begins, and persists until the drag either completes
|
||
|
or times out. A pointer to it is stored in
|
||
|
dataset-data for the GdkDragContext, however there
|
||
|
is no ownership. If the SourceInfo is destroyed
|
||
|
before the context, the field is simply cleared.
|
||
|
|
||
|
A GtkDestInfo is attached to each GdkDragContext
|
||
|
that is received for an incoming drag. In contrast
|
||
|
to the SourceInfo the DestInfo is "owned" by the
|
||
|
context, and when the context is destroyed, destroyed.
|
||
|
|
||
|
The GDK API
|
||
|
===========
|
||
|
|
||
|
It is expect that the GDK DND API will never be
|
||
|
used by anything other than the DND code in GTK+.
|
||
|
|
||
|
/* Drag and Drop */
|
||
|
|
||
|
GdkDragContext * gdk_drag_context_new (void);
|
||
|
void gdk_drag_context_ref (GdkDragContext *context);
|
||
|
void gdk_drag_context_unref (GdkDragContext *context);
|
||
|
|
||
|
These create and refcount GdkDragContexts in a
|
||
|
straightforward manner.
|
||
|
|
||
|
/* Destination side */
|
||
|
|
||
|
void gdk_drag_status (GdkDragContext *context,
|
||
|
GdkDragAction action,
|
||
|
guint32 time);
|
||
|
void gdk_drop_reply (GdkDragContext *context,
|
||
|
gboolean ok,
|
||
|
guint32 time);
|
||
|
void gdk_drop_finish (GdkDragContext *context,
|
||
|
gboolean success,
|
||
|
guint32 time);
|
||
|
GdkAtom gdk_drag_get_selection (GdkDragContext *context);
|
||
|
|
||
|
/* Source side */
|
||
|
|
||
|
GdkDragContext * gdk_drag_begin (GdkWindow *window,
|
||
|
GList *targets,
|
||
|
GdkDragAction actions);
|
||
|
gboolean gdk_drag_get_protocol (guint32 xid,
|
||
|
GdkDragProtocol *protocol);
|
||
|
void gdk_drag_find_window (GdkDragContext *context,
|
||
|
GdkWindow *drag_window,
|
||
|
gint x_root,
|
||
|
gint y_root,
|
||
|
GdkWindow **dest_window,
|
||
|
GdkDragProtocol *protocol);
|
||
|
gboolean gdk_drag_motion (GdkDragContext *context,
|
||
|
GdkWindow *dest_window,
|
||
|
GdkDragProtocol protocol,
|
||
|
gint x_root,
|
||
|
gint y_root,
|
||
|
GdkDragAction action,
|
||
|
guint32 time);
|
||
|
void gdk_drag_drop (GdkDragContext *context,
|
||
|
guint32 time);
|
||
|
void gdk_drag_abort (GdkDragContext *context,
|
||
|
guint32 time);
|
||
|
|
||
|
GdkAtom gdk_drag_get_selection (GdkDragContext *context);
|
||
|
|
||
|
Retrieves the selection that will be used to communicate
|
||
|
the data for the drag context (valid on both source
|
||
|
and dest sides)
|
||
|
|
||
|
Cursors and window heirarchies
|
||
|
==============================
|
||
|
|
||
|
The DND code, when possible (and it isn't possible over
|
||
|
Motif window) uses a shaped window as a drag icon.
|
||
|
Because the cursor may fall inside this window during the
|
||
|
drag, we actually have to figure out which window
|
||
|
the cursor is in _ourselves_ so we can ignore the
|
||
|
drag icon properly. (Oh for OutputOnly windows!)
|
||
|
|
||
|
To avoid obscene amounts of server traffic (which are only
|
||
|
slighly observerable locally, but would really kill a
|
||
|
session over a slow link), the code in GDK does
|
||
|
XGetWindowAttributes for every child of the root window at
|
||
|
the beginning of the drag, then selects with
|
||
|
SubstructureNotifyMask on the root window, so that
|
||
|
it can update this list.
|
||
|
|
||
|
It probably would be easier to just reread the entire
|
||
|
list when one of these events occurs, instead of
|
||
|
incrementally updating, but updating the list in
|
||
|
sync was sort of fun code, so I did it that way ;-)
|
||
|
|
||
|
There is also a problem of trying to follow the
|
||
|
mouse cursor as well as possible. Currently, the
|
||
|
code uses PointerMotionHint, and an XQueryPointer
|
||
|
on MotionNotify events. This results in pretty
|
||
|
good syncing, but may result in somewhat poor
|
||
|
accuracy for drops. (Because the coordinates of
|
||
|
the drop are the coordinates when the server receives
|
||
|
the button press, which might actually be before
|
||
|
the XQueryPointer for the previous MotionNotify
|
||
|
event is done.)
|
||
|
|
||
|
Probably better is doing MotionNotify compression
|
||
|
and discarding MotionNotify events when there
|
||
|
are more on the queue before the next ButtonPress/Release.
|
||
|
|
||
|
Proxying
|
||
|
========
|
||
|
|
||
|
A perhaps rather unusual feature of GTK's DND is proxying. A
|
||
|
dest site can be specified as a proxy drop site for another
|
||
|
window. This is most needed for the plug-socket code - the
|
||
|
socket needs to pass on drags to the plug since the original
|
||
|
source only sees toplevel windows. However, it can also be
|
||
|
used as a user visible proxy - i.e., dragging to buttons on
|
||
|
the taskbar.
|
||
|
|
||
|
Internally, when the outer drag enters a proxy dest site, a
|
||
|
new source drag is created, with SourceInfo and
|
||
|
GdkDragContext. From the GDK side, it looks much like a
|
||
|
normal source drag; on the GTK+ side, most of the code is
|
||
|
disjoint. The need to pass in a specific target window
|
||
|
is the reason why the GDK DND API splits
|
||
|
gdk_drag_find_window() and gdk_drag_motion().
|
||
|
|
||
|
For proxy drags, the GtkDestInfo and GtkSourceInfo for the
|
||
|
drag point at each other.
|
||
|
|
||
|
Because the abstraction of the drag protocol is at the GDK
|
||
|
level, a proxy drag from Motif to Xdnd or vice versa happens
|
||
|
pretty much automatically during the drag, though the
|
||
|
drop can get complicated. For Xdnd <-> Motif,
|
||
|
Motif <-> Xdnd, or Motif <-> Motif drags, it is necessary to
|
||
|
for the Proxy to retrieve the data and pass it on to
|
||
|
the true destination, since either the selection names
|
||
|
differ or (Motif<->Motif), the proxy needs to know
|
||
|
about the XmDRAG_SUCCESS/FAILURE selection targets.
|
||
|
|
||
|
Further Reading:
|
||
|
================
|
||
|
|
||
|
Xdnd:
|
||
|
|
||
|
The spec is at:
|
||
|
|
||
|
http://www.cco.caltech.edu/~jafl/xdnd/
|
||
|
|
||
|
Motif:
|
||
|
|
||
|
The Motif DND protocol is best described in the
|
||
|
Hungry Programmers _Inside Lesstif_ book, available
|
||
|
from:
|
||
|
|
||
|
http://www.igpm.rwth-aachen.de/~albrecht/hungry.html
|
||
|
|
||
|
Harald Albrecht and Mitch Miers have done a far
|
||
|
better job at documenting the DND protocol then
|
||
|
anything the OpenGroup has produced.
|
||
|
|
||
|
|
||
|
|
||
|
Owen Taylor
|
||
|
otaylor@redhat.com
|
||
|
Oct 18, 1998
|