In my attempt to understand the drag/dock implementation of Lazarus, I came across some general questions. Is there a developer who is familiar with the current implementation of these things?
1. Model
--------
IMO a separation into drag/dock execution and dock site layout manager is suggested.
The drag/dock part only has to deal with the implementation of the graphical user feedback, based on the dock source properties DragMode and DragKind. It interacts with a drag/dock target at the current mouse position, to adjust the drag image appropriately. Depending on DragKind, the target should have UseDockManager=True or react on DockOver for dkDock, or it should react on DragOver for dkDrag. The search for a possible target continues into the parents of the control, until an accepting target is found, or the parent is Nil. This part of drag/dock execution IMO should be feasable for every platform, somehow. For testing purposes or on poorly equipped platforms it would be sufficient to change the drag image (or mouse pointer) according to the response of the target control.
The layout manager(s) are responsible for inserting and removing components, and can be shared with the layout managers of the form designers, and the handling of client/container size and other layout managers could be added freely. The anchordocking sample demonstrates an layout manager with no docking capabilities (yet).
Dock-aware layout managers provide additional means to determine the parameters for visual feedback and a later DockDrop action. They also are responsible for starting an undock action at runtime, or for a rearrangement of components at design time.
2. Platform support
-------------------
I don't know what features are provided by the currently supported platforms, and nobody can know the features of future platforms. That's why I would restrict the expected support for drag/drop operations to very few features:
2.1 Determination of the component at the mouse position. This feature should be available on every platform, since the system itself must know which target window has to be notified of mouse events. Using this feature should allow to easily fill the white spots in the current determination of the dock site.
2.2 Mouse capture
Every platform should support kind of mouse capture, i.e. sending messages to a dedicated capture window, instead of informing the window under the mouse position. Information about active capturing and dragging should be stored in a dedicated place (Application object?), so that it can canceld easily and properly, whenever required. I dunno about the synchronization between the IDE and a debugger and debugged application, but I assume that there exists according communication, which should allow to properly cancel unrecoverable actions.
2.3 Visual feedback
Every platform should at least support mouse pointer shapes, for visual feedback about the actual target of a drag action. This primitive kind of feedback also requires no special management of temporary images on the screen, or in design state. More elaborated feedback can be added for platforms with better support for dragging operations (drag images etc.). Question is: how to deal with platform specific features in the LCL? Another kind of standard support could use the hint-window mechanism, and drag around such a temporary window. Then the target of a dock operation only had to supply the dock zone (RECT), so that the dragged window can be properly resized and repositioned. This solution is really useful only with transparent windows, which do not hide the possible dock targets, both from the sight of the user, nor from the sight of the system in the determination of the drag target. Perhaps not a good idea?
3. Dock zones
-------------
IMO the current integration of drag/dock support bloats the TWinControl class too much. Much could be moved into the DockManager, except for the event handlers, UseDockManager and the DockManager itself. Even the information about the undocked size of docked components could be stored in the dock manager. While the data bloat may be acceptable, the code bloat hinders a proper (re-)design of the buggy or still missing parts of the drag/dock code.
3.1 Dock sites
IMO dock sites could be implemented in dedicated container controls, which react on docking controls also can implement support for floating themselves, e.g. as detachable frames or toolbars. Delphi compatibility will not be affected, but a choice of ready-to-use docking components can simplify the design of new Lazarus applications.
3.2 Dockable components
IMO the docking of simple components is useful only at design time, when components are dropped from the toolbar onto a form. Docking at runtime can be restricted to forms and other floatable components, like toolbars or frames. Undocking of docked components requires that the structure of the docked components stays intact, i.e. forms or frames must be undocked in their original composition and framing. This forbids unpacking the components of a form, when the form is docked. When a form or frame instead is docked entirely, including a (small) title bar as in Delphi, every docked item can be restored exactly into it's previous appearance and behaviour, what currently simply is impossible.
3.3 Determination of the docking position.
I'll not go into details here, it's mostly a matter of implementation of the an docking manager. The default (tree) docking manager implementation lacks many docking features which can hardly be added, or even be made work, without a deep redesign of all related classes and other data types.
Therefore I'd suggest a separation of any but the abstract TDockManager from the Controls unit, eliminating the need for a rebuild of the IDE after every change to the Controls unit. It also will show up the undocumented dependencies between controls, messages, and the code that implements the details of dragging and docking. I'm willing to perform that separation myself, if no official developer is willing to do so - but I'm not willing to spend much time when afterwards the new structure doesn't become part of the trunk.
DoDi