This needs the ITEM_MODIFICATION_ROUTINE to learn to delete items.
Condense the item change handlers into a single injected object
(ITEM_MODIFICATION_ROUTINE::CHANGE_HANDLER) and provide the basic
implementation that just takes some callables.
This simplifies the construction of the routines and also would make
a CHANGE_HANDLER object possible that can be reused between different
tools.
In the EDIT_TOOL::ModifyLines method, there was some
remaining special-casing for FP childen commits.
One bit was put back by mistake (in
0b32ae3f4b) because I thought I'd
left it out of the refactor. In fact it was removed slightly later in
1218f61d0a.
THe other bit is omitting Modify commits in the FP-editor (what the
reinstatement of the above code was trying to cause). This shouldn't
be needed any more.
However, something is still incomplete here as the tools still don't
work correctly in the footprint editor. However, I don't think it's
substantially _more_ broken with the special casing removed and at
least it's less confusing.
Related to: https://gitlab.com/kicad/code/kicad/-/issues/15253
Using the new ITEM_MODIFICATION_ROUTINE system, drop in two new
tools: chamfer and line extend. These are two geometric operations
that are relatively common when editing footprints in particular.
Chamfer delegates the geometric calculations to a dedicated unit
in kimath/geometry.
Describe the actions of the fillet tools is a generic way, so that the
same general pattern can be used for other tools that modify shapes on
the BOARD.
Basically, an "ITEM_MODIFICATION_ROUTINE" is defined, which is
configured and called multiple times, calling back to the EDIT_TOOL when
it modifies or creates an item.
The motivation here is to make it easier to slot in new line-based
tools like chamfer, extend and so on without having to redo the
complicated item, selection and commit handling each time, and keep the
core "routines" simple and decoupled from the EDIT_TOOL's
internals.
This also resolves#15094 because the new commit handling does the right
thing when items were "conjured up" for the fillet (e.g. when a
rectangle is decomposed into lines).
Fixes: #15094
Note that "immediate" doesn't mean quite the same thing: while it will
enter the tool immediately, it won't necessarily finish the tool during
the call if the tool has an event loop. So for something like Rotate
"immediate" and "synchronous" have the same behaviour, but for something
like Move they do not.
Fixes https://gitlab.com/kicad/code/kicad/-/issues/15085
It had several encapsulation leakage issues, as well as poorly-defined
behaviour of undo for chained-actions (append-to-board, and then rotate
while moving, for instance).
Using a boolean argument just leads to a lot of trailing booleans in the
function calls and is not user friendly. Instead, introduce PostAction()
to send an action that runs after the coroutine (equivalent to passing
false or the default argument), and leave RunAction as the immediate
execution function.
Using std::any from C++17 allows for proper type handling in the
parameter field, removing the need for casting to void* and then casting
the void* to the desired type.
Also adds some debugging to try to catch dangling pointers.
Also adds a cache for group bounding boxes (which will be expensive
to calculate for large groups).
Fixes https://gitlab.com/kicad/code/kicad/issues/12875
Jeff Young
John Beard
jean-pierre charras
Ian McInerney
Mike Williams
Seth Hillbrand
Alex