Set the current seat in seatop handlers

[rpigott]

Fixes #5939, see comment.

The functions in container.c all like to reference the current seat, so we better make sure it is set before we do anything that may deal with focus. Currently, when the seatop_ commands run, the handler_context.seat can be anything, even NULL, so I've chosen to set it in all the seatop_begin* functions. I don't suppose it matters for some, like floating_resize, but I set it in all of them anyway for symmetry.

[rpigott]

This is the only line changed in this commit. Otherwise these three functions were lifted directly from commands/swap.c

[rpigott]

I don't see why container_split should use the default seat while the reset use the current seat, so I've changed it.

[emersion]

config->handler_context is mostly meant for command handlers, not for tree manipulation functions. Can we instead pass the seat via an argument to container_swap?

[rpigott]

config->handler_context is mostly meant for command handlers, not for tree manipulation functions. Can we instead pass the seat via an argument to container_swap?

Technically yes, but I think this way is better. My reasoning is as follows:

The container* functions need a reference to a seat whenever they may mess with the focus, but none accept a seat argument. They all get a seat reference currently with input_manager_current_seat() with the exception of container_split, which I believe is an error, hence the 3rd commit.

The implementation of input_manager_current_seat is:

struct sway_seat *input_manager_current_seat(void) {
	struct sway_seat *seat = config->handler_context.seat;
	if (!seat) {
		seat = input_manager_get_default_seat();
	}
	return seat;
}

So they do already reference the handler_context.seat to use the correct seat when invoked from the command handlers.

Well the seatop actions are just like commands, and setting handler_context.seat in seatop_begin* is analogous to setting it in execute_command. This change basically just treats the seatop actions like the commands that they are. If anything I think it should go further and set the handler_context node/container etc.

[rpigott]

Actually I suppose it would be better to analogize each tiling dropzone with a command, rather than the whole tiling_move operation. In that case seatop_begin* is the wrong place to set handler_context.seat, it should be when the action ends before we start modifying the tree. Similarly the other seatops should only set it when they actually do something.

[emersion]

I'm not a fan of setting global state like this. We did it because it was the less intrusive way to do it for commands, but I'd prefer not to make more stuff rely on it.

[rpigott]

I generally agree, and I think there are still bugs in the PR as written. I'll need to think about this one a bit, so I've converted to a draft for now.