Take the DAP lock on the R thread to avoid deadlock

[lionel-]

Addresses posit-dev/positron#5024

I can't reproduce the deadlock anymore, but the approach taken here should fix all deadlock issues: We only take the lock on the DAP state when we're on the R thread.

[DavisVaughan]

I managed to deadlock it in dplyr with dev ark, i.e. cargo build without --release.

# in dplyr
devtools::load_all()
debugonce(mutate_col)
mutate(mtcars, x = mpg+cyl)

I stepped in and out of some functions and it deadlocked while coming out of eval_all_mutate()

I think I can confirm your fix here based on the frozen stack in ark:

• the main R thread is waiting on the lock in debug_start()
• but the ark-dap thread is indeed paused in collect_r_variables() waiting on the r_task() to run while its holding the lock on state
• so the r_task() can't run because the main R thread can't advance, etc, etc, deadlock

You'll also see a request for variables was just received in the logs, but then we got stuck

[DavisVaughan]

I can't seem to deadlock it with this PR, though it took me a long time to do it with dplyr the first time around so I can't be 100% certain. But based on my screenshots above, this looks like the right fix!

[DavisVaughan]

Suggested change

	let state = self.state.clone();
	let state = Arc::clone(&self.state);

Just so it's very clear this is a cheap clone. I had to go look it up myself to make sure.

This is a typical rust pattern i see with Arc

[lionel-]

yep same diagnostics as ^^ was provided in posit-dev/positron#5024, it's good to confirm we're observing the same thing (or was, I can't reproduce anymore)