[r2r] Fix task::*::cancel if the RPC task is an awaiting status

[sergeyboyko0791]

The issue was:

1. cancel_init_trezor:
   https://github.com/KomodoPlatform/atomicDEX-API/blob/6c90c28a29723dad6c48296bed3093686188eeb2/mm2src/mm2_main/src/lp_init/init_hw.rs#L209
2. RpcTaskManager::cancel_task:
   https://github.com/KomodoPlatform/atomicDEX-API/blob/6c90c28a29723dad6c48296bed3093686188eeb2/mm2src/rpc_task/src/manager.rs#L111-L112
3. A corresponding item is removed from the RpcTaskManager::tasks:
   https://github.com/KomodoPlatform/atomicDEX-API/blob/6c90c28a29723dad6c48296bed3093686188eeb2/mm2src/rpc_task/src/manager.rs#L21-L22
4. If the task is in Awaiting status, abort_handle AND action_sender are dropped
   https://github.com/KomodoPlatform/atomicDEX-API/blob/6c90c28a29723dad6c48296bed3093686188eeb2/mm2src/rpc_task/src/manager.rs#L269-L274
5. I don't know why, but action_sender channel is closed before abort_handle, so the task finishes with an error, but we expect that it should abort.

[shamardy]

I don't know why, but action_sender channel is closed before abort_handle, so the task finishes with an error, but we expect that it should abort.

Do you think the drop order can affect this? If we reversed action_sender and abort_handle positions, would that solve the problem. Just curious :)

[sergeyboyko0791]

I haven't checked, but this behaviour is dependent on the executor, and we can't guarantee that the abort_handle receiver will be notified at first, and the action_sender later.
So I decided to make the 100% solution :)

Do you think the drop order can affect this? If we reversed action_sender and abort_handle positions, would that solve the problem. Just curious :)

[sergeyboyko0791]

@shamardy I've checked - the order doesn't matter 😁
I.e. it matters, but not so much that the order of awaken futures depends on it

[shamardy]

I've checked - the order doesn't matter 😁

I thought that it won't, I was just curious as I said :)

So I decided to make the 100% solution :)

Agreed, it's better than leaving it in the hands of a race condition.

[shamardy]

LGTM!

[smk762]

Just gave it a test in CLI.
PIN grid stays on screen when cancelling the init task, but I can init again and the grid changes each time I cancel and re-init.

[sergeyboyko0791]

As expected. It's turned out to be difficult to fix, so I created the issue #1586

Just gave it a test in CLI. PIN grid stays on screen when cancelling the init task.

[artemii235]

👍