[FEATURE] Interruption behaviour of Conversations

[hyzyla]

Issue I am facing

Hi, I have faced a problem where I have a few conversations and a few commands:

    dp.add_handler(
        ConversationHandler(
            entry_points=[CommandHandler('command1', command1)],
            states={
                STATE1: [MessageHandler(Filters.text, do_something1)],
            },
            fallbacks=[AnyHandler(do_something1_fallback)],
            allow_reentry=True,
        )
    )
    dp.add_handler(CommandHandler('command2', command2))
    dp.add_handler(
        ConversationHandler(
            entry_points=[CommandHandler('command3, command3)],
            states={
                STATE2: [MessageHandler(Filters.video, do_something3)],
            },
            fallbacks=[AnyHandler(do_something3_fallback)],
            allow_reentry=True,
        )
   )
   dp.add_handler(AnyHandler(global_fallback))

Now when my bot handles conversation for command3, I can easy to interrupt the conversation by entering command2 or command3. But interruption doesn't reset state of conversation for command3.
For example:

U: /command3
B: Send video
U: (sends a text)
B: Send video, please
U: /command1
B: Send text
U: (sends a text)
B: Thanks, I saved it
U: BLABLA
B: Send video, please.

The last sentence is not obvious, why I need a video (do_something3_fallback) when I finished the conversation in the middle of full conversational? More generaly I expect to see the result of global_fallback.

My idea to add a parameter isolated to conversational that doesn't allow interruption by any command and when such conversational started than dispatcher can only view handlers of started conversation.

I know that I can implement it by nesting Conversational Handlers, but then my code looks very ugly and messy.

[Bibo-Joshi]

Please habe a look at the group argument of do.add_handler. TLDR: only one handler per group is excecuted. Putting your global_fallback in it's own group should suffice for it to be triggered.

[hyzyla]

@Bibo-Joshi Sorry, I read documentation and source code for add_dispatcher and handle_update and still don't understand how it solves my problem when I need to handle or prevent conversation interruptions? Could you please, help me with that?

[Bibo-Joshi]

Well, it doesn't prevent a conversation from being interrupted but can be used to solve this:

But interruption doesn't reset state of conversation for command3.

See the discussion in #1447
However, I do find the idea of non-interruptable ConversationHandlers interesting.

[hyzyla]

I added groups to handlers and it's solved my problem (interrupt conversation after some command), but such a method seems a bit hacky to achieve my goal.

    dp.add_handler(
        ConversationHandler(
            entry_points=[CommandHandler('command1', command1)],
            states={
                STATE1: [MessageHandler(Filters.text, do_something1)],
            },
            fallbacks=[MessageHandler(Filters.command, cancel_conversation), AnyHandler(do_something1_fallback)],
            allow_reentry=True,
        ),
        group=1,
    )
    dp.add_handler(CommandHandler('command2', command2), group=100)
    dp.add_handler(
        ConversationHandler(
            entry_points=[CommandHandler('command3, command3)],
            states={
                STATE2: [MessageHandler(Filters.video, do_something3)],
            },
            fallbacks=[MessageHandler(Filters.command, cancel_conversation), AnyHandler(do_something3_fallback)],
            allow_reentry=True,
        ),
        group=2,
   )
   dp.add_handler(AnyHandler(global_fallback), group=100)

I do not have idea hot improve that, maybe some flag to the conversational handler that interrupts the conversation when update handled by some handler outside of conversation and some parameters interrupt handler. What do you think about it?

[Bibo-Joshi]

Well your solution fallbacks=[MessageHandler(Filters.command, cancel_conversation), ...] seems pretty reasonable to me.
Still, I've added this issue to the v12.5 milestone. At this point I think a PR should add two kwargs to ConversationHander:

• cancel_on_interruption (or some similar naming): If True, the CHs state will be set to END, if coversation is interruptet. One could also think about allowing to pass a callback that's executed on cancellation or even a dict (mapping states to callbacks) so that the callback depends on the state in wich the conversation is cancelled.
• allow_interruption (or some similar naming): If False the conversation will not be interrupted until it has ended

Would you like to give at a go yourself?

[Bibo-Joshi]

PS: Preventing your CH from interruption can be done by adding fallback in the "isolated" CH that catches e.g. entrypoints of other conversations. However, this will only work for handlers in the same group.
Also, the DispatcherHandlerStop may be useful for you.

Also, if you'd like to PR, please tell us about your proposed changes beforehand, so we can keep in touch b/c CH is quite complex …

[hyzyla]

Sorry, I don't have time to work on this feature

[TempAccountForResponseToQuestion]

A very dirty hack that ends all conversational handlers. Modify under your task.

all_hand = dispatcher.handlers
for dict_group in all_hand:
    for handler in all_hand[dict_group]:
        if isinstance(handler, ConversationHandler):
            handler.update_state(ConversationHandler.END, handler._get_key(update))

c2hpdCBjb2Rl

2021-07-01, Comment by @Bibo-Joshi : The update_state method was never documented and should be considered internal. By now (v13.6) it's private. I highly recommend not to use it, as its behavior may change without notice.

[Bibo-Joshi]

Leaving this here for completeness sake: Nested CHs need to be considered, if this is tackled.