Bot State Persistence #55
Conversation
solomon-b
force-pushed
the
feature/bot-persistence
branch
from
7f13b36 to
2452f86
Compare
|
You should probably change the target of the PR to the branch for #54 to make this easier to review. |
|
@solomon-b Thanks. Looks like some of the persistence stuff went into #54 though. |
solomon-b
force-pushed
the
feature/todo-behavior
branch
from
6ef179e to
f58db2b
Compare
solomon-b
force-pushed
the
feature/bot-persistence
branch
from
2452f86 to
857871e
Compare
solomon-b
force-pushed
the
feature/todo-behavior
branch
from
f58db2b to
f189b7a
Compare
solomon-b
force-pushed
the
feature/bot-persistence
branch
from
857871e to
0f4aec2
Compare
|
@masaeedu yeah sorry about that. Its fixed now. I just made some interface changes in the other branch and the rebasing got mixed up. I'll paste a script for how to use it in like an hour. You can also look at the parser to figure it out. |
solomon-b
force-pushed
the
feature/todo-behavior
branch
from
f189b7a to
9b11eee
Compare
solomon-b
force-pushed
the
feature/bot-persistence
branch
2 times, most recently
from
4233541 to
c911b2f
Compare
solomon-b
force-pushed
the
feature/todo-behavior
branch
from
9b11eee to
9bc7e02
Compare
solomon-b
force-pushed
the
feature/bot-persistence
branch
2 times, most recently
from
2ed52b1 to
bb69fda
Compare
solomon-b
force-pushed
the
feature/bot-persistence
branch
from
bb69fda to
b5622ee
Compare
app/Main.hs
Outdated
| simplifyMatrixBot $ | ||
| bot | ||
| process | ||
| state <- readState "state" |
There was a problem hiding this comment.
what type does this get?
src/CofreeBot/Bot.hs
Outdated
|
|
||
| fixBotPersistent :: forall m s i o. (MonadIO m, Serializable s) => Bot m s i o -> s -> IO (Behavior m i o) | ||
| fixBotPersistent (Bot bot) initialState = do | ||
| saveState "state" initialState |
There was a problem hiding this comment.
Probably callers of this function should give a key, otherwise they will all clobber one another, right?
There was a problem hiding this comment.
Are you imagining having multiple bots running on the same linux user?
There was a problem hiding this comment.
I'm imaging each sub bot wants to manage its own state separately from the global composition
There was a problem hiding this comment.
Do you mean that you would want bots to be able to state explicitly how their individual state is persisted?
There was a problem hiding this comment.
I'm imaging each sub bot wants to manage its own state separately from the global composition
I think @isovector is talking about a scenario in which a user fixBotPersistents several independent bots, then laterally composes the resulting Behaviors (or whatever it is we're calling them now), instead of doing it the other way around as we're doing now.
This is a totally reasonable use case (although not one we need right now), and it'd be pretty easy to support by just adding a parameter for the file path here. We can just pass in xdgCache </> "state" or whatever it is as that file path closer to Main.
There was a problem hiding this comment.
Oh ok that is a totally reasonable user expectation. I'll thread in the path from main rather then hard code it here.
| @@ -30,7 +31,7 @@ main = do | |||
| TokenCmd TokenCredentials {..} -> do | |||
| session <- createSession (getMatrixServer matrixServer) matrixToken | |||
| matrixMain session xdgCache | |||
| CLI -> cliMain | |||
| CLI -> cliMain xdgCache | |||
There was a problem hiding this comment.
We probably don't have enough config yet to warrant this change, but we might want to eventually have an app-specific monad in which "global" config like this is accessible via MonadReader, so that we don't have to pass it around.
There was a problem hiding this comment.
Yeah I agree once the executable gets a bit more sophisticated.
| liftIO $ saveState cachePath newState | ||
| pure (output, go) | ||
|
|
||
| readState :: Read s => String -> IO (Maybe s) |
There was a problem hiding this comment.
It would be nice if we could avoid opening and closing this file over and over, but I don't have any good ideas at this moment about how to do it.
There was a problem hiding this comment.
I guess we could read the file once and then persist it in an IORef/MVar/TVAR for the lifecycle of the bot? If we go that route we should put it in a followup PR
Adds a context function which moves a bot's in-memory state to the file system.