Inject/Project find effects nested on the left #219
Conversation
| eff (R (R (R other))) = CullC (eff (R (R (R (handleCoercible other))))) | ||
| eff (R (L (L Empty))) = empty | ||
| eff (R (L (R (Choose k)))) = k True <|> k False | ||
| eff (R (R other)) = CullC (eff (R (R (handleCoercible other)))) |
There was a problem hiding this comment.
Basically just reassociating here.
| inj = reassoc . inj where | ||
| reassoc (L l) = L (L l) | ||
| reassoc (R (L l)) = L (R l) | ||
| reassoc (R (R r)) = R r |
There was a problem hiding this comment.
The reassociation here won’t be too bad a performance hit as long as we’re reasonable about using left-nested sums.
There was a problem hiding this comment.
Can you elaborate on what you mean by this?
There was a problem hiding this comment.
And can we document why we need to reassociate here (and maybe drop a link to DTALC for people who are wondering about why all these OVERLAPPABLE pragmas need to happen?
There was a problem hiding this comment.
Can you elaborate on what you mean by this?
Basically, this assumes that trees will generally be right-leaning. So anything exclusively right-chained won’t ever see this instance, and if you have (A :+: B) :+: C that’s an extremely minor hit (as your benchmark demonstrates). But a long left-chain—which is extremely unlikely to happen, and never by accident; you would have to work quite hard to design your effects thus—will take more of a hit because it has to do a bunch more reassociation.
There was a problem hiding this comment.
And can we document why we need to reassociate here (and maybe drop a link to DTALC for people who are wondering about why all these OVERLAPPABLE pragmas need to happen?
I’m gonna do this in #223 to avoid conflicts.
| inj = reassoc . inj where | ||
| reassoc (L l) = L (L l) | ||
| reassoc (R (L l)) = L (R l) | ||
| reassoc (R (R r)) = R r |
There was a problem hiding this comment.
Can you elaborate on what you mean by this?
| inj = reassoc . inj where | ||
| reassoc (L l) = L (L l) | ||
| reassoc (R (L l)) = L (R l) | ||
| reassoc (R (R r)) = R r |
There was a problem hiding this comment.
And can we document why we need to reassociate here (and maybe drop a link to DTALC for people who are wondering about why all these OVERLAPPABLE pragmas need to happen?
I can’t think of anything in |
This PR allows effects to occur on the left as well as the right, which will enable us to split
Errorup into bothThrowandCatch, reintroduceNonDetas a synonym forEmpty :+: Choose, etc.NonDetas a synonym forEmpty :+: Choose.NonDetin theCarrierinstances forNonDetC,CutC, andCullC.SplitsDeferring to a future PR.ErrorintoThrow&Catch.RedefinesDeferring to a future PR.Failas a synonym forThrow String.