defined Alternative

[BebeSparkelSparkel]

No description provided.

[andreasabel]

Thanks @BebeSparkelSparkel !
Could you please provide proofs of the monad and alternative laws in a comment with the instance declaration?

[BebeSparkelSparkel]

Added to tests.

[andreasabel]

I was a bit surprised to read that there isn't a consensus about what the laws for Alternative should be: https://en.wikibooks.org/wiki/Haskell/Alternative_and_MonadPlus#Other_suggested_laws

I am a bit hesitant here still. Does it make sense to have Alternative for a monad that does destructive updates like ST? In, general, even if a fails and b succeeds, a <|> b will then have also updates coming from a.

@josefs What do you think?

[BebeSparkelSparkel]

@andreasabel Alternative is defined for IO so it is hard to say that monads with destructive updates should not have an Alternative instance.

[andreasabel]

Thanks @BebeSparkelSparkel ! Published as 0.4.8.

[josefs]

Apologies for the late reply, I've been on an extended sick leave.

I would have liked to see a good motivating example here.

Side effects in STT cannot be "undone" so these kinds of instances are rather dodgy. Now, as far as I understand, the laws of Alternative are not strong enough to require that side effects be undone, unlike MonadPlus. So I'm not going to ask to revert this. But whoever decides to use this better know what they're doing.

Below is the kind of program I'm worried about. The result of this function is 1 but someone unfamiliar with the difference between Alternative and MonadPlus might think it should be 0. So there's a real risk of confusion here.

foo = do
  r <- newSTRef 0
  (modifySTRef (+1) r >> lift empty <|> return ())
  readSTRef r

[andreasabel]

Apologies for the late reply, I've been on an extended sick leave.

Sorry to hear, and glad that your are back! I wish you a good 2024!

I would have liked to see a good motivating example here.

@BebeSparkelSparkel: Would you provide such an example?

Side effects in STT cannot be "undone" so these kinds of instances are rather dodgy. Now, as far as I understand, the laws of Alternative are not strong enough to require that side effects be undone, unlike MonadPlus. So I'm not going to ask to revert this. But whoever decides to use this better know what they're doing.

Below is the kind of program I'm worried about. The result of this function is 1 but someone unfamiliar with the difference between Alternative and MonadPlus might think it should be 0. So there's a real risk of confusion here.

I am not good a precedences for these operators. I suppose this is the following, right?

foo = do
  r <- newSTRef 0
  ((modifySTRef (+1) r >> lift empty) <|> return ())
  readSTRef r

[BebeSparkelSparkel]

This is not a finished example but is where I am using it

 86             (whileM
 87               (do
 88                 addDigit =<< charToDigit =<< index
 89                 inc
 90                 decExponent
 91                 return True
 92               <|> return False))

[josefs]

@andreasabel :

I am not good a precedences for these operators. I suppose this is the following, right?

Yes.

@BebeSparkelSparkel can you explain where in the do block the computation can fail and where you update references?

[BebeSparkelSparkel]

Both index and charToDigit can fail.

addDigit, inc, and decExponent all mutate.

Runs in the Maybe monad.

[josefs]

Thank for the example @BebeSparkelSparkel . That is indeed a valid usecase as there doesn't seem to be any potential failures after mutation.

[BebeSparkelSparkel]

@josefs Thanks for checking on this