Add Map.applyWithDefaults to satisfy Applicative laws #8
Conversation
test/Test/Data/Map.purs
Outdated
| -- Identity: (pure identity) <*> v = v | ||
| log "applyWithDefault abides applicative laws: Identity" | ||
| quickCheck \(TestMap x :: TestMap Int Int) -> | ||
| let out = M.applyWithDefault M.empty (Just identity) x Nothing |
There was a problem hiding this comment.
Maybe this Nothing should be any Maybe Int? It looks like it's important to the property that it's Nothing, but it isn't. I had to write down the equations to see that it made no difference. So, maybe it'll prevent someone else going through the same thing?
test/Test/Data/Map.purs
Outdated
| left2 = M.applyWithDefault left1 (fd <*> gd) h hd | ||
| right0 = M.applyWithDefault f fd g gd | ||
| right1 = M.applyWithDefault right0 (fd <*> gd) h hd | ||
| in left2 == right1 |
There was a problem hiding this comment.
Is this checking associativity? I haven't tried to look if it's an equivalent formulation, but it looks like the values are grouped the same way. Is this the associative grouping?
let left0 = M.applyWithDefault M.empty (Just compose) f fd
left1 = M.applyWithDefault left0 (Just compose <*> fd) g gd
left2 = M.applyWithDefault left1 (Just compose <*> fd <*> gd) h hd
right0 = M.applyWithDefault g gd h hd
right1 = M.applyWithDefault f fd right0 (gd <*> hd)
in left2 == right1I haven't checked that either.
There was a problem hiding this comment.
You're right, I completely messed this up 😅
| Map k i -> Maybe i -> | ||
| Map k o | ||
| applyWithDefault fns defFn vals defVal = | ||
| let build k = Tuple k <$> ((lookup k fns <|> defFn) <*> |
There was a problem hiding this comment.
Wouldn't it be possible to use union here, like the implementation in Data.Align? That way if we get a faster hedge union, this will improve for free too.
There was a problem hiding this comment.
Another thought: why not add this in the form of newtype MapWithDefault k a = MapWithDefault a (Map k a) with an Applicative instance? I think Conal has a Haskell library which does that.
There was a problem hiding this comment.
I'll have a think about the union, thanks.
Found the library you mentioned: https://hackage.haskell.org/package/total-map-0.0.6/docs/Data-TotalMap.html
I think both the type you propose and mine are useful for difference purposes.
With the one I proposed you get to model merges where one/both/neither have a default.
|
Should this PR be closed? It's been open for quite some time. |
|
We merged the |
|
Closing for inactivity, I'd also forgotten it existed. |
Inspired by @joneshf's comments here: purescript-contrib/purescript-these#21
This gives us a way of modelling an applicative map, for example something like:
Whether or not the above should live in this library is a different question (thoughts?), but the function is useful nonetheless.