augustss / orthotope Goto Github PK

These two take their type argument (the "shape" and the payload type) in different order:

https://hackage.haskell.org/package/orthotope-0.1.2.0/docs/Data-Array-RankedS.html#v:fromList

https://hackage.haskell.org/package/orthotope-0.1.2.0/docs/Data-Array-ShapedS.html#v:fromList

I'm fine with either of these and I will standardise in horde-ad on the one chosen in orthotope. The chosen order is not a big deal thanks to the @_ idiom that permits skipping type arguments when applying, but having to remember which operation uses which order is problematic and it obfuscates error messages.

I'm seeing thunks created in function

https://hackage.haskell.org/package/orthotope-0.1.2.0/docs/src/Data.Array.Internal.RankedS.html#scalar

most probably due to

https://hackage.haskell.org/package/orthotope-0.1.2.0/docs/src/Data.Array.Internal.RankedG.html#Array

not having strict fields. For context, note how the following has some strict fields:

https://hackage.haskell.org/package/orthotope-0.1.2.0/docs/src/Data.Array.Internal.html#T

Shall we make both or the second field of data Array (n :: Nat) v a = A ShapeL (T v a) strict? I'm seeing

Data/Array/Internal/DynamicG.hs:data Array v a = A !ShapeL !(T v a)

so probably the policy is that shapes are strict and strides are lazy, which seems to match my use case very well.

How about declaring a global StrictData extension in the data file and mark (with ~) only the fields that should not be strict?

I'm ready to prepare a PR, but it's easy to make a design mistake here, so I can't start before brainstorming.