foldMap, foldMap' for PrimArray by chessai · Pull Request #186 · haskell/primitive

chessai · 2018-07-15T00:06:20Z

No description provided.

andrewthad · 2018-07-16T17:53:32Z

I'm good with this. If there aren't any objection in the next few days, I'll merge this.

RyanGlScott · 2018-07-16T18:12:17Z

No objections here, although these deserve to be mentioned in the changelog.

chessai · 2018-07-16T19:23:05Z

Sorry I keep forgetting that. I'll add it in.

chessai · 2018-09-17T18:47:31Z

Is this good to be merged? The travis failure is unrelated.

treeowl · 2018-09-17T19:00:17Z

+foldMapPrimArray f = foldrPrimArray (\a acc -> acc `mappend` f a) mempty
+
+-- | Strictly map each element of the primitive array to a monoid, and combine the results.
+foldMapPrimArray' :: forall a m. (Prim a, Monoid m) => (a -> m) -> PrimArray a -> m


The documentation is insufficient. It should explain which way the combination associates. Also, it's notably not (necessarily) strict in the function results. The documentation should reflect this and there should be a comment explaining that decision.

treeowl · 2018-09-17T19:02:51Z

 sizeofPrimArray (PrimArray arr#) = I# (quotInt# (sizeofByteArray# arr#) (sizeOf# (undefined :: a)))

+-- | Lazily map each element of the primitive array to a monoid, and combine the results.
+foldMapPrimArray :: forall a m. (Prim a, Monoid m) => (a -> m) -> PrimArray a -> m


The documentation is insufficient. It should explain how the combination associates. In fact, I think we should almost surely offer more than one version: associating to the left, associating to the right, and (for certain tree-like monoids) associating in a balanced fashion are all potentially useful.

Tree-like or in some cases numerical. Balanced association can reduce floating point errors somewhat, for example.

treeowl · 2018-09-17T19:03:40Z

Sorry for the delayed review

chessai · 2018-09-17T19:06:00Z

That is okay, this is valuable feedback. I will make these changes shortly.

…rimarray

… blue fish

chessai · 2018-09-17T20:50:47Z

most recent commit does not add balanced foldMap. I think I'd prefer for that to be in a separate PR, as it's a little more complicated.

treeowl · 2018-09-17T22:57:20Z


+-- | Map each element of the primitive array to a monoid, and combine the results.
+--   The combination is right-associated, and the accumulation is lazy.
+foldrMapPrimArray :: forall a m. (Prim a, Monoid m) => (a -> m) -> PrimArray a -> m


I'd rather see the Map before the r.

treeowl · 2018-09-17T23:00:51Z

+
+-- | Map each element of the primitive array to a monoid, and combine the results.
+--   The combination is right-associated, and the accumulation is strict, though
+--   this function is not necessarily strict in its result.


"This function is not necessarily strict in its result" doesn't mean anything whatsoever to me. What I meant needed explanation is that at each step, we force the accumulator value, but we don't force the value of f a. So if mappend is lazy in its first argument, f a will not be evaluated. This is probably the correct behavior, but needs better explanation.

treeowl · 2018-09-17T23:01:21Z

 ## Changes in version 0.6.4.1

- * Add instances for the following newtypes from `base`:
+ * Add `foldMapPrimArray` and `foldMapPrimArray'`


This needs to be updated now that you've added more functions.

…l in changelog

…ct some documentation on strict folds

chessai · 2018-09-18T01:42:44Z

requested changes attempted, @treeowl

treeowl

Much better. Just a couple more little changes.

treeowl · 2018-09-18T02:03:44Z

Both examples are supposed to show associativity. I think they can be much briefer, TBH. Just give a three-element example application and show how it unfolds as a single expression.

…

On Mon, Sep 17, 2018, 9:54 PM chessai ***@***.***> wrote: ***@***.**** commented on this pull request. ------------------------------ In Data/Primitive/PrimArray.hs <#186 (comment)>: > +-- +-- @mysum = 'Data.Monoid.getSum' '$' ('Data.Monoid.Sum' 1 '<>' ('Data.Monoid.Sum' 2 '<>' ('Data.Monoid.Sum' 3 '<>' 'Data.Monoid.Sum' 4)))@ +-- +-- @mysum = 'Data.Monoid.getSum' '$' ('Data.Monoid.Sum' 1 '<>' ('Data.Monoid.Sum' 2 '<>' ('Data.Monoid.Sum' 7)))@ +-- +-- @mysum = 'Data.Monoid.getSum' '$' ('Data.Monoid.Sum' 1 '<>' ('Data.Monoid.Sum' 9)@ +-- +-- @mysum = 'Data.Monoid.getSum' '$' 'Data.Monoid.Sum' 10@ +-- +-- @mysum = 10@ +foldMapRPrimArray :: forall a m. (Prim a, Monoid m) => (a -> m) -> PrimArray a -> m +{-# INLINE foldMapRPrimArray #-} +foldMapRPrimArray f = foldrPrimArray (\a acc -> f a `mappend` acc) mempty + +-- | Map each element of the primitive array to a monoid, and combine the results. +-- The combination is left-associated, and the accumulation is lazy. Clarification: Are you asking to provide an example for foldMapLPrimArray, to show left-associativity (for this particular lazy fold) — You are receiving this because you were mentioned. Reply to this email directly, view it on GitHub <#186 (comment)>, or mute the thread <https://github.com/notifications/unsubscribe-auth/ABzi_V_eja4orM-34lJJs1TwInJE6ERrks5ucFJVgaJpZM4VQDt8> .

chessai · 2018-09-18T02:10:30Z

Both examples are supposed to show associativity. I think they can be much
briefer, TBH. Just give a three-element example application and show how it
unfolds as a single expression.

By 'briefer', do you mean less 'tutorial-like', to assume that the example is illustrative enough sans written explanation?

treeowl · 2018-09-18T02:14:07Z

By 'briefer', do you mean less 'tutorial-like', to assume that the example is illustrative enough sans written explanation?

Right. This is primitive, not base or containers. Users are expected to know what they're doing. But language is slippery, so "associates to the left" might well evoke opposite meanings in the minds of two different people.

chessai · 2018-09-18T02:17:28Z

Right. This is primitive, not base or containers. Users are expected to know what they're doing. But language is slippery, so "associates to the left" might well evoke opposite meanings in the minds of two different people.

Makes sense - understood

treeowl · 2018-09-18T02:29:09Z

Aha! I've changed my mind :). I believe we should force each function value in the strict versions. Laziness here is kind of silly, but more to the point, I believe the laziness reveals semantic differences in fold associativity that are only supposed to be performance differences.

…

On Mon, Sep 17, 2018, 10:17 PM chessai ***@***.***> wrote: Right. This is primitive, not base or containers. Users are expected to know what they're doing. But language is slippery, so "associates to the left" might well evoke opposite meanings in the minds of two different people. Makes sense - understood — You are receiving this because you were mentioned. Reply to this email directly, view it on GitHub <#186 (comment)>, or mute the thread <https://github.com/notifications/unsubscribe-auth/ABzi_a3CzXzv1s-TcI2ZeePIIZX1mptbks5ucFe5gaJpZM4VQDt8> .

chessai · 2018-09-18T02:53:05Z

@treeowl Examples are now simplified and function values are being forced

treeowl · 2018-09-18T02:54:37Z

Humor me and check whether what I said about forcing is really true. I was going off intuition.

…

On Mon, Sep 17, 2018, 10:53 PM chessai ***@***.***> wrote: @treeowl <https://github.com/treeowl> Examples are now simplified and function values are being forced — You are receiving this because you were mentioned. Reply to this email directly, view it on GitHub <#186 (comment)>, or mute the thread <https://github.com/notifications/unsubscribe-auth/ABzi_VlFO4Cj6buIYDs2COP8-OJAPoAPks5ucGASgaJpZM4VQDt8> .

chessai · 2018-09-18T14:16:50Z

OK

andrewthad · 2020-06-12T19:13:55Z

I like the stuff in the PR, although I dislike the naming convention that @treeowl suggested. I actually prefer foldrMapPrimArray to foldMapRPrimArray. Anyone else got any thoughts on the color of the bikeshed?

Boarders · 2020-11-24T21:46:37Z

I would still like this and also a foldl for ByteArray at some point. Happy to help to make that happen in whatever way I can.

andrewthad · 2020-11-25T13:39:14Z

I'll change the names and merge this. @Boarders Open a separate PR for a foldl on ByteArray. You should be able to nearly copy and paste the foldl on PrimArray.

foldMap, foldMap'

ff7f68a

update changelog

6948536

treeowl requested changes Sep 17, 2018

View reviewed changes

Comment thread Data/Primitive/PrimArray.hs Outdated

chessai added 3 commits September 17, 2018 15:50

Merge branch 'master' of github.com:haskell/primitive into foldable-p…

853b996

…rimarray

fix ordering of mappend in foldMapPrimArray

af41b39

have 4 foldMaps; two right, two left, two lazy, two strict, red fish,…

0bb9dd6

… blue fish

bump quickcheck-classes for travis happiness

8aaa9d1

treeowl requested changes Sep 17, 2018

View reviewed changes

chessai added 2 commits September 17, 2018 21:02

change fold{dir}MapPrimArray to foldMap{dir}PrimArray, and mention al…

96eb0cc

…l in changelog

provide an example to demonstrate association and accumulation, corre…

534a20c

…ct some documentation on strict folds

treeowl reviewed Sep 18, 2018

View reviewed changes

Comment thread Data/Primitive/PrimArray.hs

Comment thread Data/Primitive/PrimArray.hs

chessai added 2 commits September 17, 2018 22:43

simplify examples for foldMapRPrimArray foldMapLPrimArray

6f0c84e

make strict foldMaps on PrimArray force function value at each step

d4eb3d0

Conversation

chessai commented Jul 15, 2018

Uh oh!

andrewthad commented Jul 16, 2018

Uh oh!

RyanGlScott commented Jul 16, 2018

Uh oh!

chessai commented Jul 16, 2018

Uh oh!

chessai commented Sep 17, 2018 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

treeowl Sep 17, 2018

Choose a reason for hiding this comment

Uh oh!

treeowl Sep 17, 2018

Choose a reason for hiding this comment

Uh oh!

treeowl Sep 17, 2018

Choose a reason for hiding this comment

Uh oh!

treeowl commented Sep 17, 2018

Uh oh!

chessai commented Sep 17, 2018

Uh oh!

Uh oh!

chessai commented Sep 17, 2018

Uh oh!

treeowl Sep 17, 2018

Choose a reason for hiding this comment

Uh oh!

Uh oh!

treeowl Sep 17, 2018

Choose a reason for hiding this comment

Uh oh!

treeowl Sep 17, 2018

Choose a reason for hiding this comment

Uh oh!

chessai commented Sep 18, 2018

Uh oh!

treeowl left a comment

Choose a reason for hiding this comment

Uh oh!

Uh oh!

Uh oh!

treeowl commented Sep 18, 2018 via email

Uh oh!

chessai commented Sep 18, 2018

Uh oh!

treeowl commented Sep 18, 2018

Uh oh!

chessai commented Sep 18, 2018

Uh oh!

treeowl commented Sep 18, 2018 via email

Uh oh!

chessai commented Sep 18, 2018

Uh oh!

treeowl commented Sep 18, 2018 via email

Uh oh!

chessai commented Sep 18, 2018

Uh oh!

andrewthad commented Jun 12, 2020

Uh oh!

Boarders commented Nov 24, 2020

Uh oh!

andrewthad commented Nov 25, 2020

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

5 participants

chessai commented Sep 17, 2018 •

edited

Loading