Redesign #23

athanclark · 2018-12-15T22:29:45Z

An absolute ton went into this revision. Here is a brief summary:

Testing

There's a much more thorough test suite, first off - the TypedArray mechanism was most important - to prove that all generated values, for the different bit-caste sizes, operate exactly as expected for all operations, was a critical milestone.

Foreign Interface

The previous version relied on Uint8Arrays as the fundamental representation in JavaScript for most types - that is no longer the case; the actual JavaScript DataView, ArrayBuffer, and TypedArray types are the underpinning representation for the PureScript types.

Class-based API

Because of the commonality between APIs among the different TypedArray types, I took the liberty of designing it with classes and instances in-mind. The performance overhead should be negligible, especially when using rollup-plugin-purs.

There's also a classy interface for DataViews, to choose the getter and setters for each binary representation.

No more Data.ArrayBuffer.Show, No more String encodings

As @AlexaDeWit mentioned in #20, string encodings will be done with purescript-arraybuffer-codecs. The Show module used a foreign npm module, inspect I think, and it just felt gnarly, so I got rid of it.

Generators

MonadGen values live in Data.ArrayBuffer.*.Gen, for unit test generation.

Type-Level Kit

There's some extra mappings in Data.ArrayBuffer.ValueMapping, where byte length and application-level type casts are defined for each ArrayViewType. This was necessary to keep DataView and TypedArrays consistent at compile-time.

Value-Level Kit

The TypedArray interface at Data.ArrayBuffer.Typed provides all the tools you'd expect from an iterable structure - folds, maps, traversals, searching, querying, filtering, slicing, etc. are all in the form of Snoyman's mono-traversable.

That's about it! Please let me know what you think, any feedback is helpful.

anonymous foreign functions for stack traces

functions

this is not a correct implementation currently in DataView

the intended values that would be expected in javascript - negative encodings for proper Uint32 is invalid in this case. Because of this, Uint32s are represented in the positive domain of 64bit floats (js Number)

…r into redesign

athanclark · 2019-01-27T09:10:57Z

Oooh, I just had a tricky thought... should offsets and byte length be represented as UInts? Because they can't be negative...? It feels against-the-grain though :\

jacereda · 2019-01-27T09:14:20Z

In the manual bounds-checking they're interpreted as UInts... If there was compiler support so that literals could become UInts automagically it would probably be the way to go, but as it's now I guess it will just be burdensome.

athanclark · 2019-01-27T09:16:41Z

Yeah I was thinking the same. We gotta get some kind of purescript v0.13 wish list going or something

jacereda · 2019-01-27T21:38:38Z

src/Data/ArrayBuffer/Typed.purs

+
+-- | Returns a new typed array view of the same buffer, beginning at the index and ending at the second.
+-- |
+-- | **Note**: there is really peculiar behavior with `subArray` - if the first offset argument is omitted, or


Is this behaviour for subarray documented somewhere?

I wish. I can't find it in the ECMA spec, but it's pretty obvious if you try it in psci. I'm not sure how to find a doc on this either

jacereda · 2019-01-27T22:12:02Z

src/Data/ArrayBuffer/Typed.purs

+foreign import joinImpl :: forall a. Fn2 (ArrayView a) String String
+
+-- | Prints array to a delimiter-separated string - see [MDN's spec](https://developer.mozilla.org/en-US/docs/Web/JavaScript/Reference/Global_Objects/TypedArray/join) for details.
+toString' :: forall a. ArrayView a -> String -> String


join? toStringSep? toStringSep? toStringIntercalate?

I'll go with join :) sorry

jacereda · 2019-01-27T22:35:24Z

I've pushed more changes to the v10 branch making more functions effectful. I guess some more would need to be converted, particularly I guess at should also be effectful, right?

athanclark · 2019-01-28T05:49:09Z

Definitely, at is just Ref.read in a way

jacereda · 2019-01-28T22:58:29Z

I've unified the signatures for slice/subArray/fill to mirror Array/Node.Buffer APIs in the v10 branch. Also, I can't reproduce the behaviour you specify for subArray and removed the note. I seem to remember fixing one of the subArray tests, maybe that was the problem (or maybe you found a broken node version?).

athanclark · 2019-01-29T02:20:23Z

:s shoot I don't think so, but I wouldn't doubt it. iirc I was using nvm v8.12.0, but either way I'm glad the mystery is cleared up.

But awesome, I definitely should have thought of that - the API mirroring.

jacereda · 2019-01-30T21:58:34Z

src/Data/ArrayBuffer/Typed.js

-exports.asInt8Array = function(v) {
-  return new Int8Array(v.buffer, v.byteOffset, v.byteLength);
-}
+exports.polyFill = function polyFill () {


Why do you export this? Would it be problematic to just ejecute the polyfill directly?

I don't think it would, but I was just trying to be on the safe side - it technically causes an effect, and although I know ffi files only get executed once, I just felt like someone might want finer granularity of control. But honestly I think it would be fine to just execute - I pulled it from MDN.

jacereda · 2019-02-01T13:44:19Z

I've added some unsafe instances (Show/Eq/Ord/Semigroup/Monoid), but I don't think these belong in this repo, I guess they could be somewhere else. See #24

athanclark · 2019-02-01T21:36:26Z

Beautiful! I think it looks great. The newtype was exactly what I was going to suggest.

jacereda · 2019-02-01T22:22:01Z

I've merged to master but will hold the release until I get comments from @AlexaDeWit regarding the new API.

On a different subject, do you think TypedArrays could have a Foldable instance? I've been unable to write it...

athanclark · 2019-02-01T22:58:05Z

Awesome, that will be great. I think a cereal compatible library is next on my plate after that. But technically, no, because it's not a container over "any type a"; only the type that it's related to - like a Uint or whathaveyou; it's got that extra constraint predicate.

jacereda · 2019-02-01T23:07:03Z

But somehow it's a container for "some type a". Is it possible to specify that a could be any of [UInt,Int,Float32,...]?

athanclark · 2019-02-02T02:28:25Z

No, because an instance of Foldable only exposes it's container type f for an instance constraint. Because Foldable.foldr adds an additional forall a. in it's member definition - there's no way to link that skolemized a with the f at the top level.

athanclark · 2019-02-02T02:29:59Z

It's almost the same idea for why a Set can't be a Functor - because we have no way to say that the member elements must also have an Ord constraint.

jacereda · 2019-02-02T08:52:17Z

Awesome, that will be great. I think a cereal compatible library is next on my plate after that. But technically, no, because it's not a container over "any type a"; only the type that it's related to - like a Uint or whathaveyou; it's got that extra constraint predicate.

Not cereal compatible, but some time ago I started https://github.com/jacereda/purescript-binary and I guess I'll update it at some point to work with this.

athanclark · 2019-02-05T22:23:03Z

Ping - hey sorry to pester. Were you intending on publishing the new changes in master? I know the v9.0.0 version doesn't include the Gen modules and all of that. I'm just trying to publish some other libraries that rely on this one :) thanks for your help!

jacereda · 2019-02-05T22:29:32Z

I was just waiting to see if @AlexaDeWit had any comment regarding the new API. In any case, I'll try to release on Thursday.

athanclark · 2019-02-06T00:21:54Z

Awesome, thank you!

AlexaDeWit · 2019-02-06T00:41:50Z

Sorry about that, I didn't mean to ignore things. I've just not been following this super closely, as legacy versions will work for me, and I kind of trust you two to come up with something coherent so that when I get around to making my codec lib it should be fine

athanclark and others added 30 commits December 4, 2018 10:29

ripping out untrue functionality of arraybuffers

53c7b23

rename to stay true to implementation

97a49c6

converting to Effect.Uncurried, using exception catching, naming

d650148

anonymous foreign functions for stack traces

module documentation

a4c7cd0

preparing for major redesign of typed interface - supporting all

addf664

functions

polyfill

5200641

module doc

5d49921

using new as binder - also, remainder function

75a91e6

copyWithin

827b29e

all, any ~ every, some

7a3435f

fill semantics

ba3f089

toString

b203cd9

subarray

e2b70aa

in-place sort

16f384c

slice

95a79cb

moving set around a little

b281b9d

reverse in-place

f8ae9ad

typed map

be5e28b

joining an array

f2be9a4

filter

fa53dca

toArray typed properly

a3da940

folds

588284d

elem, need to do find and indexing

3b246b9

better arraybuffer slice, more stuff

4759fe4

the uint library stores values greater than 2^31 as a negative integer -

aa789b7

this is not a correct implementation currently in DataView

index of

d0c4038

more instances

2ffbb00

honestly I forgot about floating points

3c7066e

the binary serialization of javascript Numbers for TypedArrays follows

fd29b0b

the intended values that would be expected in javascript - negative encodings for proper Uint32 is invalid in this case. Because of this, Uint32s are represented in the positive domain of 64bit floats (js Number)

Remove text encoding dependency

ea66b5f

jacereda and others added 5 commits January 25, 2019 21:27

Make empty effectful

23c5135

Restore specific setters/getters in the API, changed implementation

86fd6e8

Merge branch 'v10' of git://github.com/jacereda/purescript-arraybuffe…

e06a9a2

…r into redesign

fixing tests

0fa0d7a

restricted imports

45f310a

jacereda reviewed Jan 27, 2019

View reviewed changes

jacereda reviewed Jan 30, 2019

View reviewed changes

jacereda merged commit 45f310a into purescript-contrib:master Feb 1, 2019

jamesdbrock mentioned this pull request Jun 19, 2021

typelevel dependency #29

Closed

jamesdbrock mentioned this pull request Sep 8, 2021

Polyfill is not defined when using spago bundle-app #34

Closed

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

Redesign #23

Redesign #23

athanclark commented Dec 15, 2018

athanclark commented Jan 27, 2019

jacereda commented Jan 27, 2019

athanclark commented Jan 27, 2019

jacereda Jan 27, 2019

athanclark Jan 28, 2019

jacereda Jan 27, 2019

athanclark Jan 28, 2019

jacereda commented Jan 27, 2019

athanclark commented Jan 28, 2019

jacereda commented Jan 28, 2019

athanclark commented Jan 29, 2019

jacereda Jan 30, 2019

athanclark Jan 31, 2019

jacereda commented Feb 1, 2019

athanclark commented Feb 1, 2019

jacereda commented Feb 1, 2019

athanclark commented Feb 1, 2019

jacereda commented Feb 1, 2019

athanclark commented Feb 2, 2019

athanclark commented Feb 2, 2019

jacereda commented Feb 2, 2019

athanclark commented Feb 5, 2019

jacereda commented Feb 5, 2019

athanclark commented Feb 6, 2019

AlexaDeWit commented Feb 6, 2019

Redesign #23

Redesign #23

Conversation

athanclark commented Dec 15, 2018

Testing

Foreign Interface

Class-based API

No more Data.ArrayBuffer.Show, No more String encodings

Generators

Type-Level Kit

Value-Level Kit

athanclark commented Jan 27, 2019

jacereda commented Jan 27, 2019

athanclark commented Jan 27, 2019

jacereda Jan 27, 2019

Choose a reason for hiding this comment

athanclark Jan 28, 2019

Choose a reason for hiding this comment

jacereda Jan 27, 2019

Choose a reason for hiding this comment

athanclark Jan 28, 2019

Choose a reason for hiding this comment

jacereda commented Jan 27, 2019

athanclark commented Jan 28, 2019

jacereda commented Jan 28, 2019

athanclark commented Jan 29, 2019

jacereda Jan 30, 2019

Choose a reason for hiding this comment

athanclark Jan 31, 2019

Choose a reason for hiding this comment

jacereda commented Feb 1, 2019

athanclark commented Feb 1, 2019

jacereda commented Feb 1, 2019

athanclark commented Feb 1, 2019

jacereda commented Feb 1, 2019

athanclark commented Feb 2, 2019

athanclark commented Feb 2, 2019

jacereda commented Feb 2, 2019

athanclark commented Feb 5, 2019

jacereda commented Feb 5, 2019

athanclark commented Feb 6, 2019

AlexaDeWit commented Feb 6, 2019