Revert #409?

[saschanaz]

  AudioParam setValueCurveAtTime (sequence<float> values, double startTime, double duration);

dictionary WaveShaperOptions : AudioNodeOptions {
  sequence<float> curve;
  OverSampleType oversample = "none";
};

@iccir suggested in #408 that we should use Float32Array citing an older spec, but the latest one says it's sequence<float>.

#409 also changed VRLayer and it also has sequence<float>. (It's now VRLayerInit.)

So I suggest that we revert #409, what do you think? @mhegazy

[iccir]

I likely pulled up the W3C specification or an old webaudio.github.com one by mistake. This changed in WebAudio/web-audio-api#1189

My original issue was a warning when passing a Float32Array as an argument of setValueCurveAtTime(). From my understanding, this is still allowed by the latest spec ("This is
a backward-compatible change since Float32Array is a sequence…").

Should sequence<float> map to a union type, like number[] | Float32Array?

[saschanaz]

Should sequence map to a union type, like number[] | Float32Array?

Sequences cannot be used for attributes, so it may make sense.

[mhegazy]

@saschanaz what about the other APIs that changed?

Maybe the right fix is to revert the change and add a specific override for setValueCurveAtTime.

Sorry for making the changes without checking the updated spec

[saschanaz]

what about the other APIs that changed?

I think that's all.

Full diff:

--- a/baselines/dom.generated.d.ts
+++ b/baselines/dom.generated.d.ts
@@ -1033,8 +1033,8 @@ interface PeriodicWaveConstraints {
 }

 interface PeriodicWaveOptions extends PeriodicWaveConstraints {
-    imag?: Float32Array;
-    real?: Float32Array;
+    imag?: number[];
+    real?: number[];
 }

 interface PointerEventInit extends MouseEventInit {
@@ -1569,8 +1569,8 @@ interface VRDisplayEventInit extends EventInit {
 }

 interface VRLayer {
-    leftBounds?: Float32Array | null;
-    rightBounds?: Float32Array | null;
+    leftBounds?: number[] | null;
+    rightBounds?: number[] | null;
     source?: HTMLCanvasElement | null;
 }

@@ -1581,7 +1581,7 @@ interface VRStageParameters {
 }

 interface WaveShaperOptions extends AudioNodeOptions {
-    curve?: Float32Array;
+    curve?: number[];
     oversample?: OverSampleType;
 }

@@ -1984,7 +1984,7 @@ interface AudioParam {
     linearRampToValueAtTime(value: number, endTime: number): AudioParam;
     setTargetAtTime(target: number, startTime: number, timeConstant: number): AudioParam;
     setValueAtTime(value: number, startTime: number): AudioParam;
-    setValueCurveAtTime(values: Float32Array, startTime: number, duration: number): AudioParam;
+    setValueCurveAtTime(values: number[], startTime: number, duration: number): AudioParam;
 }

 declare var AudioParam: {

PeriodicWaveOptions is also from Web Audio.

dictionary PeriodicWaveOptions : PeriodicWaveConstraints {
  sequence<float> real;
  sequence<float> imag;
};

[iccir]

That reverts it, but the original issue remains: setValueCurveAtTime needs to be able to take a Float32Array without warning.

I'm also not sure what the official policy is regarding specifications that haven't been adopted by all browsers. Firefox and WebKit both require Float32Array and throw an exception on other types.

[saschanaz]

Another general solution is creating new WebIDL-compatible Sequence<T> type.

The spec says iterable objects can be casted as IDL sequence type, so we may do:

type Sequence<T> = T[] | { [Symbol.iterator](): IterableIterator<T> };

// all of the followings should be allowed

declare function receiveNumberSequence(sequence: Sequence<number>): void;
receiveNumberSequence([0]);
receiveNumberSequence(new Set([0]));
receiveNumberSequence(new Uint8Array([0]));
receiveNumberSequence(new Float32Array([0]));

declare function receiveStringMap(map: Sequence<[string, string]>): void;
receiveStringMap(new Map([["abc", "bcd"]]));
receiveStringMap(new URLSearchParams());
receiveStringMap(new Headers());

I have no idea how to make this ES5-compatible though. microsoft/TypeScript#13031

[mhegazy]

ES5 vs ES6 is the issue here..

[mhegazy]

So every use of Sequence<float> should be number[] | Float32Array ?

Looks like all the changes in #409 were in input positions, so the use of union should be fine.

[saschanaz]

So every use of Sequence should be number[] | Float32Array ?

For now, yes.

Looks like all the changes in #409 were in input positions

Actually it's guaranteed by the WIDL spec 😁 Ah, methods can return sequences.

[iccir]

Is number[] | Float32Array too narrow? Isn't sequence<float> any object with a length that returns float-like values via its subscript operator?

Chrome appears to accept Float64Array and Int16Array values for setValueCurveAtTime(), but I'm not sure if this is what the spec intended.

[saschanaz]

That's the intended behavior, but there is no good general way in TS to allow it. Explicitly allowing every possible interfaces will be too exhaustive.

[iccir]

Makes sense, thanks! number[] | Float32Array (and maybe Float64Array?) should cover most use cases.

[saschanaz]

ArrayBufferView may be the best option for those typed arrays.

[iccir]

ArrayBufferView includes DataView, which I believe isn't allowed in sequence<float> due to the lack of [] and length.