Spec glrc.drawingBufferStorage. #3222

kdashg · 2021-03-03T21:14:48Z

An alternative to #3220.

kenrussell

This is pleasingly small and gives applications the chance to feature detect. Personally I like this approach. Other opinions / feedback?

A few questions.

specs/latest/1.0/index.html

kenrussell · 2021-03-04T00:21:10Z

specs/latest/1.0/index.html

@@ -1671,6 +1703,7 @@ <h3><a name="WEBGLRENDERINGCONTEXT">The WebGL context</a></h3>

    const GLenum RGBA4                          = 0x8056;
    const GLenum RGB5_A1                        = 0x8057;
+    const GLenum RGBA8                          = 0x8058;


Some WebGL conformance tests like conformance/context/constants-and-properties.html and conformance2/context/constants-and-properties-2.html (and maybe others) will have to be updated for this change.

Yep, that's on the list of test changes that we'll need.

@ccameron-chromium points out in https://chromium-review.googlesource.com/c/chromium/src/+/4887172 that it's necessary to add RGB8 to WebGL 1.0 too, because {alpha:false} contexts use this as drawingBufferFormat even though drawingBufferStorage can't be called for such contexts afterward.

kenrussell · 2021-03-04T00:21:22Z

specs/latest/1.0/index.html

@@ -1728,6 +1761,8 @@ <h3><a name="WEBGLRENDERINGCONTEXT">The WebGL context</a></h3>
    sequence&lt;DOMString&gt;? getSupportedExtensions();
    object? getExtension(DOMString name);

+    undefined backbufferStorage(GLenum sizedFormat, unsigned long width, unsigned long height);


Similar for this new method.

kenrussell · 2021-03-04T00:23:29Z

specs/latest/1.0/index.html

+            similar to <code>renderbufferStorage</code> as used on Renderbuffers.
+
+            Clearing behavior is equivalent to setting <code>HTMLCanvasElement.width</code> and <code>HTMLCanvasElement.height</code>, followed by changing the backbuffer format.
+            This method respects <code>WebGLContextAttributes.antialias</code>.


How can the user know (aside from calling gl.getError, which is discouraged) that the setting actually took effect? Should we add a new read-only property alongside drawingBufferWidth and drawingBufferHeight?

You can't tell today for textures or renderbuffers, so I'm not super worried about going above-and-beyond for backbuffers.

kenrussell · 2021-03-04T00:34:00Z

@lexaknyazev @ccameron-chromium please review this version of this functionality and see how you like it compared to #3220 . (I don't have the ability to explicitly add you as reviewers.) Thanks.

kainino0x · 2021-03-04T00:41:32Z

I haven't thought on this too deeply, but it seems like quite a nice approach. I initially thought the dynamicness would be scary - but the semantic I imagine is simply "throw out the drawing buffer you allocated for this frame (if any) and replace it with new blank one with this format". As long as implementations can get that texture to the compositor, that's great.

nit: The "backbuffer" naming makes the "back"ness central when it's not necessarily relevant (though with the semantic above, maybe it is, since you couldn't have that semantic on a frontbuffer...?) Could be "default framebuffer", "default framebuffer color attachment", "drawing buffer", ?

kdashg · 2021-03-04T00:48:16Z

I'm definitely being cheeky and saying "backbuffer" when I mean "default framebuffer", but "default framebuffer" is relatively poorly understood, I believe. I can s/backbuffer/default framebuffer/ though, it's technically more correct.

kdashg · 2021-03-04T00:52:44Z

Actually, since I would like to avoid defaultFramebufferStorage, I would prefer to keep backbufferStorage.
It certainly behaves as the backbuffer as per spec (for dom atomicity and such) and I think this is probably the least bad name I can think of. Does anyone else have ideas?

I like this as-is, because I feel like practically 100% of people will understand what's going on here, even without a good background in WebGL details.

lexaknyazev · 2021-03-04T06:46:14Z

Looks great overall. Couple thoughts:

The WebGL backbuffer format complements the color space set during canvas creation. Are all possible combinations valid? For example, will future BT.2020 canvas work with RGBA8 backbuffers?
It is not immediately clear to me how respecifying backbuffer's width affects HTMLCanvasElement.width, CSS width, and WebGLRenderingContext.drawingBufferWidth.

ccameron-chromium · 2021-03-04T09:04:39Z

The WebGL backbuffer format complements the color space set during canvas creation. Are all possible combinations valid? For example, will future BT.2020 canvas work with RGBA8 backbuffers?

I think all combinations should be valid. They are all equally feasible technically, but have varying degrees of usefulness. Most APIs (Metal, Vulkan, D3D) allow all combinations.

Would we want to include the color space as an argument to backbufferStorage, too? (I'm bringing the question up for completeness -- I have no opinion on that topic. For WebGPU, the plan is to add color space to GPUSwapChainDescriptor, which is vaguely analogous. Should we leave open the possibility of adding it as an optional argument?)

kdashg · 2021-03-04T18:45:22Z

TBH moving all the way to a WebGLSwapChain sounds better. I'll see what that looks like.
It'd be fine if colorspace was fixed, but I agree that it's a little weird to have one fixed and not the other.

kdashg · 2021-03-04T19:25:29Z

And +1 on keeping colorspace and format orthogonal, even if some combos are bad ideas. (rec2020 rgba8)

kainino0x · 2021-03-06T01:25:42Z

TBH moving all the way to a WebGLSwapChain sounds better. I'll see what that looks like.

Moving to WebGLSwapChain would be extremely cool, but I worry it would make the road to implementation a lot longer. Seems ideal to do things incrementally: design something like backbufferStorage() now that can still be used in a future with a WebGLSwapChain and maybe later even headless WebGL contexts.

On the topic of WebGLSwapChain, I bet WebGPU's GPUCanvasContext could serve dual purpose. Note the GPUDevice isn't bound to the GPUCanvasContext until GPUCanvasContext.configureSwapChain.

interface GPUCanvasContext {
    GPUSwapChain configureSwapChain(GPUDevice device, GPUSwapChainDescriptor descriptor);

    WebGLSwapChain configureSwapChain(WebGLRenderingContext context, GPUSwapChainDescriptor descriptor);
    // or
    WebGLFramebuffer configureSwapChain(WebGLRenderingContext context, GPUSwapChainDescriptor descriptor);
};

or something.

kenrussell · 2021-03-09T23:00:05Z

Thanks @kainino0x for your input.

Per discussion today with @jdashg and @ccameron-chromium - we're in agreement that adding a swap chain concept to WebGL at this late stage in its evolution is too much to add.

Here's the current plan:

Add drawingBufferFormat read-only attribute to WebGLRenderingContextBase
Add drawingBufferStorage(format, width, height) method to WebGLRenderingContextBase, which allows setting it

Still discussing how colorSpace will be handled. Since the drawing buffer's format is changeable, colorSpace really needs to be changeable at run time, too, rather than being specified during context creation time.

RafaelCintron · 2021-03-13T00:03:22Z

specs/latest/1.0/index.html

+            Respecify the size and format of the backbuffer,
+            similar to <code>renderbufferStorage</code> as used on Renderbuffers.
+
+            Clearing behavior is equivalent to setting <code>HTMLCanvasElement.width</code> and <code>HTMLCanvasElement.height</code>, followed by changing the backbuffer format.


The spec talks about clearing behavior here.

I presume this means that after the developer calls backbufferStorage, subsequently querying HTMLCanvasElement.width, HTMLCanvasElement.height, drawingBufferWidth and drawingBufferHeight will return the values passed to `backbufferStorage'?

Yes, modulo the same limitations the browser already enforces. Recall that drawingBufferWidth and drawingBufferHeight may be clamped due to platform limitations. The intent here is to make this new API behave exactly the same as resizing the HTMLCanvasElement or OffscreenCanvas, to avoid adding yet another behavior to the web platform.

RafaelCintron · 2021-03-13T00:05:19Z

specs/latest/1.0/index.html

+            Respecify the size and format of the backbuffer,
+            similar to <code>renderbufferStorage</code> as used on Renderbuffers.
+
+            Clearing behavior is equivalent to setting <code>HTMLCanvasElement.width</code> and <code>HTMLCanvasElement.height</code>, followed by changing the backbuffer format.


If the user first calls backBufferStorage and subsequently sets the HTMLCanvasElement.width and HTMLCanvasElement.height attributes, will the default framebuffer be resized to the last format passed as the sizedFormat parameter? Or will the default framebuffer be resized to whatever the default format the user agent decided when it first allocated the default framebuffer? The former seem preferable but I don't have a strong opinion.

It'll use the last format passed in to backBufferStorage - which per the last WebGL WG meeting will be renamed to drawingBufferStorage in the next iteration of this proposal.

…rFormat.

kenrussell

Thanks Jeff for putting together and revising this; it looks good to me!

I'm really excited about trying this out! How do you think we should move forward? Land this and put implementations behind flags, for example?

kenrussell · 2021-03-25T00:55:13Z

specs/latest/1.0/index.html

+                </p>
+                <p>
+                    In WebGL 1.0, <code>RGBA16F</code> requires the extension <code>EXT_color_buffer_half_float<code>.
+                    In WebGL 2.0, <code>RGBA16F</code> requires the extension <code>EXT_color_buffer_float<code>.


EXT_color_buffer_half_float's legal in WebGL 2.0 contexts now - to support iOS devices specifically. It would be better to say that WebGL 2.0 requires either EXT_color_buffer_float or EXT_color_buffer_half_float.

kenrussell · 2021-03-25T00:58:23Z

specs/latest/1.0/index.html

+        <dd>
+            The current effective format of the drawing buffer.
+            Initially <code>RGBA8</code> or <code>RGB8</code> for <code>alpha:true</code> and <code>alpha:false</code> respectively.
+            The actual height of the drawing buffer. May be different from the


This text and below's accidentally duplicated.

greggman · 2022-02-08T21:52:27Z

how do does this work with antialias? Does it ignore it? Up to browser? If up to browser can the browser switch from antialias true to false?

kdashg · 2022-02-09T01:07:16Z

Good question. Fortunately, antialias (like depth and stencil) is fixed after context creation, so the logic would be similar to e.g. canvas.width = x, where if resize failed, and no satisfactory backbuffer could be created, our failure case of last resort is context-loss.

ccameron-chromium · 2023-12-18T12:54:35Z

Is there anything more that we need to do in order to get this merged?

kenrussell · 2023-12-20T19:54:37Z

@kdashg told me on chat it was OK if I merge this after final cleanup, so doing that now.

Follow-on to KhronosGroup#3222 .

Follow-on to #3222 .

Elchi3 · 2024-01-25T12:21:08Z

This PR seems to be rendered on https://registry.khronos.org/webgl/specs/latest/1.0/ but not on https://registry.khronos.org/webgl/specs/latest/2.0/. Can the WebGL 2 spec be updated, too?

kainino0x · 2024-01-25T20:20:34Z

It is defined on WebGLRenderingContextBase, so WebGL 2.0 inherits it from 1.0:

WebGL2RenderingContext includes WebGLRenderingContextBase;
WebGL2RenderingContext includes WebGL2RenderingContextBase;

(Same as for many other entrypoints.)

Elchi3 · 2024-01-26T08:49:26Z

ahhh of course! Thank you!! Sorry for the noise.

kdashg requested a review from kenrussell March 3, 2021 21:14

kdashg mentioned this pull request Mar 3, 2021

Add WebGLContextAttributes.pixelFormat RGBA8/SRGB8_ALPHA8/RGBA16F. #3220

Closed

kenrussell reviewed Mar 4, 2021

View reviewed changes

kenrussell requested a review from kainino0x March 4, 2021 00:33

Spec glrc.backbufferStorage.

8b7b88a

kdashg force-pushed the backbuffer-storage branch from c81e9fa to 8b7b88a Compare March 4, 2021 02:09

RafaelCintron reviewed Mar 13, 2021

View reviewed changes

kenrussell mentioned this pull request Mar 16, 2021

Add HDR compositing of HTMLCanvasElement proposal w3c/ColorWeb-CG#15

Merged

s/backbufferStorage/drawingBufferStorage/, add read-only drawingBuffe…

4c44560

…rFormat.

kenrussell approved these changes Mar 25, 2021

View reviewed changes

kenrussell changed the title ~~Spec glrc.backbufferStorage.~~ Spec glrc.drawingBufferStorage. Mar 25, 2021

kenrussell reviewed Mar 25, 2021

View reviewed changes

kenrussell mentioned this pull request Jun 15, 2021

Add drawingBufferColorSpace/unpackColorSpace attributes to WebGLRenderingContextBase. #3292

Merged

ccameron-chromium mentioned this pull request Jun 19, 2021

Canvas 2D color management w3ctag/design-reviews#646

Closed

1 task

ccameron-chromium mentioned this pull request Jul 19, 2021

WebGL, WCG and HDR w3c/ColorWeb-CG#2

Open

kainino0x mentioned this pull request Dec 21, 2021

Rely on canvas size instead of allowing size to be configured gpuweb/gpuweb#2416

Closed

kainino0x mentioned this pull request Jan 21, 2022

Open questions about swap chain formats gpuweb/gpuweb#1231

Closed

5 tasks

ccameron-chromium mentioned this pull request Feb 19, 2023

Roadmap for a color-managed workflow in three.js mrdoob/three.js#23614

Closed

20 tasks

ccameron-chromium mentioned this pull request Apr 25, 2023

Minimal HDR Canvas proposal w3c/ColorWeb-CG#95

Merged

ccameron-chromium mentioned this pull request Sep 5, 2023

HDR on the web w3c/tpac2023-breakouts#74

Open

ccameron-chromium mentioned this pull request Nov 15, 2023

WebGL drawingBufferStorage API WebKit/standards-positions#278

Open

ccameron-chromium mentioned this pull request Dec 20, 2023

WebGL drawingBufferStorage mozilla/standards-positions#943

Open

Merge branch 'main' into backbuffer-storage

342d77d

kenrussell merged commit 1025e76 into KhronosGroup:main Dec 20, 2023
2 checks passed

kenrussell mentioned this pull request Dec 20, 2023

Upgrade extract-idl.py for Python3; regenerate IDLs. #3606

Merged

kenrussell added a commit to kenrussell/WebGL that referenced this pull request Dec 20, 2023

Fix "code" tags causing incorrect HTML rendering.

f50f2e1

Follow-on to KhronosGroup#3222 .

kenrussell mentioned this pull request Dec 20, 2023

Fix "code" tags causing incorrect HTML rendering. #3607

Merged

kenrussell added a commit that referenced this pull request Dec 20, 2023

Fix "code" tags causing incorrect HTML rendering. (#3607)

036f520

Follow-on to #3222 .

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

Spec glrc.drawingBufferStorage. #3222

Spec glrc.drawingBufferStorage. #3222

kdashg commented Mar 3, 2021

kenrussell left a comment

kenrussell Mar 4, 2021

kdashg Mar 4, 2021

kenrussell Sep 23, 2023

kenrussell Mar 4, 2021

kenrussell Mar 4, 2021

kdashg Mar 4, 2021

kenrussell commented Mar 4, 2021

kainino0x commented Mar 4, 2021

kdashg commented Mar 4, 2021

kdashg commented Mar 4, 2021

lexaknyazev commented Mar 4, 2021

ccameron-chromium commented Mar 4, 2021

kdashg commented Mar 4, 2021

kdashg commented Mar 4, 2021

kainino0x commented Mar 6, 2021

kenrussell commented Mar 9, 2021

RafaelCintron Mar 13, 2021 •

edited

Loading

kenrussell Mar 15, 2021

RafaelCintron Mar 13, 2021 •

edited

Loading

kenrussell Mar 15, 2021

kenrussell left a comment

kenrussell Mar 25, 2021

kenrussell Mar 25, 2021

greggman commented Feb 8, 2022

kdashg commented Feb 9, 2022

ccameron-chromium commented Dec 18, 2023

kenrussell commented Dec 20, 2023

Elchi3 commented Jan 25, 2024

kainino0x commented Jan 25, 2024

Elchi3 commented Jan 26, 2024

Spec glrc.drawingBufferStorage. #3222

Spec glrc.drawingBufferStorage. #3222

Conversation

kdashg commented Mar 3, 2021

kenrussell left a comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

kenrussell commented Mar 4, 2021

kainino0x commented Mar 4, 2021

kdashg commented Mar 4, 2021

kdashg commented Mar 4, 2021

lexaknyazev commented Mar 4, 2021

ccameron-chromium commented Mar 4, 2021

kdashg commented Mar 4, 2021

kdashg commented Mar 4, 2021

kainino0x commented Mar 6, 2021

kenrussell commented Mar 9, 2021

RafaelCintron Mar 13, 2021 • edited Loading

Choose a reason for hiding this comment

Choose a reason for hiding this comment

RafaelCintron Mar 13, 2021 • edited Loading

Choose a reason for hiding this comment

Choose a reason for hiding this comment

kenrussell left a comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

greggman commented Feb 8, 2022

kdashg commented Feb 9, 2022

ccameron-chromium commented Dec 18, 2023

kenrussell commented Dec 20, 2023

Elchi3 commented Jan 25, 2024

kainino0x commented Jan 25, 2024

Elchi3 commented Jan 26, 2024

RafaelCintron Mar 13, 2021 •

edited

Loading

RafaelCintron Mar 13, 2021 •

edited

Loading