wpt: rendering to hidden canvas

[austinEng]

Various ways a canvas might be hidden

• canvas not visible and off the page
• canvas not in the DOM
• CSS display:none

[greggman]

How would a reftest test any of these? If I understand correctly, a ref test is a test where you make 2 HTML pages and compare screenshots. How do you compare blank pages in any useful way? What should these tests do?

[austinEng]

Whoops. I suppose they ought to be op tests then?

Additional things that also come to mind:

• canvas texture expiry still occurs

[greggman]

canvas texture expiry still occurs

I'm not sure how I can test this. The spec seems to make it ambiguous when a texture will expire. It says

Queue a global task on the automatic expiry task source

It then goes on to say when that task executes could be

1. Immediately after running any task.
2. Immediately after running animation frame callbacks.
3. Immediately before rendering. (which might be never for an un-rendered canvas, un-rendered because it's not visible, offscreen, not in DOM)

Or, maybe 'before rendering' means before rendering the page, not the canvas, in which case I can force a page render in various way. @kainino0x ?, @toji ?

I actually wrote a test, and just waited for timeout (which fits 1 above) and the test passed. I could change it to rAF (that would fit 1 and 2). But for 3?

As for these

• canvas not visible and off the page
• canvas not in the DOM
• CSS display:none

AFAICT all I can do here is

• draw the webgpu canvas to a 2d and check the color
• call toDataURL -> image -> draw image to canvas 2d, check the color
• call toBlob -> image -> draw image to canvas 2d, check the color
• create an imagebitmap from the canvas -> draw image to canvas 2d, check the color

Those are already done in src/webgpu/web_platform/canvas/readbackFromWebGPUCanvas.spec.ts though that canvas is always a webgpu canvas that is not in the DOM. I can add "in dom but display: none" and "in dom but positioned offscreen" if you think that's important.

[kainino0x]

I wrote up some stuff about testing this for @shrekshao on https://bugs.chromium.org/p/chromium/issues/detail?id=1404421#c6

[kainino0x]

cc #2385

[kainino0x]

I did a little reading to remember what the heck state we left canvas texture expiry in, and I think I said the wrong thing earlier iiuc officially it's allowed for a canvas texture to not expire if the canvas isn't getting displayed

Also I reopened https://github.com/gpuweb/gpuweb/issues/3295#issuecomment-1461104644 because we need to write stuff about OffscreenCanvas in the spec.