More SVG as image with subresources test coverage

[annevk]

All credit to Said Abou-Hallawa for doing most of the work here in WebKit many years ago.

[annevk]

I think my cache bust identified a legitimate bug in Firefox. @emilio want to take a look? WebKit will be racy here as well, but we'll mark it as flaky until that's fixed.

@jgraham can this still be merged?

[emilio]

Hmm, are you sure per spec this works? Shouldn't you image.decode().then(? What guarantees that the inner image:

• Is decoded sync?
• Is decoded on time?

[shallawa]

I think the web developer should not worry about loading or decoding internal images. An image should be looked at as one entity especially if you do not have access to its internal structures (like the image in this test case). So for an image there should have only one loading time and only one decoding time.

For drawing an image into a canvas, I think this should force sync decoding. Unlike HTMLImageElement in the page, the image will be drawn into the canvas only once. If the image is not drawn because it has not been decoded yet, HTMLImageElement in the page will be repainted once it is ready. But canvas can't repaint the image.

So I think image.onload() should be enough to guarantee the image can be drawn to the canvas regardless of whether it contains internal images or not and regardless its decoding attribute is sync or async

[emilio]

@annevk not sure, you sure you're just not relying on the inner <image> decoding sync? But yeah maybe worth aligning in this case and forcing sync decoding on SVG-as-image documents...

[annevk]

Yeah, note that Gecko already seems to have that behavior in most cases. It's just the post load case that's flaky. But overall I think that all images should behave in the same way.