Node removal should not trigger "update the image data"

[yoavweiss]

Following an IRC discussion I don't think that node removal should trigger the "update the image data" algorithm as it may trigger spurious downloads, and there's no real benefit in doing that.

Once the image is removed from the picture parent, nothing will be displayed. Also, once it will get added elsewhere, "update the image data" algorithm should run.

Maybe we should just make sure that once it was removed from picture, the algo should run at next insertion (even if not to a picture parent or a separate document)?

[yoavweiss]

/cc @zcorpan

[zcorpan]

It might not be displayed but it can be painted on a canvas and its naturalWidth etc can be read. I'm not convinced we should complicate this. Is it a problem in practice? The "relevant mutations" are intentionally pretty dumb to be more likely to be interoperable.

[yoavweiss]

I'd argue that it's more likely that a subtle "removing <img> from one parent and adding it to another" will cause spurious downloads than it is that handling of a detached img node would result in the wrong image being displayed by accident.

[yoavweiss]

More irc discussions lead to the conclusion that the above scenario won't trigger spurious downloads if done in the same script, but the "removed an <img> and forgot about it" would.

[yoavweiss]

A way to resolve this would be to "update the image data" with a null "selected source" in the removal scenario.

[zcorpan]

That would mean an extra flag to "update the image data", and check it in step 3 and 8. Step 9 would then set the image to the broken state and fire an error event. (Maybe we should avoid the error event also?)

[yoavweiss]

Yeah, avoiding the error is probably better

[zcorpan]

But that would still leave the image broken if you insert it to the document again... I'm still unconvinced we should change things here.