UI events containing lone surrogates

[annevk]

Per WebAssembly/interface-types#21 (comment) it's possible for input events to contain lone surrogates, at least on Windows. Perhaps we should require browsers to coalesce such events, if possible, to prevent none scalar values from being handed to JavaScript.

cc @Pauan @alexcrichton @lukewagner

[Pauan]

@annevk Thanks a lot for getting the ball rolling on this!

I created a reduced test case (JS only, no wasm): http://paaru.pbworks.com/w/file/fetch/132853233/input-bug.html

I was able to reproduce it with Windows 10 (64-bit) with the following browsers:

• Mozilla Firefox Developer Edition 67.0b7 (64-bit)
• Google Chrome 73.0.3683.86 (Official Build) (64-bit)

Interestingly, I could not reproduce it with these browsers (they do not have this bug):

• Microsoft Edge 42.17134.1.0
• Microsoft Internet Explorer 11.648.17134.0

[hsivonen]

https://hsivonen.com/test/moz/input.html is probably of interest when testing.

[hsivonen]

The Windows 10 onscreen keyboard generates a keydown/keyup pair for each surrogate in Firefox but a keydown/keyup pair per scalar value in Edge. (Note that multi-scalar value emoji generate multiple keydown/keyup pairs in Edge, i.e. every scalar value of an emoji is a keystroke.)

[hsivonen]

Hah. https://docs.microsoft.com/en-us/windows/desktop/inputdev/wm-unichar says this goes back to XP!

[hsivonen]

Correction: Edge doesn't issue keyup events for emoji.

[hsivonen]

Gecko bug

[hsivonen]

Chrome seems to expose the unpaired surrogate in input and beforeinput events as well as the observable value of the HTML input element but not as the key of keydown and keyup.

[hsivonen]

Chrome bug

[Pauan]

@hsivonen Thanks! I'm excited to see this fixed in browsers proper, so we won't need any more workarounds.

[hsivonen]

IE has been able to report a single keystroke (with a surrogate pair in the key string) per astral keystroke since at least IE9 on Windows 7. (The test case is incompatible with earlier IE.)

[hsivonen]

@travisleithead, are you aware of any reason why other browsers shouldn't adopt the Trident/EdgeHTML behavior here?

(Firefox on Linux appears to have been shy to report astral keystrokes as keystrokes and instead emits fake composition events for them. Based on the source comments, it looks this wasn't motivated by a concrete Web compat reason and was done just in case.)