Lazy transcoding of DOM strings

[kmcallister]

DOM APIs and SpiderMonkey require us to use UCS-2 strings, but the vast majority of content comes off the wire in UTF-8 or another ASCII-compatible encoding. We can save memory and potentially time by storing a DOM string as UTF-8 and converting it to UCS-2 when first touched by script.

Not sure how this would interact with interning.

See #282, #1701.

[SimonSapin]

Perhaps we’ll want different strategies for different types of strings: content of a text node, element name, attribute name, attribute value, …

[SimonSapin]

This may impact #1879 in the choice of the encoding for the input to the HTML tokenizer.

[kmcallister]

Yep, that's why I'm thinking about it right now :) Will send an email regarding this today.

[jdm]

Dropping a link to Ms2ger's UCS-2 string experiments: Ms2ger/servo@mozilla:master...Ms2ger:strings

[Ms2ger]

cc @bzbarsky

[kmcallister]

This is part of my current plan for strings in html5ever.

[pshaughn]

Is this issue talking about something that's still relevant, or is it all stuff that was later decided to just make UTF8 all the time?

[jdm]

Given that spidermonkey is adding more and more APIs that accept UTF-8, I think we can close this.