On Tue, Nov 14, 2017 at 10:57:23PM +0000, Mark Svancarek wrote:
Maybe this is semantics, but I don't think of it as "the browser" doing the conversions. It would need to be some JS coded into the page. The browser is just the sandbox the page's script would be running inside.
Strictly of course you're right, which is why the specification talks about "user agents". The point is that the thing the user interacts with is not necessarily the data that qualifies as "input". If those must be 1:1, then there is obviously a problem in the spec because users can't reliably enter A-labels or Punycode-form IDNA2003/UTS#46 labels or even NFC-normalized UTF-8. Alternatively, maybe the specification is doing some filtering but not actually producing valid email addresses, in which case the spec has a serious problem as written in not permitting Unicode in the server-part. But since it explicitly allows the user agent to do transformations for display, that doesn't sound like what's intended (at least to me). I've now said enough about this :) A -- Andrew Sullivan ajs@anvilwalrusden.com