[whatwg] Spellchecking mark III

[Peter Kasting]

On Wed, Jan 28, 2009 at 10:27 AM, Křištof Želechovski <kkz at mimuw.edu.pl>wrote:

>  Spelling quizzes are an artificial example; they are not interesting once
> spell checking is commonly available because the user can cheat by
> temporarily using another control that is being checked.
>
They can cheat today by pasting something into Microsoft Word.  Or looking
it up in a dictionary.  That doesn't mean there's no value in this.  There
are many "internet quizzes" where you can cheat by looking answers up with a
search engine, but they're still fun and wildly popular.  Your argument that
because people could cheat no author would, or should, want to write such a
thing does not seem supported by evidence.

>  Address fields contain data in a technical language, not in a natural
> language.  Of course, the browser can support technical languages by
> checking the syntax and validity of data as well (e.g. matching the zip code
> against the place using an external database).
>
This seems rather far afield from spellchecking.  There's a whole section of
the spec (forms) that deals with validation of various kinds of form input.
 It's separate from spellchecking for a reason: the algorithms are
completely different (and a potential rabbit-hole).  From my perspective as
a UA author that actually writes the code to do this stuff, you're
conflating too many kinds of input validation here.

> People do not say "this is English" but machines do (Content-Language MIME
> header).
>
And that header content is not generally set by explicit user action.  In
fact, it's often not set at all, or set incorrectly.  Hoping that this will
change seems naive.

> I want incorrect input, including input in an unexpected language, to be
> marked as such by the spell checker.
>
I already said that having a general-purpose, JS-accessible language
detector might be a good thing.  It would certainly be a necessary thing for
this request.  Once one had it, the request would be better addressable
without touching the behavior of the browser's spellchecker at all, because
the author could use the output of the language detector to display any
message or take any action he desired, rather than simply having the UA draw
a line under every word and thus look completely broken.

What you want is better accomplished by means other than what you propose,
and what you propose does not address the use cases for the spellcheck
attribute.  I'm not sure we can reach further agreement, so I leave this
subthread in Hixie's hands.

PK
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.whatwg.org/pipermail/whatwg-whatwg.org/attachments/20090128/8d05515f/attachment.htm>