Re: [9fans] mothra

["Howard Trickey" <howard@research.bell-labs.com>]

> Exactly. An httpfs that just serves HTTP raw MIME content (and handles
> caching).
>
> We have to divorce HTTP from HTML. They're two distinct problems.

You're kidding yourself if you think this comes anywhere near solving
the big problems in writing a web browser.  Actually fetching the bits
and passing them along is trivial.  (And, in any case, fetching the bits
is more closely tied to the logic of the browser than you might think:
you have to deal with redirections, errors, and authorization requests.
And, it is good to be able to start rendering before all of the HTML
has arrived, and certainly before all of the images have arrived.)

I wrote the first version of the charon browser with a "webget" filesystem
to serve the pages.  I abandoned it in later rewrites, mainly for speed
reasons, but also because it wasn't buying me anything.  We only ever
had one web client attached to the damn thing anyway.  But that could
change in a Plan 9 environment...

Separating things into pieces is easy, but doesn't help much if the pieces
are hard.

The first real hard part is lexing/parsing the html in a way that is
forgiving
of errors in exactly the same ways as Netscape and IE. The next real hard
part
is getting the layout (especially tables!) exactly the same as Netscape and
IE.
Another hard part is SSL, just because ASN1 is a pain in the butt.
The hardest hard part is making Javascript objects and methods that behave
exactly the same as Netscape and IE (especially if you want to do something
different with respect to the concepts of "top level windows" or "frames").

And don't say "it doesn't have to be exactly the same as Netscape and IE"
until you've had users.

- Howard Trickey, erstwhile fool