W3C home > Mailing lists > Public > www-style@w3.org > September 2006

Re: [css3-selectors] :parent selector

From: Boris Zbarsky <bzbarsky@mit.edu>
Date: Sun, 24 Sep 2006 11:08:58 -0500
Message-ID: <4516AD9A.10501@mit.edu>
To: Mihai Sucan <mihai.sucan@gmail.com>
CC: David Woolley <david@djwhome.demon.co.uk>, www-style@w3.org

Mihai Sucan wrote:
> A several mega bytes streamed document is something I have yet to use 
> over the web.

I'm sorry, but this is the funniest thing I've read so far today.  Granted, it's 
still morning.

Documents in the megabytes to tens of megabytes range are not all that rare. 
Documentation, mailing list archives, intranet stuff of various sorts.

> A several MB document is something that poses problems in 
> current proeminent UAs as well (e.g. Opera, Gecko, KHTML, WebKit and 
> Trident).

There are plenty of multi-megabyte documents I've seen that pose no problems. 
Yes, it takes 30-40 seconds to render the document.  But the browser is doing 
incremental layout as it goes and is responsive the whole time.  At least in 
some of the UAs you list; others have known issues in this regard that are being 
worked on.

> We can't talk of MBs since current CSS implementations are not usable 
> today with huge documents as these.

I'm sorry, that's false.

> Thus, :parent would not be a problem all by itself.

This is what's known as a logical fallacy.

> Yes, :parent adds further more to the processing, any new 
> CSS selector, any new CSS property, any new spec implementation (such 
> as, but not limited to, XBL2 and SVG) make The Web slower.

The question is whether it makes it algorithmically slower or not.  That is, 
whether layout goes from, say, O(N) to O(N) with a different constant factor or 
whether it goes from O(N) to O(N^2).

The former is acceptable, generally.  The latter is something to avoid.

> There are already tutorials on how to optimize your JavaScript code, DOM 
> manipulation and CSS layouts, for performance purposes.

For what it's worth, a lot of those tutorials are flat-out wrong.  At least in 
terms of optimizing for multiple UAs at once.

> There are currently implemented CSS properties slowing down pages: 
> nobody complains.

What gave you that idea?  ;)

> We don't yet work with documents of several mega bytes.

Again, false.  As a simple example, the single-page version of the MySQL 
reference manual (which is a lot more usable than the multi-page version because 
you can actually search it) is 9.7MB of HTML.  Given the number of bugs I've 
seen filed on it rendering slowly in Gecko, people are for sure working with it.

> Having :parent is not only about speed. It's about limiting or not the 
> capabilities given to web developers.

Of course.  It's a tradeoff.

> My request for :parent and possibly all other similar requests come from 
> web authors

They come from a subset of web authors; generally those working on user 
interfaces, not documents.  The ones who do documents care a LOT about 
incremental rendering, from what I've seen.

-Boris

P.S.  If you can show a way to implement :parent without algorithmic slowdown, 
that's great, of course.
Received on Sunday, 24 September 2006 16:09:39 GMT

This archive was generated by hypermail 2.2.0+W3C-0.50 : Monday, 27 April 2009 13:54:47 GMT