Hi, pages with complex CSS feel a bit slow with the current css-prototype and gprof shows a huge number of CssSelector::match() calls. Attached is a patch that tries to improve that by: * putting CssSelectors into hashtables so that we only need to match against a small subset of CssSelectors. * avoiding repeated matches of CssSimpleSelectors against the same part of the doctree. Please test and report if there is a real world speedup. Cheers, Johannes
Johannes wrote:
pages with complex CSS feel a bit slow with the current css-prototype and gprof shows a huge number of CssSelector::match() calls. Attached is a patch that tries to improve that by:
* putting CssSelectors into hashtables so that we only need to match against a small subset of CssSelectors.
* avoiding repeated matches of CssSimpleSelectors against the same part of the doctree.
Please test and report if there is a real world speedup.
Wow! In the past, my best example had been pages on reddit.com that had at least maybe 300 comments. They had been unusably slow--although this would encourage a person not to waste time looking at reddit. Quite usable now.
On Tue, Jan 13, 2009 at 08:00:37PM +0000, corvid wrote:
Johannes wrote:
pages with complex CSS feel a bit slow with the current css-prototype and gprof shows a huge number of CssSelector::match() calls. Attached is a patch that tries to improve that by:
* putting CssSelectors into hashtables so that we only need to match against a small subset of CssSelectors.
* avoiding repeated matches of CssSimpleSelectors against the same part of the doctree.
Please test and report if there is a real world speedup.
Wow!
In the past, my best example had been pages on reddit.com that had at least maybe 300 comments. They had been unusably slow--although this would encourage a person not to waste time looking at reddit.
Quite usable now.
Great. reddit.com is looking strange though. There is this weird blue circle...
Johannes wrote:
On Tue, Jan 13, 2009 at 08:00:37PM +0000, corvid wrote:
Johannes wrote:
pages with complex CSS feel a bit slow with the current css-prototype and gprof shows a huge number of CssSelector::match() calls. Attached is a patch that tries to improve that by:
* putting CssSelectors into hashtables so that we only need to match against a small subset of CssSelectors.
* avoiding repeated matches of CssSimpleSelectors against the same part of the doctree.
Please test and report if there is a real world speedup.
Wow!
In the past, my best example had been pages on reddit.com that had at least maybe 300 comments. They had been unusably slow--although this would encourage a person not to waste time looking at reddit.
Quite usable now.
Great. reddit.com is looking strange though. There is this weird blue circle...
I think it's an iframe's bullet hypertrophied by #ad-frame { border: 0px; overflow: hidden; width: 300px; height: 300px; } <iframe id="ad-frame" frameborder="0" scrolling="no" name="ad-frame" src="/static/ad_default.html"></iframe>
On Wed, Jan 14, 2009 at 03:40:23PM +0000, corvid wrote:
Johannes wrote:
On Tue, Jan 13, 2009 at 08:00:37PM +0000, corvid wrote:
Johannes wrote:
pages with complex CSS feel a bit slow with the current css-prototype and gprof shows a huge number of CssSelector::match() calls. Attached is a patch that tries to improve that by:
* putting CssSelectors into hashtables so that we only need to match against a small subset of CssSelectors.
* avoiding repeated matches of CssSimpleSelectors against the same part of the doctree.
Please test and report if there is a real world speedup.
Wow!
In the past, my best example had been pages on reddit.com that had at least maybe 300 comments. They had been unusably slow--although this would encourage a person not to waste time looking at reddit.
Quite usable now.
Great. reddit.com is looking strange though. There is this weird blue circle...
I think it's an iframe's bullet hypertrophied by
#ad-frame { border: 0px; overflow: hidden; width: 300px; height: 300px; }
<iframe id="ad-frame" frameborder="0" scrolling="no" name="ad-frame" src="/static/ad_default.html"></iframe>
Ah right. We should probabely disable style processing for content that dillo generates on its own (e.g. <object> or <(i)frame> replacements). Cheers, Johannes
participants (2)
-
corvid＠lavabit.com -
Johannes.Hofmann＠gmx.de