Projects/KHTML: Difference between revisions

From KDE TechBase
No edit summary
Line 27: Line 27:
=== Potentially high-impact bugs ===
=== Potentially high-impact bugs ===
* {{Bug|150662}}
* {{Bug|150662}}
* {{Bug|158532}} --- assert failure in the outline painting code. Didn't manage to sort it out myself - S.E.
* derstandard.at
* derstandard.at
* tokenizer script-recursion crashes - (what bug #? (Spart))
* tokenizer script-recursion crashes - (what bug #? (Spart))
** Might not be a good name, but I mean: {{Bug|170185}}, part of {{Bug|165601}}, part of {{Bug|162498}}, later bts and dupes of {{Bug|145666}}, and likely some others. My initial description is probably wrong, though --- the problem is the hack in handling window.document that creates the document by calling begin/write/end is doing so on a document with a parser active; one of the triggers is probably because the restoration of frames from KHTMLPageCache may happen in the wrong order, but the various reports suggest it's not the only case. (The reason for the original description is that I recalled seeing quite a few reports mentioning the same m_scriptExecuting assert, e.g. {{Bug|118104}} for a long time, and figured it's time to get rid of this; some of those are actually mixed in with the dupes of the above). -S.E.
** Might not be a good name, but I mean: {{Bug|170185}}, part of {{Bug|165601}}, part of {{Bug|162498}}, later bts and dupes of {{Bug|145666}}, and likely some others. My initial description is probably wrong, though --- the problem is the hack in handling window.document that creates the document by calling begin/write/end is doing so on a document with a parser active; one of the triggers is probably because the restoration of frames from KHTMLPageCache may happen in the wrong order, but the various reports suggest it's not the only case. (The reason for the original description is that I recalled seeing quite a few reports mentioning the same m_scriptExecuting assert, e.g. {{Bug|118104}} for a long time, and figured it's time to get rid of this; some of those are actually mixed in with the dupes of the above). -S.E.
*** I analyzed this some more, and it's very hard to fix; basically the whole restoreState on multiple frames thing is broken, as it tries to restore kids independently of parents, and each may have <script> fragments, etc. The code likely needs to be reworked to do deferred application of saved info such as scroll position, etc., once loaded. -Maks

Revision as of 16:14, 20 October 2008

In-progress bugfixes

  • SadEagle:bug #169988 --- focus/blur on all elements, head/body parsing #?????, multiple problems with onchange (bug #170451, many others)
  • I have a patch for doing dns-prefetch as was suggested on http://www.kdedevelopers.org/node/3701#comment-8086
    • Maksim, to answer your question, the point is not to retrieve on hover but as soon as an anchor's href is parsed, so it is wasteful, but the Chromium guys say they observe it doesn't put a significant pressure on DNSes as the average page as few external links. My approach would be to cap the number of resolution to something (configurable) like 40 per page, getting most of the benefit on news site, without the DDOSing behaviour. It does give a kind of "what could possibly go wrong" feeling though :-)

Acid3 stuff

See Projects/Acid3

Major targets

  • SVG --- vtokarev
  • DOM namespace changes --- vtokarev --- done (in trunk)
  • Class attribute hashconsing for improved css selectors performance --- vtokarev
    • Implementation that we have currently is significantly slower on specific test case (probably less on real-life pages). Anyway, it should improve performance (maybe even memory usage), with small drawback on complexity
  • HTML5 Audio/Video --- Spart
  • getComputedStyle w/display:none --- SadEagle
  • iframe stuff --- SadEagle, non-immediate
    • XSS permissions for about: and javascript: URL affect tinyMCE
  • canvas --- SadEagle, non-immediate:
    • New params to putImageData, createImageData, parameter checking updates
    • Depends on getComputedStyle:
      • Add support for currentColor
      • text

Potentially high-impact bugs

  • bug #150662
  • bug #158532 --- assert failure in the outline painting code. Didn't manage to sort it out myself - S.E.
  • derstandard.at
  • tokenizer script-recursion crashes - (what bug #? (Spart))
    • Might not be a good name, but I mean: bug #170185, part of bug #165601, part of bug #162498, later bts and dupes of bug #145666, and likely some others. My initial description is probably wrong, though --- the problem is the hack in handling window.document that creates the document by calling begin/write/end is doing so on a document with a parser active; one of the triggers is probably because the restoration of frames from KHTMLPageCache may happen in the wrong order, but the various reports suggest it's not the only case. (The reason for the original description is that I recalled seeing quite a few reports mentioning the same m_scriptExecuting assert, e.g. bug #118104 for a long time, and figured it's time to get rid of this; some of those are actually mixed in with the dupes of the above). -S.E.
      • I analyzed this some more, and it's very hard to fix; basically the whole restoreState on multiple frames thing is broken, as it tries to restore kids independently of parents, and each may have <script> fragments, etc. The code likely needs to be reworked to do deferred application of saved info such as scroll position, etc., once loaded. -Maks