Mozilla

Highlights

• Mail/news eng+qa team ("You've Got Margaritas") wins the CPD margarita blend-off!
• Fixed 43 bugs
• Threads mostly work (although navigating [e.g. next unread] doesn't work)
• Command enabling is ready for people to use (documentation coming soon)
• Fixed bad bugs in corruption of outgoing messages
• URL dispatching is checked in (turn on with a pref)
• Choosing format of outgoing messages (after composition) is working
• Posted results of weekly UI meeting

Priorities

• Fix remaining dogfood bugs
• Get a handle on our M12 bug list (M12 tree closure == 12/6)
• davidmc - 10837 Import tool for 4.x address books
• alecf - 16929 Crash drag-scrolling tree, 18420 Scrolling screws up after deleting all messages
• jefft - 8889 Can't post attachments to newsgroups, 12929 Close IMAP connection when deleting folder
• rhp - 18410 Make URL scanning for plain text messages work for non-ASCII, 18008 Message loading performance on Linux
• mscott - 14928 URL Dispatching, 19166 Crash sending some messages
• putterman - Make threading work as described in the UI meeting
• ducarroz - 17771 Email addresses parsed wrong when replying, 10932 Fix tabbing in message compose window
• sspitzer, bienvenu, and chuang are out next week, back after Thanksgiving

Lowlights

• Regressions continue to be... well, you know
• Hardware purchases are slower than scrolling the thread pane

I'm basically working on PDT+ bugs for hitting our dogfood milestone. There was one thing I did this week, though, that might be interesting for mozilla developers:

The stack traces for Addrefs and Releases are now getting dumped properly on Windows. Earlier they were only working on Linux. (See http://www.mozilla.org/performance/refcnt-balancer.html for details on how to dump refcount logs).

I've added a new environment variable (XPCOM_REFCNT_LOG_ENABLE_VIA_PREF) which works in concert with the pre-existing variables (XPCOM_MEM_REFCNT_LOG, XPCOM_MEM_LOG_CLASSES, and XPCOM_MEM_LOG_OBJECTS). If you set XPCOM_REFCNT_LOG_ENABLE_VIA_PREF to 1, you will be able to use a pref in the debug pane of mozilla to turn refcount logging on and off. This is useful for cases where you want to get refcount logs of finer granularity than the lifetime of the app.

Also, now the refcount log is synced up on each addref and release. This ensures that your refcount log will be as current as possible in case you crash. I was debugging a crash caused by premature frees where my refcount log was left out of date because the last batch of addref/release logs weren't committed to the disk.

Brian reports that he's got mousewheel support checked in for unix. He'd like people to check it out and report any bugs to him about it. Also, a Win32 implementation is almost done. Mac might be somewhat hard...

Brendan Eich landed some changes -

"I implemented function expression statements with dead-code elimination for constant if and if/else statements, so users can conditionally define functions and 'if(0)'-out dead or unready functions. Let's see, what else did I do? Oh yeah, the strict option, for more JS warnings. And the -Werror option (my post called it error_on_warn, but I think I'll follow gcc's lead)."

"Bigger news: brutal sharing of JS scripts in XUL. This sped up new-window by 40ms, will get another 40 or so when waterson and hyatt finish brutally sharing the 'master document' (outermost .xul file, currently not shared brutally or otherwise). 'Brutal Sharing' avoids reparsing XUL and recreating content nodes for window chrome when making additional windows - the first window's nodes can be reused with copy-on-write sharing. Compiled JavaScript scripts and functions can also be reused."

John Bandhauer is continuing to refactor XPConnect internals to correctly reflect the relationship between JavaScript execution contexts (e.g. stacks) and the global objects that executing scripts see. (There isn't much of one!) This architectural fix will save much future confusion for users of XPConnect.

John says - "Landing this stuff is dependent on 17736 (DOM code's assumptions about JSContexts). Brendan offered to help fix that bug. For now I'm happy improving xpconnect away from the trunk."

Mike McCabe is working on an IDL-like way to express information about XPCOM components, e.g. whether a given component is to be considered thread-safe for proxying purposes, and the interfaces it is guaranteed to implement. This will allow us to skip many QueryInterface calls in components manipulated through XPConnect, and is a step towards accessing the DOM components through XPConnect. Mike is also tithing pork to 18638.

MathML is continuing its torture of the lizard, and so far the lizard is enduring everything. I landed a prototype work for the rendering of stretchy characters. This is one of the distinctive features of mathematical typesetting. It reflects the ability of some characters to "auto-grow" depending on their surrounding context. Central to the support of stretchy symbols is the Symbol font and the GetBoundingMetrics() method that was landed previously. Not all stretchy chars are supported given that fonts that are needed are not readily available and/or are not supported yet within GFX. Also due to some differences in the character coordinate system between platforms, strecthy rendering is currently only tuned for Win32 (see bug 19024 for screenshots). However, Shyjan will be tuning the rendering on Linux soon. And when Shyjan is at something, one can seat back with expectation :-)

Also done was the support of the mfenced tag, together with the chasing and catching up of changes that are going on in layout!

Roger has also included a screenshot.

1. Eric Vaughan reimplemented box on top of linked list to fix the 100 element  limitation and to lower memory consumption.  (evaughan)
2. Mike Pinkerton added the ability to drag the urlbar proxy icon. (pinkerton)
3. Mike also wrote some clipboard and drag&drop documentation: Introduction To The Clipboard and Drag & Drop, Toolbars and Toolboxes, The Clipboard, Drag and Drop,  The Tree Widget.  (pinkerton)
4. Stuart Parmenter added nsIFilePicker interface to replace nsIFileWidget and nsIFileSpecWithUI. (pavlov)
5. Chris Saari got menu positioning to work, with help from Dan Matejka and Stuart Parmenter. (saari, pavlov, danm)
6. David Hyatt helped Alec Flett fix a problem the tree widget has with  recreateframesonattributechanged. (alecf, hyatt)
7. Dave also helped Travis Bogard with some chrome vs content issues in Travis' webshell redesign. (travis, hyatt)
8. Dave also helped Steve Elmer make contiguous (shift-click) selection work. (selmer, hyatt)
9. Steve Dagley implemented the MacOS version of Truncate() in nsIFileSpec in support of the new cache implementation. (sdagley)
10. The XPToolkit team resolved 45 bugs in the last week, fixing 29 of them.  See our resolved bug list below for details.

Lowlights

• Steve Dagley has not finished the Mac implementation of nsIFile.  He plans to finish it while on vacation.  (sdagley)

Priorities
    (current open bugs prefixed with a *)

1. pinkerton: *18409 Pasting from clipboard causes extraneous characters for any CR's in the data
2. danm: * 17065 Messenger stalls opening IMAP  Inbox
    * 18005 Leave mail window for a long time, GetMsg, crash
   13695 Window close box needs to call JS close routines
3. saari: * 19421 No menubar on Mac,
   *9454 XP & popup menus can appear partially offscreen,
   * 18983 html forms picking up carets for each field with text,
   Menu scrolling.
   Command dispatching
4. hyatt: *12687 chrome registry support
   Write up a ui document on css restrictions/guidelines for our skin.
   Help Chris Saari with command dispatching.
5. evaughan: *12122 overflow property within XUL == disappearing content
    *19142 Styles broken with gfx scrollbars on
6. sdagley: *17949 Mac implementation of NSIfile

Issues

• Regressions are continuing to be a huge problem for us.  Things are breaking the day after being fixed, which keeps us busy continually fixing things that should still be working.  I'm not sure of the best solution to this problem, but I'm sure it involves using more care when changing code that you don't own, using extreme caution when changing any interfaces, and doing much more thorough testing for any possible regressions before checking code in.

People

• Stuart Parmenter will be on vacation next week. (pavlov)
• Steve Dagley will be on vacation next week.  (sdagley)

• Worked on nsIFile implementation for Unix
• The JavaScript loader unloads JavaScript components
• Native and JS components understand deferred component loading as discussed briefly in .xpcom

Highlights:

• 43 editor bugs were resolved this past week, 9 were PDT+
• 66 open M12 bugs, 20 are PDT+

Lowlights:

• We currently have 265 open bugs against the editor
• Steve is on vacation
• Our bug list keeps growing and growing, with Steve and Simon on other projects, we are starting to feel the impact

Accomplishments:

• Kin
• performed regression testing in regard to the numerous focus bug fix check-ins
• fixed 17989, this helped resolve bugs 18353, 16286 (partial fix) and 15982
• fixed 19371
• worked on resolving 18092
• modified code to fix 1413, which in turn fixes 18002 & 18006

• Kathy
• fixed 14750, 14624, 15442, 18892
• worked on 9074
• lots of bug list triage for the editor team and saari
• discussing dom key spec (and lack thereof) with joki, and possible complications with deadkeys on Windows
• helped rods with some bugs on his plate (Mac-related)
• filed many bugs against editor
• Steve
• continued to work on webshell issues
• did a big check-in for gfx text-control
• Joe
• reviewed and checked in buster's fix fro 18537
• fixed mail blockers 18922, 18938, 18723
• fixed 15696
• wrote system for editor HTML rules to track what portion of the document is changed by an operation
• wrote whitespace handling that validates whitespace over changed document region
• wrote mozBR handling that inserts a special BR on empty list items and in empty table cells and at the end of structures when a return is entered (deals with the new line issue)
• debugging other whitespace issues
• debugging text paste performance issues
• Simon
• investigated image rendering performance, found some image refresh issues in layout's image drawing code, and experimented with the effects of necko buffering sizes in image rendering speed. Have a suite of changes in my tree that make image rendering faster than 4.5.
• checked in a change that improves threading performance on Mac, helping image rendering and other things.
• reviewed some imageLib memory leak fixes for neeti.
• removed a redundant JPEG shared lib from the mac build by merging it with the decoder library.
• debugged a subtle problem exposed by brendan's activation of brutal sharing for scripts, that was causing the Mac build to do odd things at startup, and, with brendan's assistance, checked in a workaround.
• helped valeski land some HTTP changes
• worked on some Mac build cleanup issues
• Mike
• rebuilt his tree and merged back in the fixes he had in the tree before he left.
• Charley
• on vacation Mon-Thurs.
• fixed bug 1900
• Akkana
• fixed 19237
• fixed 11818, but waiting for code review
• worked on bug 18033
• checked in a contribution from Ben Bucksch for anchor and img tags, and fixed some of the extra blank lines in plaintext mail while I was in there.

Priorities:

• International development support by the entire Ender team when needed and when necessary
• Review and triage M12 bug list
• Next week:
• Kin: work on 19371, and deal with buster's open M12 bug list, triage bugs, jump on regressions
• Kathy: bug 19372, review buster's bug list, help triage the bug list
• Steve: on vacation
• Joe: merge with tip (see list of fixes below) -- get it reviewed, check it in and resolve any issues, work on table deletion rules, work on post-operation selection, work on whitespace handling documentation, continue to work on text paste performance
• Simon: Porkjokey, get image rendering changes fully debugged, reviewed, and checked in. Look at XP menu performance, and other layout performance issues
• Mike: work on the API for Akkana, work on lower priority bugs until the saari's focus work is checked-in
• Charley: work on bugs 14344 (Page Properties) and 6256 (table editing)
• Akkana: check in the fix for 11818, work on 18033 -- which will fix 18046 (Sammy Ford contributed an IDL file (yay!), so what remains is to work with Mike and Saari to

get the focus information needed to implement these APIs.); and work on some minor changes for html mail quoting (change API to return the blockquote node, to make it easier for mail to put the caret in the right place); related to 19237

Issues:

• The editor team needs to iron out functionality issues and document action policies. This will be part of the overall Editor Functional spec.
• Mike is waiting for saari to check-in his focus work so he can wrap up 2 PDT+ bugs.
• Joe is running into placement issues with the caret (UBER-Caret), for example: If you have a list, for instance, and you create another list in front of the first list:  the moment you close out the top list, your caret will be an uber caret after that top list.  I can't add a br between the lists, because that br would actually cause new vertical whitespace.  There's just nowhere else for the caret to go.
• Simon needs to have time with Troy to discuss Troy's plans for future layout performance work, so duplication of effort can be avoided. Simon is also gated to some extent on the remaining incremental reflow work that nisheeth and troy are working on, Simon needs to know this information before he can get a feel for what performance issues remain with layout.
• Simon is continuing to work with Porkjockey effort, editor tasks are on hold
• Steve is 1. on vacation, 2. working on webshell rewrite, other editor tasks are on hold

Joe's big check-in will fix these bugs:

• 2073 & 12337 - whitespace trouble when typing (you can end up with all nbsp's instead of spaces)
• 16715 - caret gone or wrong after hitting return
• 16988 - aim foiled by our mozDIVs (since these are history, maybe we don't need aim-specific rules prior to beta?)
• 17908 - enter at end of document does nothing
• 16085 - 4.x mail replied foiled by mozDIVs
• 19368 - inline style lost after typing return
• 17059 - "normal" made a paragraph, but typing made a div.  Now they both just use direct text in the body (or wherever)
• 18717 - exiting list gives UBER caret (note: this will only be fixed most of the time, not all)
• 17778 - cant type beside an image (it ends up above or below it) - this was a rather strange ramification of the old mozDIV rules which are now gone.

People:

• Everybody is out on 11/25-26
• Mike was out on 11/15-17
• Charley was on vacation 11/15-18
• Steve is on vacation until 12/6
• Kathy will be in MV the first week in December.

Previous Updates