Re: [Gnash-commit] gnash ChangeLog server/character.cpp server/cha...

[Sandro Santilli]

On Sat, Sep 01, 2007 at 01:20:47AM +0000, Sandro Santilli wrote:

> Log message:
>               * server/dlist.{cpp,h}: when unloaded character's unload() 
> method
>                 returns true (character or any of its childs has onUnload 
> methods to
>                 be run) don't really remove it from the list, but just shift 
> it
>                 down to the "removed" zone. Advance and display only 
> non-removed
>                 characters (depth-zone based for now, possibly too weak).
>                 Add lots of paranoid invariant testing (not enough, missing 
> the
>                 one for no unloaded characters out of "removed" depth zone).
>               * server/character.{cpp,h}: more info about removedDepthOffset,
>                 assertion preveing double unload of a character (this can be
>                 probably easily broken by a focused testcase using 
> removeMovieClip
>                 on a removed but still-reachable character).
>               * server/movie_root.{cpp,h}: added cleanupDisplayList() method, 
> and
>                 call it at the end of actions execution to properly cleanup 
> removed
>                 but still-reachable characters.
>               * server/sprite_instance.{cpp,h}: implement a 
> cleanupDisplayList, to
>                 be called by movie_root, make sure to not include unloaded
>                 characters when visiting the DisplayList for bounds 
> extractions and
>                 similar.

This is a pretty big change. We don't always really remove characters from the 
DisplayList,
but we keep them alive if they (or their children) have any onUnload event 
handler to be
executed. In that case we only shift them down to a different depth (verified 
to be the
expected behaviour).

Things to test (important):

 1) what happens if the target depth after the shift is occupied.
    For example:
        - mc1 placed with onUnload event at depth 2, then removed (shifted down)
        - mc2 placed with onUnload event at depth 2, then removed (shifted down 
occupied)
        - both onUnload handlers will need to reference 'this' and verify they 
point
          to the correct instance (mc1 or mc2) Better if they both have the 
same name!

 2) How is it possible for an unloaded character to be out of the unloaded 
depth zone
    I've seen this with bombgame.swf, for which line 2360 of 
sprite_instance.cpp currently
    cleans out unloaded characters from the newlyAdded DisplayList.
    Still, I don't know how was it possible for a character at depth -16381 (3) 
to be
    tagged as "unloaded" (it should have been moved down to -16388 when removed
    (-32769-(-16381))

 3) Implement pixel-checking for some of the available depth-based tests (IIRC 
displaylist_depths_test.c
    had some weird visual behaviour, which may now be explainable)

 4) Add a test for characters out of the "removed" zone not being marked as 
'unloaded' in the
    DisplayList::testInvariant() and run as many SWF till gnash crashes :)

--strk;