Sounds like we have a leak, then... I will dig into it. On 2010-08-20, at 14:58, Maynard Demmon wrote:
> There are no debugger messages being generated when this fails. > > -Maynard > > P T Withington wrote: >> Backtraces should be garbage-collected as soon as the call returns. But the >> are held on to by debugger output (for later inspection). Is there a lot of >> debugger output in the failing case? >> On Aug 19, 2010, at 19:02, "\"Maynard Demmon\" <[email protected]> >> (JIRA)"<[email protected]> wrote: >>> flash halts in backtrace mode but not in debug mode when dealing with >>> deeply nested views >>> ----------------------------------------------------------------------------------------- >>> >>> Key: LPP-9311 >>> URL: http://jira.openlaszlo.org/jira/browse/LPP-9311 >>> Project: OpenLaszlo >>> Issue Type: Bug >>> Components: Kernel - swf9 >>> Affects Versions: 5.0 (TRUNK) >>> Reporter: Maynard Demmon >>> Assignee: P T Withington >>> Fix For: 5.0 (TRUNK) >>> >>> >>> [15:42] mdemmon: so the other issue i was seeing is a halting state in the >>> flash player that only happens with backtrace on >>> [15:42] Max Carlson: hmmm >>> [15:42] Max Carlson: okay >>> [15:42] mdemmon: it's happening in deeply nested views that trigger >>> onheight events all the way up the tree >>> [15:43] mdemmon: by deep i mean maybe 20 levels deep >>> [15:43] Max Carlson: okay >>> [15:43] Max Carlson: what happens? >>> [15:43] Max Carlson: the player just freezes? >>> [15:43] mdemmon: flash hangs >>> [15:43] mdemmon: yes >>> [15:43] mdemmon: no error is generated, or at least none that i'm privy to >>> [15:43] mdemmon: identical code runs at lower levels of view nesting >>> [15:43] Max Carlson: Are you running the debug player? >>> [15:43] mdemmon: i am >>> [15:44] Max Carlson: do you have a testcase :) >>> [15:44] Max Carlson: It sounds like some sort of stack/heap overflow >>> [15:44] mdemmon: not really it's the filemanager i'm working on for WT >>> [15:44] mdemmon: i'm doing a tree nav >>> [15:44] Max Carlson: where the backtrace tool is generating too much >>> information >>> [15:45] Max Carlson: is it available in perforce? >>> [15:45] Max Carlson: That could be the testcase... >>> [15:45] Max Carlson: Just have Tucker or Henry look at it. >>> [15:45] Max Carlson: Also, did you try Flash 10.1 >>> [15:46] mdemmon: 10,1,53,64 installed >>> [15:46] Max Carlson: aha! >>> [15:46] Max Carlson: okay >>> [15:46] Max Carlson: well, if you can file a bug and include the p4 path to >>> get it running, that's a start. >>> [15:46] mdemmon: it is, or it will be when i check in my latest changes >>> [15:47] Max Carlson: Another thing you could do is try adding >>> fixed/constrained sizes to the containing views to help prevent computing >>> the size/cascading >>> [15:47] Max Carlson: That's good for perf anyway >>> [15:47] mdemmon: i'm not sure i understand what you mean >>> [15:48] Max Carlson: if a view doesn't have a width/height it attempts to >>> measure its children >>> [15:48] Max Carlson: which is expensive >>> [15:48] mdemmon: ic >>> [15:48] mdemmon: that's kinda what i want, but i guess i could measure it >>> myself >>> [15:48] Max Carlson: so, if you can give the containing views explicit >>> sizes, that should help >>> [15:48] Max Carlson: but the backtrace tool should be fixed to not blow up. >>> [15:48] Max Carlson: Does the memory usage explode? >>> [15:48] mdemmon: i mean the views need to reflect the size of their children >>> [15:49] Max Carlson: right, well then no worries :) >>> [15:49] mdemmon: haven't checked, but i can >>> [15:49] Max Carlson: kk >>> [15:51] mdemmon: memory use went up by about 10 megs after i started >>> opening lots of dirs down to where it would blow up normally >>> [15:51] mdemmon: that's in debug non-backtrace >>> [15:51] Max Carlson: okay >>> [15:51] Max Carlson: how bout with backtrace >>> [15:51] mdemmon: trying that now >>> [15:53] mdemmon: it went up by about 18 megs >>> [15:53] Max Carlson: that's not too bad >>> [15:53] mdemmon: then when i did the action that causes the freeze it >>> rapidly jumped up by about 60 megs >>> [15:53] Max Carlson: aha! >>> [15:53] Max Carlson: there you go >>> [15:53] Max Carlson: so, I think the backtrace tool needs to only save the >>> last n backtraces... >>> [15:54] Max Carlson: Or sumpthin' >>> [15:55] mdemmon: it's definitley a depth thing >>> [15:56] mdemmon: i can do the same folder operation 1 level up and it works >>> fine but as soon as i'm in the 11th level of nested folder it halts >>> [15:56] Max Carlson: which browser? >>> [15:57] mdemmon: ff >>> [15:57] Max Carlson: u could try chrome >>> [15:57] Max Carlson: it may give better data about how/why flash is crashing >>> [15:57] Max Carlson: Anyhow, pls file a bug and Tucker or Henry can put in >>> a circular buffer >>> [15:58] mdemmon: i'm gonna see if i can reduce the number of stack entries >>> in my code >>> [15:58] Max Carlson: maybe keep the first n and last n entries >>> [15:58] Max Carlson: kk >>> [15:58] Max Carlson: the view measurement code uses lots of recursion, so >>> that may be part of the issue >>> [15:59] Max Carlson: same with getAttributeRelative() >>> >>> -- >>> This message is automatically generated by JIRA. >>> - >>> If you think it was sent incorrectly contact one of the administrators: >>> http://jira.openlaszlo.org/jira/secure/Administrators.jspa >>> - >>> For more information on JIRA, see: http://www.atlassian.com/software/jira >>> >>> > >
