We're still getting the odd problem with destroy, even with an update to 0.83. The situation is the following:
An Event object exists in the system. An event is sent to this object (and the 'event' method of this object is called). The event method calls another method in another object which is set to be the event callback. As a result of this call the object gets destroyed. Or rather it destroys another object (its parent), which in turn destroys the Event object.
According to the "destroy" logic, this should work fine. And in fact it does.. for a while. At some point during the execution of the callback I get this message:
destroy failed (probably wrong # args?) or object called after physical destroy while executing "$ob event [$cmd getAttrs]" (procedure "newData" line 24) ::eniak::EniakServer0 ::eniak::EniakServer->newData invoked from within " ::eniak::EniakServer0 newData"
This comes at some point during the execution of the event method (which, as mentioned, is a method of the Event object). This seems to be very difficult to replicate as the simple cases I've built wont do the trick. To give you some idea why it's so difficult I'll mention that in my first case it crashed (as above) just after (successfully) creating a SelectionList object in my callback. To see if it was the amount of created objects that mattered, I instantiated a Text object before this. It still crashed with the SelectionList. The natural assumption then is that there might be something funny with the SelectionList. Just to be certain, I created a Button between the Text and SelectionList. Now it crashed immediately after the Text object -- before it even reached the Button or SelectionList in the code! Removing the Button made it crash after the SelectionList again...
This would lead me to believe that there may just be a small possibility of some kind of memory overwriting taking place or something else very funny. I'm really sorry to be this vague. Maybe I missed something after all, but I've been debugging this all day and I just can't find anything suspicious myself and am running so tight with time I can't really spend much on this -- especially as it's rather random. The problem is also that it's not very obvious how to go round this particular case without seriously altering the structure of the code.
If there is anything that can cause this I'd really appreciate some tips. Or if there is a bug somewhere in there a patch would be splendid. As we've already missed one deadline I'd appreciate any help I can get. I can send the code that causes the problem, although I'm not sure how much it will help as there's quite a bit of it. I might be able to cut some irrelevant parts off, though...
Meanwhile I'll sleep over it and see if I solve it in my dreams.
- ---------- = = ---------//--+ | / Kristoffer Lawson | www.fishpool.fi|.com +-> | setok@fishpool.com | - - --+------ |-- Fishpool Creations Ltd - / | +-------- = - - - = --------- /~setok/
It's amazing how clever you get after you've just sent a vague bug report. Here's the problem:
[~] package require XOTcl 0.81 [~] Class Foo Foo [~] Foo instproc init {} { [self] destroy safohasofhisa } [~] Foo ob destroy failed (probably wrong # args?) or object called after physical destroy while evaluating {Foo ob}
Ie. the Tcl bytecode should cause an error when it executes init because, of course, safohasofhisa is not a valid command. This is apparently not picked up by XOTcl. So I had an error in my callback method further along the method, but I never received the error but just got that "destroy failed" thing.
Am I right in believing this has something to do with the previous bug report I sent with the funny inits and destroys that caused a segfault...?
- ---------- = = ---------//--+ | / Kristoffer Lawson | www.fishpool.fi|.com +-> | setok@fishpool.com | - - --+------ |-- Fishpool Creations Ltd - / | +-------- = - - - = --------- /~setok/
-
Kristoffer Lawson