The Lounge is rated PG. If you're about to post something you wouldn't want your
kid sister to read then don't post it. No flame wars, no abusive conduct, no programming
questions and please don't post ads.
Vikram, there's a "using" statement in C# that does the try...finally thing in a nice syntax that guarantees that Dispose() gets called. I'd have a link to the MSDN, but my access to MSDN seems to be oddly broken at the moment.
You do need to implement IDisposable and ensure that the Dispose method actually does something! I think there may be a couple of articles here that give a good overview (you can find them). If not I can send you some code if you like.
Thanks for the reply, Gary, but I wasn't looking for coding tips to handle the memory leak. I know perfectly well what needs to be done, I was wondering if there is a tool that can do this for me or I had to review the entire code.
I ran into a similar problem a while back, but we owned the code for the type (you could always disassemble/assemble).
My solution was to store the stack trace of the constructor of every instance of that type as a member in that type, and then in the finalizer, log the stack trace. I also ran a timer in the background that periodically called GC.Collect(); GC.WaitForPendingFinalizers(); GC.Collect();. That way, the finalizer would run for all instances that did not get disposed, and I would be able to see where those instances got created.
I would then run the application and try to exercise all paths in the code that deal with that type. I would then go through the log and fix code that was not disposing instances.
The trick is to make sure that all code paths that involve that type are exercised, and that's what makes this non-deterministic. A tool like NCover would probably help track code coverage.
You just gave me an idea - it would be nice to have a tool automate the first part (track instances and report where they are created in code). Now I know what I'll be doing over the Republic Day weekend
Making any changes to the type itself is out of question. I ploughed through the code today, adding Dispose() calls, and will see tomorrow if it has had any effect. If not, I'll go with your idea of logging the stack trace by inheriting the class (I hope they didn't seal it ) and replacing all HDElement instances with my subclass.
If you are still working on fixing the leak, I have a basic version of my app here[^]. It uses the CLR Profiling API to track object allocations and finalizations and shows the constructor stack trace for finalized objects.
Just give it the list of fully qualified names (including namespace) of the types you want to monitor and run the application - It will log all data to a file you specify.
It has a couple of kinks - you'll have to give some type name i.e. you can't monitor all types, inner classes don't work if the class name is fully qualified, and the stack trace doesn't include method parameter/return types.
Worth a try if you are still struggling with the leak
Senthil, this is much appreciated. I later realized the numbers I was seeing in task manager were not accurate, and found some articles that demonstrated this (the bookmarks are at work).
I have come around to the conclusion that there is probably no leak, but all that work did not go waste: I was able to bring down the memory usage by more than 25% by rewriting the code to use Dispose() and making a few calls to GC.Collect().
I've downloaded and mailed your app to my office ID so I can play with it on Monday (I don't do any dev work at home).
I've never ever worked anywhere where there has not been someone who given the choice I would not work with again. It's a job, you do your work, put up with the people you don't like, accept there are probably people there that don't like you a lot, and look forward to the weekends.
- Josh Gray.
So I spent 4 hours (2 x 2 hours), trying to figure out why a BindingList was going apeshit.
1. The fool that wrote the generated code (coming from an ORM tool), decided it would be a good idea to fire an event on a separate thread. Problem was easily solved by wrapping the BindingList and redirecting the event to a UI thread.
2. The same fool as above decided that using BeginInvoke (now I could see where the above issue came from), was a good idea to fire off PropertyChanging/PropertyChanged events. Subsequently, LINQ2SQL's object tracker 19/20 times could not detect that a property was changed and would not submit the data.
Problem solved now, hopefully tomorrow less issues
xacc.ide - now with TabsToSpaces support IronScheme - 1.0 beta 1 - out now! ((lambda (x) `((lambda (x) ,x) ',x)) '`((lambda (x) ,x) ',x))