Hi Alexander, I think you forgot big chunk of leak, NhLinqExpression leak reference to constants that used in linq expression. If linq query reference for object it end up in cache and object not released. I wrote simple task case for demostration https://github.com/radiy/nhibernate-core/blob/master/src/NHibernate.Test/Linq/ExpressionSessionLeakTest.cs#L23 . I made over version of fix https://github.com/radiy/nhibernate-core/commit/82706daafcbfe4ec32aca94a51bbbb16369fe83e . Please review.
2014/1/21 Alexander I. Zaytsev <[email protected]> > Hi all, > > I've made a pull request with a fix. > > https://github.com/nhibernate/nhibernate-core/pull/248 > > Please review. > > Best Regards, > Alexander > > > 2014/1/21 Oskar Berggren <[email protected]> > >> >> 2014/1/20 Lauri Kotilainen <[email protected]> >> >>> >>> >>> On Saturday, January 18, 2014 11:44:53 PM UTC+2, Oskar Berggren wrote: >>>> >>>> >>>> Unfortunately, even with that patch applied, the unit test in my other >>>>> post fails -- even though a heap inspection with WinDBG confirms that the >>>>> session is no longer rooted and is eligible for collection. >>>>> >>>> >>>> My attempts with dubious knowledge of Windbg seems to confirm that the >>>> test itself did indeed maintain a reference to the SessionImpl instance, as >>>> a local variable in the stack fram of IsGarbageCollected<TObject>(ref >>>> TObject @object, Action<TObject> useObject). I don't know how to explain >>>> that, given that the test code attempted to nullify the reference. >>>> >>> >>> The second GC.Collect() pass fixed that for me, like Gunnar suggested. >>> I'll take another look at it soon-ish to see if your modified test still >>> behaves the way I think it should. >>> >> >> I added a bunch more, plus some other GC.xxx() methods, didn't help. >> >> >> >>> My first attempt at a fix actually was to discard the _expression >>> reference. However, that didn't work -- IIRC because the results of the >>> translation maintained another reference to the expression. Changing that >>> would have had a considerably bigger change footprint, one that I wasn't >>> comfortable with. >>> >> >> I later tried that too and discovered that Translate() is actually called >> twice and apparently produces different output on the second go. I haven't >> yet figured out what input data (or how) is modified to achieve this effect. >> >> >> /Oskar >> >> -- >> >> --- >> You received this message because you are subscribed to the Google Groups >> "nhibernate-development" group. >> To unsubscribe from this group and stop receiving emails from it, send an >> email to [email protected]. >> For more options, visit https://groups.google.com/groups/opt_out. >> > > -- > > --- > You received this message because you are subscribed to the Google Groups > "nhibernate-development" group. > To unsubscribe from this group and stop receiving emails from it, send an > email to [email protected]. > For more options, visit https://groups.google.com/groups/opt_out. > -- --- You received this message because you are subscribed to the Google Groups "nhibernate-development" group. To unsubscribe from this group and stop receiving emails from it, send an email to [email protected]. For more options, visit https://groups.google.com/groups/opt_out.
