Weak references.

[Jason House]

Bill Baxter wrote:

> Jason House wrote:
>> Jarrett Billingsley wrote:
>> 
>>> "Jarrett Billingsley" <kb3ctd2 at yahoo.com> wrote in message
>>> news:fttdej$1nfs$1 at digitalmars.com...
>>>> "Sean Kelly" <sean at invisibleduck.org> wrote in message
>>>> news:fttd02$1mp1$1 at digitalmars.com...
>>>>> Are these tables AAs?  And how do the weak references store their
>>>>> reference?
>>>>> The type information provided to AAs is a bit weak, and so some
>>>>> non-pointer
>>>>> data is currently still scanned by the GC.  I'm going to improve this
>>>>> a bit in the
>>>>> next Tango release, but it still won't be perfect (because the
>>>>> compiler doesn't
>>>>> provide enough info for it to be).
>>>> In my tests I've just been doing
>>>>
>>>> auto weakref = new WeakRef!(A)(new A);
>>>>
>>>> and then checking that after doing god-knows-what to try to get it to
>>>> collect that instance of A, so it's not even that I'm doing anything
>>>> that crazy.
>>>>
>>> Testing my code in Linux causes a crash in glibc's free() after the
>>> program
>>> exits.  (And the assertion still fails.)
>>>
>>> Something really weird is happening.
>> 
>> Ok, my latest testing (and testing by others for configurations I don't
>> have) are collected in the list below.  I don't see any pattern.
>> 
>> tango+linux+dmd = works
>> tango+linux+gdc = crash in glibc
>> tango+windows+dmd = does not collect
>> tango+windows+gdc = ???
>> phobos+linux+dmd = works
>> phobos+linux+gdc = does not collect
>> phobos+windows+dmd = ???
>> phobos+windows+gdc = ???
>> 
>> These tests are based off the code at http://tango.pastebin.com/m10aa57b6
>> Note that phobos users must comment out the version=Tango line.  Any
>> additional tests would be appreciated.
>> 
> 
> I'd be worried about stack references in the second example.  That inner
> function makeWa may be getting inlined or something with some compiler
> combos.
> 
> The first unittest there works ok right?  Maybe explicitly overwriting a
> with null after using it would help.
> 
> Anyway, I'm pretty sure Sean said after he added the functions to Tango
> that he thought they should work but they weren't tested very well.  So
> I wouldn't be surprised if there were issues.  And really, Phobos's gc
> functions for weak ref support probably haven't been tested all that
> well either.
> 
> --bb

I'm mostly trying to extract as much info as I can from this example to
better help me file a bug report.  The unit test can't differentiate
between "never cleans up" and "didn't clean up from register values hanging
around"...  Nevertheless, it's a good start for further investigation. 
Certainly the crash inside glibc is a bug in either gdc, tango, or both.