> On Mar 31, 2016, at 22:35, Christiaan Hofman <cmhof...@gmail.com> wrote:
>
>
>> On Mar 31, 2016, at 19:17, Maxwell, Adam R <adam.maxw...@pnnl.gov
>> <mailto:adam.maxw...@pnnl.gov>> wrote:
>>
>> The custom zone is there for performance; Apple’s malloc is extremely fast
>> for repeated use of small blocks, but when I wrote that it was dismally slow
>> for anything like the image tile sizes I needed for scaling. Last I profiled
>> it (10.9 or 10.10), mine was still significantly faster. The only potential
>> problem is if they change the zone layout again, but the last time was in
>> 10.8, and I think there was enough padding in fv_zone to allow that.
>>
>> FWIW, I’m running 10.11, but haven’t done much with BibDesk recently; I’m
>> using my old test version here and have never seen the webkit crashes, so
>> I’ve always assumed that was elsewhere than FileView :-). I’ve seen enough
>> crashes due to a stray autorelease in completely separate code to be pretty
>> skeptical of backtraces in finding a heap smasher. Does anyone have a file
>> with only remote icons that can make it reproducible?
>>
>
> There are some databases linked in some bug reports, where the user says it’s
> reproducible for him. Though not necessarily with only remote URLs. But I
> have never been able to see it myself.
>
> Christiaan
>
>>> On Mar 31, 2016, at 07:49, Christiaan Hofman <cmhof...@gmail.com
>>> <mailto:cmhof...@gmail.com>> wrote:
>>>
>>> Adam Maxwell, you’re still around here? On 10.11 there are quite a few
>>> reports on crashes due to file view icons, due to invalid objects. The
>>> memory management looks to be OK, no overrelease in sight, but apparently
>>> it does happen. I wonder whether this has to do with the custom zone code.
>>> Perhaps in combination with the system,’s support for ARM (even if we don’t
>>> use that). Unfortunately I have no clue about how zones work and what’s
>>> going on in that part. I also cannot reproduce that (for one because I
>>> still don’t have 10.11, in part because that seems to have a pretty buggy
>>> PDFKit). Would you have any idea about that? What’s actually the reason for
>>> the custom zone?
>>>
>>> Christiaan
>
>
I think I found the problem. In several init implementations, when the receiver
is discarded, it calls [super dealloc] to clean up the discarded self. That’s
dangerous, dealloc should never be called except from a dealloc override. It
looks replacing these calls by [self release] (and making some dealloc
implementations safer to be called by uninitialized objects) and let the object
be cleaned up automatically gets rid of the crashes.
Christiaan
------------------------------------------------------------------------------
Transform Data into Opportunity.
Accelerate data analysis in your applications with
Intel Data Analytics Acceleration Library.
Click to learn more.
http://pubads.g.doubleclick.net/gampad/clk?id=278785471&iu=/4140
_______________________________________________
Bibdesk-develop mailing list
Bibdesk-develop@lists.sourceforge.net
https://lists.sourceforge.net/lists/listinfo/bibdesk-develop
