Re: gworkspace - ddbd connection problems

[Fred Kiefer]

On 20.02.2013 01:34, Riccardo Mottola wrote:
> While I experimented with removing certain autoreleasepools around
> connections, as tit was suggested here on the list, without results,
> Sebastian empirically fount out that this aptch
>
> Index: Tools/ddbd/ddbd.m
> ===================================================================
> --- Tools/ddbd/ddbd.m   (revision 36176)
> +++ Tools/ddbd/ddbd.m   (working copy)
> @@ -223,14 +223,12 @@
>
>   - (NSArray *)userMetadataForPath:(NSString *)apath
>   {
> -  CREATE_AUTORELEASE_POOL(arp);
>     NSArray *usrdata = nil;
>
>     [pathslock lock];
>     usrdata = [pathsManager metadataForPath: apath];
>     [pathslock unlock];
>
> -  RELEASE (arp);
>
>     return usrdata;
>   }
>
>
> helps a lot, for example it does not crash on MDKit indexing.. But why?
> I cannot see anything bad to instantiate a new autoreleasepool and
> releasing it later. Why woul this change make connections die or not?

That is easy to explain. You get back an autoreleased array from 
metadataForPath:. As there is an autorelease pool around that code, this 
array will get released within the autorelease pool. Next you return 
that now invalid object from your method. The caller will may get an 
invalid pointer.
This is the second basic memory management rule you have to learn when 
coding Objective-C (the first being to match release and retain calls), 
if you need to hold on to a variable retain it.
I don't see why you would need an autorelease pool here, but if there is 
reason for it, you need to retain usrdata when assigning it and 
autorelease it in the return line.

Hope this helps,
Fred