Re: [swarm-hackers] Memory leak in Swarm

[Marcus G. Daniels]

Steve Railsback wrote:
> A few months ago Steve Jackson went through a big exercise with 
> Valgrind to get memory leaks out of our big fish model. Some of them 
> turned out to be in Swarm, and I finally got him to take a moment or 
> two from his new job as math professor to tell me at least the one 
> that bugged him the most.
>
> Apparently when you call Symbol getName, the method uses a string dup 
> that leaves a string in the scratch zone.
>
> Perhaps Scott or somebody mucking in the code could deactivate that 
> feature?
The current intent is to have all callers of getName _within Swarm_ free 
that object as soon as they are done with it.   There are some cases of 
that as I recall, but I thought they were covered.

If the caller is your code, then you can free that object.

The reason it is allocated in scratch space is because in the case of 
Java (or other foreign languages), there is no preallocated and fixed 
memory for the object.   Foreign objects are potentially being moved 
around by a garbage collector, so any memory handle you get out of them 
is volatile.  Thus it must be immediately copied.