Re: big memory leak in GSString

[David Chisnall]

On 8 Jan 2013, at 00:28, Pirmin Braun wrote:

> the substrings of a substring retain their string from which they where 
> splitted off;
> when those 2nd generation substrings are retained, the 1st generation 
> substrings won't get deallocated;
> but they should, they are just intermediate objects;

This is a known difference in implementation between the Apple and GNUstep 
implementations.  GNUstep assumes that strings created as substrings are likely 
to have a shorter lifespan than the original string and so, if safe to do so 
(i.e. the original string is immutable and owns its contents).  This behaviour 
is very useful when tokenising a string, because it means that the memory usage 
is 

Now, in this case, you are looking at very short strings so it's less clear 
that it's an advantage.  If you are on a 64-bit system, then this will use 
GSTinyString, which hides a 7-character ASCII string inside a pointer, so these 
strings are effectively free, their retain count methods are no-ops, and they 
don't refer to the original string.  

We should possibly provide some kind of heuristic that says if the substring is 
less than the size of a couple of pointers that it's better to copy the data 
than to reference it.  

David

--
This email complies with ISO 3103