On 4/7/2012 12:44 AM, Manu wrote:
On 7 April 2012 01:08, Rainer Schuetze <r.sagita...@gmx.de <mailto:r.sagita...@gmx.de>> wrote: I don't think there are many places in the code where these hints might apply. Are there known ways of hunting down false references? Still, my main concern are the slow collections that stall the application when a decent amount of memory is used. Removing false pointers won't change that, just make it happen a little later. An obvious best-practise is to allocate in fewer-larger blocks. Ie, more compounds and aggregates where possible. I also expect you are doing a lot of string processing. Using D strings directly? I presume you have a stack-string class? Put as many working strings on the stack as possible...
There isn't a lot of string processing involved: tokens take a slice on the original text, and nodes of the AST seldomly save more than the identifier which just the same slice. So the full text always remains in memory, but this is only small part of the actual footprint, the AST is a lot bigger.
The nodes have child and parent references, so you keep the whole AST once there is a false pointer to any node. I could try breaking up this dependencies when I think the AST is no longer used, but that brings me back to manual memory management and thread synchronization (parsing uses std.parallelism).
