On Thursday 06 December 2007 14:57, Thomas Fischbacher wrote:
> Richard Jones wrote:
> > If you want to do this in pure OCaml, probably your best bet would be
> > to just Marshal the structure and count how big it is.  It'll be slow
> > of course.
>
> Actually, the situation that brought up this question is that I have a
> complicated internal data structure which will free 300 MB of RAM if I
> delete it, while serializing it produces a file of 94 MB only...
> So, I would like to have more clarity what is going on here, and which
> part of this data structure eats how much space.

I had never though of measuring the size of a marshalled data structure. Turns 
out its representation of ints can be more concise than the code 
representation though:

# String.length (Marshal.to_string (Array.make 1000000 0) []);;
- : int = 1000025
# String.length (Marshal.to_string (Array.make 1000000 123456789) []);;
- : int = 5000025
# String.length (Marshal.to_string (Array.make 1000000 max_int) []);;
- : int = 9000025

which is probably what you're observing.

Marshalling also handles sharing but that seems to refer to DAGs in memory 
rather than hash consing:

# String.length (Marshal.to_string (Array.make 1000000 0., Array.make 1000000 
0.) []);;
- : int = 16000031

-- 
Dr Jon D Harrop, Flying Frog Consultancy Ltd.
http://www.ffconsultancy.com/products/?e