[Gambas-user] array size limitations relative to available RAM and swap space

[Doriano Blengino]

kevinfishburne ha scritto:
> Using GAMBAS 2.19 I tried creating an array:
>
> PUBLIC Elevation AS Short[65536, 65536]
>
> With no data assigned to any of the array's elements, reading Elevation[0,0]
> shows a value of 24. I'm thinking it should be zero. Reading
> Elevation[4095,0] crashes the program. I tried reading Elevation[100,100]
> from the debug window and it also crashes the program.
>
> Granted this array if fully allocated would consume 8 gigabytes of RAM. My
> system has 4 gigs of RAM and an 8 gig swap file. Does GAMBAS use the swap
> file if enough physical RAM isn't available? What are the size constraints
> of arrays? Why would Elevation[0,0] show up as 24 when Elevation[1,1], etc.,
> are zero as they should be?
>   
Probably there is some bug here, but anyway I would use some other mean 
to achieve the goal.

This data take 4 Gib cells, each 2 bytes (bytes?) wide, for 8 Gib total. 
But if this is a dynamic array, there is the pointer to each cell to 
consider: 4 bytes for every cell on 32 bit systems, and probably 8 bytes 
for 64 bit systems; it leads to 4 Gcells * 6 bytes, (24 gigabytes), or 
40 Gib on 64 bit systems.

If you really have 65536*65536 cells, all alive together, you could use 
a binary file on disk. Disk caching will speed up things, perhaps better 
than fake ram swapped in/out from disk.

If you don't have all the cells live together, you could use an 
association, and only keep in ram the cell along with its "coordinates".

A database could also make sense.

If you insist on the ram approach, then a static array would consume 
less memory, if it is still supported.

This is an unauthoritative thought.

Regards,
Doriano