AA Troubles

[dsimcha]

On 5/10/2011 9:54 PM, Jonathan Crapuchettes wrote:
> Hey all,
>
> I have been working with a lot of associative arrays of late and been
> running into some problems with the built-in implementation. It appears
> that very heavy use in an application can cause the garbage collector to
> have issues. Most of the time I have found ways around the problems, but
> the one that I ran into today has made me wonder if there is a less
> error prone/slower implementation of AAs out there for D that doesn't
> pound the GC as hard. I have looked at the aa project at dsource and
> noticed some discussion on the forums from a little while ago, but I
> wasn't sure how up-to-date everything was.
>
> The problem that I ran into appears, according to gdb, to be in
> gc.gcx.Gcx.findPool(). During the execution of my program, the code
> halts at one point with 1 core maxed (i.e. infinite loop) about 6
> minutes into the execution with about 2.1GB of memory allocated. When I
> said that I used the AAs a lot, I mean that most of the memory is
> allocated to classes that are only accessible through the AAs. I wish
> that I could give you an example, but at the point of the halt, the
> program had pulled data from a 60M row database table and run through
> about 500 lines of code.
>
> I know that I usually don't comment in this forum unless I have a
> problem, and I appreciate that I still get replies. Just know that I am
> a great proponent of D and make very heavy use of it in a corporate
> environment doing a lot of cutting-edge economic calculations. Thank you
> to all who have contributed to D!
>
> Thank you,
>
> Jonathan

Two things:

1.  This is one of the main problems my RandAA associative array 
implementation was meant to solve.  It's usable but not completely 
polished yet.

Code:

http://dsource.org/projects/aa/browser/trunk/randAA/randaasealed.d

Docs:

http://cis.jhu.edu/~dsimcha/randaasealed.html

2.  If you're using Linux, try compiling a 64-bit binary.  Since 
pointers now have twice as many bits, false pointers are orders of 
magnitude less probable.