Hey folks,

I've been thinking about implementing a small library with custom data types. Let's say it's a matrix type that's allocated on the heap. I want to pass it around using messages, so, IIUC, I need to use A_POINTER type. Is there a way to know that this atom that I created is no longer needed/referenced anywhere? I lurked the Pd sources for a while but wasn't able to find anything relevant.

Is it possible to implement a reference counting system for a custom data type in Pd? Do you know if anyone has already done this before? How can this be implemented?

Thanks,
Eugene

Instead you could use a pool of reusable tags (as symbols) to refer to (fetch) your objects.

Great idea, thanks @nicolas-danet!

Just curious, does the symbol table in Pd get cleared up somehow? E.g. if I decide to just dump my data types as symbols that are then passed over the patch — will this be a memory leak?

IIRC, a symbol is generated once and stay there until the end. Thus you could theoretically flood the table (and slow down lookup) if you generate too many different symbols. To avoid the bloat you can split/join the data bytes as hexadecimal. That was i did for NTP timestamps for instance ("@e @0 @e @a @4 @f @d @d @3 @a @1 @a @7 @1 @9 @b").

Note that the symbols could be stored by the user and reused later. You need a mechanism to handle that case.

Cool, thanks for the info Nicolas!