[Compiler-sig] Memory leaks

[Neal Norwitz]

On 10/25/05, Jeremy Hylton <jeremy at alum.mit.edu> wrote:
> Would it make sense to have the constructors own their arguments even
> when they fail?  That is, have them free the arguments if they can't
> constructor an object?  Or have a variant of the constructor functions
> that does behave this way?

This can't work if the parameter is an asdl_seq, since we don't know
if the sequence could contain stmts, exprs, comprehensions, etc.

I kinda like the idea, but I think arenas will be better in the long
run.  I think using PyObjects might be too slow, but I'd love to be
proved wrong.  It would make the Python interface much easier.

n