Re: [ANN] A new implement of Luiz's lmapm module

[Luiz Henrique de Figueiredo <lhf@...>]

Nice to see lmapm attract some interest.

> this is a new implement of Luiz's lmapm module:
> - renamed some functions:

If you change the API, then it's not a reimplementation of lmapm:
it's a new binding to MAPM, which is fine.

> - use static work temporary variables to store intermediate result.

This is a laudable goal, but the way you have implemented it, with
static C variables, your library only works for a single Lua state or
thread. In general, Lua libraries should not have static C data that
depends on the Lua state.

In your case, you could allocate temporary variables in the registry,
but I'm not sure it's worth the trouble.

For a C library whose objects have to be created by the library it's
hard to avoid creating garbage. Some C libraries allow the user to init
objects using local variables and this makes it easier to avoid *some*
of the garbage when evaluating complex expressions. For instance, my
bindings to qd and interval support this.

However, the main source of garbage is intermediate results and this
cannot be controlled automatically unless you do something like

	begin_evaluate()
		a=<complicated expression>
	end_evaluate()

and allow the binding to keep a stack of objects that is reused at
the end; only the last object is keep.

I think we have discussed this scheme in the past here (or was it in IRC?).

Anyway, in a garbage collected language like Lua, you should not be (too)
concerned about garbage, unless you create lots of it. Moreover, the
generational mode in Lua 5.2 can probably mitigate this problem.