Re: Reducing Lua memory footprint - need help

["Jerome Vuarand" <jerome.vuarand@...>]

2008/10/12 Bogdan Marinescu <bogdan.marinescu@gmail.com>:
> As part of my "Lua for microcontrollers" project (http://elua.berlios.de),
> I'm trying to reduce the memory footprint needed by Lua. With all the eLua
> auxiliary modules enabled, "print(collectgarbage'count')" reports about 25k
> right after the Lua interpreter starts - quite a bit for a micro than can
> only access 64k of RAM. So I'm looking for ways to reduce this.
> The first idea was that I don't really need "full tables" for the Lua
> modules, I'd rather have some sort of "read only tables" that can read their
> keys directly from Flash instead of RAM (keep in mind that RAM consumption
> is "more important" than execution speed here). The obvious implementation
> was to call "luaL_register" with a table that contains only the "__index"
> function, and set the metatable of this table to itself. All table requests
> are thus redirected to the "__index" function, which searches a fixed array
> of {key, value, type (function or number} (that resides in the Flash memory)
> and returns the proper key value and type, or nil.
> I didn't yet implement this for all modules, but I already know that it
> works and that it will have a good impact on the memory consumption. What
> I'm wondering at this point is if I can go even further. The idea presented
> above assumes that each module still needs to call "luaL_register", thus
> each module creates a new table in RAM. What I'd really like is to have a
> _single_ table that can handle the access to _all_ my modules. In other
> words, given the modules m1, m2, ... mn, I'd like to differentiate between
> m1.func(), m2.func(), ... , mn.func() with a single (meta)table. And I don't
> know how (if?) this is possible, given the fact that the (meta)table somehow
> needs to know for what "table" (m1, m2, ... mn) it must do the key search.
> Do you know of any way to implement something like this? I know I can do
> this by introducing a completely new data type (something like ROTABLE) and
> implement the required semantics, but given the fact that I'm not very
> familiar with the Lua code, I'd rather skip this option for now.

Why don't you use userdata as modules ? IIRC modules don't have to be
tables. Your userdata would be zero-sized (or eventually have a simple
pointer to read-only memory), and could share a single metatable,
since the userdata itself would be passed as first argument of __index
to let you determine which module is currently indexed.