need help: some features of Lua don't fit well together

[Konstantin Osipov <kostja.osipov@...>]

Hello,

We're a database embedded into Lua, and struggling with coercing Lua
type system into the data model we adopted.

The data model of the database is a set of lists: every object
(row) is a list of fields, a field can have arbitrary type, i.e.
it can be a scalar or another list/map.

Sometimes we need to automatically convert what a Lua function
returns to a data set, e.g. to save it in the database.

So we need a simple and consistent conversion scheme between
various Lua returns and the data model.

For example: 

return -> [] (an empty set of lists)
return scalar -> [[scalar]] (a set with 1 list, the list has 1 element)
return { 1, 2, 3 } -> [[1,2,3]] (a set with 1 list, the list has 3
elements).

The task looks simple at first, but any simple conversion scheme
breaks on maps and multiple returns. 

Since there is no easy way to distinguish between a Lua map and an
array, we have to traverse each argument of conversion, and check that
its keys are numeric. If all keys are numeric, keys start with 1,
and there are no gaps in keys, it's a list. Otherwise, it's a map and
it must be treated the same way as a scalar - become a list field.

This check is very inefficient, especially on large returns, 
and there seems to be no way to optimize it.

With multiple returns, to put it simply, there is a lot of
confusion.

Should we treat each value of the multiple return as a field or as a list?
Is return with 0 or 1 values a sub-case of returns with many
values? 

If we assume that each value of the multiple return is a list, and together
they form a set, then we can't return sets larger than ~8000 elements,
that's about the limit for Lua stack size (at least this is the
case in LuaJIT, please correct me if plain Lua is differnt).

As long as we need a way to return a set which contains many lists, a
special case is needed for very large sets.

So the final (and best we could do) set of rules became like this:

function convert_stack_to_set(...)
    set = {}
    n = 1
    if the stack is 1 element and it is a Lua table and
        it has only contiguous numeric keys, and keys start with 1:

        for each table value do
            set[n] = convert_value_to_list(value)
            n = n + 1
        end
    else
        for each stack value do
            set[n] = convert_value_to_list(value)
            n = n + 1
        end
    end
    return set
end

function convert_value_to_list(value)
    if value is a Lua table and it has only numeric keys, keys start with 1:
        return value 
    else
        return { value }
    end
end

Bottom line, if we had a quick way to tell apart a Lua table from an array,
or if multiple return object was coercible to a Lua table or array, 
there would be a scheme that worked well and could be easy to understand.
The scheme that we came up with requires full traversal of the Lua
stack before a decision is made and it doesn't seem to be easy to
understand.
Any help on a better set of rules and on optimization of the functions
above is greatly appreciated.

-- 
http://tarantool.org - a NoSQL database in a Lua script