Re: [PATCH v2 2/2] lua: add key_def lua module

[Alexander Turenko <alexander.turenko@tarantool.org>]

Thank you, it works like a charm.

I added a fixup commit on top of your patchset (added a test case,
updated comments in the test a bit). Also please consider comments
below.

Vladimir, I CCed you to ask a question at end of the email (the code is
on kshch/gh-4025-lua-key-kef-methods branch).

WBR, Alexander Turenko.

On Wed, Mar 27, 2019 at 05:29:28PM +0300, Kirill Shcherbatov wrote:
> There are several reasons to add this module:
> 
> * Factor out key parts parsing code from the tuples merger (#3276).
> * Support comparing a tuple with a key / a tuple, support merging
>   key_defs from Lua (#3398).
> * Support extracting a key from a tuple (#4025).
> 
> The format of `parts` parameter in the `key_def.new(parts)` call is
> compatible with the following structures:
> 
> * box.space[...].index[...].parts;
> * net_box_conn.space[...].index[...].parts.
> 
> A key_def instance has the following methods:
> 
> * :extract_key(tuple)           -> key (as tuple)
> * :compare(tuple_a, tuple_b)    -> number
> * :compare_with_key(tuple, key) -> number
> * :merge(another_key_def)       -> new key_def instance
> * :totable()                    -> table
> 

I would add here 'Functions that accept tuple(s) also allow to pass Lua
table(s) instead'.

> +static int
> +lbox_key_def_compare(struct lua_State *L)
> +{
> +	struct key_def *key_def;
> +	if (lua_gettop(L) != 3 || (key_def = check_key_def(L, 1)) == NULL) {
> +		return luaL_error(L, "Usage: key_def:"
> +				     "compare(tuple_a, tuple_b)");
> +	}
> +
> +	struct tuple *tuple_a, *tuple_b;
> +	struct tuple_format *format = box_tuple_format_default();
> +	if ((tuple_a = luaT_tuple_new(L, 2, format)) == NULL ||
> +	    tuple_validate_parts(key_def, tuple_a) != 0)
> +		return luaT_error(L);
> +	tuple_ref(tuple_a);
> +	if ((tuple_b = luaT_tuple_new(L, 3, format)) == NULL ||
> +	    tuple_validate_parts(key_def, tuple_b) != 0) {
> +		tuple_unref(tuple_a);
> +		return luaT_error(L);
> +	}
> +	tuple_ref(tuple_b);

Consider the case when a user get tuples from a local space (or merger)
and they have a format that allows to compare faster using precalculated
offsets. I think we should not create a new tuple(s) in the case.

Applicable for other functions too.

> diff --git a/src/box/tuple.h b/src/box/tuple.h
> index 8b12fd5a8..faa42fdf7 100644
> --- a/src/box/tuple.h
> +++ b/src/box/tuple.h
> @@ -672,6 +672,39 @@ tuple_field_by_part(const struct tuple *tuple, struct key_part *part)
>  				       tuple_field_map(tuple), part);
>  }
>  
> +/**
> + * Check that tuple match with the key definition.
> + * @param key_def Key definition.
> + * @param tuple Tuple for matching.
> + * @param allow_nullable True if nullable parts are allowed.
> + *
> + * @retval 0  The tuple is valid.
> + * @retval -1 The tuple is invalid.
> + */
> +static inline int
> +tuple_validate_parts(struct key_def *key_def, struct tuple *tuple)

I don't sure it worth to inline this function: it is not so lightweight
as, say, a structure field access.

I'm tentative whether this function should be in tuple.[ch] or
key_def.[ch]. What do you think?

(If it is in tuple.[ch] maybe it is better to let a tuple being the
first parameter?)

Maybe we need to ask Vladimir (CCed).