Re: [Tarantool-patches] [PATCH] lua: introduce function for uuid comparison

[Igor Munkin <imun@tarantool.org>]

Vlad,

I glanced the patch and your review and want to clarify one point.

On 16.11.20, Vladislav Shpilevoy wrote:
> Hi! Thanks for the patch!
> 
> Lets change subsystem from 'lua:' to 'uuid:'. The former is too
> general.
> 
> See 4 comments below.
> 
> On 11.11.2020 21:04, olegrok@tarantool.org wrote:
> > From: Oleg Babin <babinoleg@mail.ru>
> > 
> > Since Tarantool has uuid data type sometimes we want to compare
> > uuid vaues as it's possible for primitive types and decimals. This
> > patch exports function for uuid comparison and implements le and
> > lt metamethods for uuid type.
> > To be consistent with decimal type this patch allows to compare
> > uuid with uuid string representation.
> > 
> > @TarantoolBot document Title: uuid compare
> 

<snipped>

> 
> 4. Hmmmmm
> 
> tarantool> u1 = 'b0b681e9-dd43-4341-98b4-a973b23f1421'
> ---
> ...
> 
> tarantool> u2 = uuid()
> ---
> ...
> 
> tarantool> type(u2)
> ---
> - cdata
> ...
> 
> tarantool> type(u1)
> ---
> - string
> ...
> 
> tarantool> u2 < u1
> ---
> - true
> ...
> 
> tarantool> u1 > u2
> ---
> - true
> ...
> 
> Doesn't this look wrong?

Yes and no. At first result is the same. The issue you mentioned above
relates to cdata comparison, but this is wrong in general. According to
Lua Reference manual[1] the objects can be compared via the metamethods
iff they are the same types. Otherwise the comparison fails. Consider
the following example:
| $ tarantool
| Tarantool 1.10.8-0-g2f18757
| type 'help' for interactive help
| tarantool> a = setmetatable({ a = 'QQ' }, { __lt = function(self, obj) if type(obj) ~= 'string' then error('invalid') end return self.a < obj end })
| ---
| ...
| 
| tarantool> str = 'QKRQ'
| ---
| ...
| 
| tarantool> a < str
| ---
| - error: '[string "return a < str"]:1: attempt to compare table with string'
| ...

Anyway, cdata comparison is badly described on LuaJIT FFI semantics
page[2], but at least there is the following note:
| Comparisons for equality/inequality never raise an error. Even
| incompatible pointers can be compared for equality by address. Any
| other incompatible comparison (also with non-cdata objects) treats the
| two sides as unequal.

At the same time, Mike implemented kinda __rlt for cdata types. In other
words, if the left operand is not cdata, but the right one is, then it
would lookup the right one for __lt. I failed to find the place this
freaking hack is described but this is how it works.

> 
> From what I see here: http://lua-users.org/wiki/MetamethodsTutorial
> they say, that for '<' Lua uses metatable of the left object. It means,
> that if string is on the left, it will use string metamethods. Although
> I may be wrong, and you need to check how it actually works. If it is
> true, I am afraid we can't support string vs cdata-uuid comparison.

Looks like we can, *but* I only checked your example, so I see no other
pitfalls for now. Anyway, I guess I will make a precise description for
the way cdata metamethod are looked up.

> Only cdata-uuid vs cdata-uuid. For string case you may need to add a new
> function. Probably in your own application.

[1]: https://www.lua.org/manual/5.1/manual.html#2.8
[2]: http://luajit.org/ext_ffi_semantics.html

-- 
Best regards,
IM