Re: Is there a reason why __eq, __lt, __le metamethods enforce returning booleans?

[Rena <hyperhacker@...>]

On Mon, Sep 17, 2018 at 1:50 PM Petri Häkkinen <petrih3@gmail.com> wrote:
>
>
> On 17 Sep 2018, at 13.33, Eike Decker <zet23t@googlemail.com> wrote:
>
>
> Am Mo., 17. Sep. 2018 um 11:04 Uhr schrieb Dirk Laurie <dirk.laurie@gmail.com>:
>>
>> Op Ma., 17 Sep. 2018 om 09:23 het Eike Decker <zet23t@googlemail.com> geskryf:
>>
>> > There are other cases where it would be useful to be able to return objects. For instance a vector math library could define that the comparison operator compare each scalar value individually and returns a new vector containing only 0/1 values.
>> >
>> > But currently, this is not possible (it seems?).
>> > Is there any particular reason why it is important to convert returned values that I don't see?
>>
>> Yes, there is. One can deduce the existence of the reason from the
>> explcit sentence "The result of the call is always converted to a
>> boolean. " in the manual.
>>
>> That does not mean that anybody outside the Lua team knows the reason,
>> but here is a guess.
>>
>> There is no '__ne', '__ge' or '__gt': they are defined implicitly via
>> "__not" and are therefore always boolean. It would be very illogical
>> to break the identity 'a<b == b>a'.
>
>
> I don't see a problem with not having a __gt / __ge metamethod, it currently simply swaps the operators when calling the __lt / __le methods - which is fine for symmetric operations. The missing __not metamethod is probably a good reason why it works how it does right now.
> It feels unsatisfying though. Having a few more metamethods instead and going through a greater length of implementation would be not the problem in the average case I guess.
>
>
> I’d have a use case for these as well. I’m using operator overloading to great success in generating HLSL / GLSL / whatever back-end shader lang variation from shader code written in Lua:
>
> http://petenpaja.blogspot.com/2016/11/shinobi-shader-system.html?m=1
>
> I can do arithmetic ops & bit ops, but not relational ops. For rel ops I have to use functional style e.g. less(a, b) which when mixed with regular infix notation produces hard to read & maintain code.
>
> One day I hopefully have some time to hack a custom version of Lua for the shader compiler...
>
> Petri

Perhaps you can use __index to have infix notation, eg: a:lt(b), or
even someWrapper(a):lt(b).

-- 
Sent from my Game Boy.