Re: Enhanced tostring

[Geoff Leyland <geoff_leyland@...>]

On 23/10/2009, at 2:07 PM, David Manura wrote:
> On Thu, Oct 22, 2009 at 4:03 PM, Geoff Leyland wrote:
> > Rima has an "enhanced" tostring that I've called repr (I pinched  
> > the name
> > from python, but I don't really know much about python's repr).
> >
> > It's the same as tostring (so there's a corresponding __repr)  
> > except that it
> > takes a second format argument.
>
> That is [1,2].

Python's repr doesn't seem to have a format object?

> I once posted, though have since removed, an idea like
> that on the wiki [3].

Why did you remove it?  I think the hard part of such an undertaking  
would be to get some agreement on what the format object might contain.

> Could someone remind me why Lua stringifies an object o by way of
> getmetatable(o).__tostring(o) rather than o:__tostring()?  Is it
> because the latter is syntactic sugar for indexing operations
> o.__tostring(o) and o["__tostring"](o) so this would lead to
> ambiguities if o represents a hashtable with an indexing operator that
> queries by key?

Certainly for me.

Rima uses proxy objects that save indexing operations for later, so  
they have no fields of their own (they want to catch ALL __index and  
__newindex events), they do have a metatable (with the appropriate  
__index and __newindex) and the metatable has a __tostring and __repr.

These objects can certainly be 'tostring'ed, but o:__tostring() and  
o.__tostring(o) won't work, they'll just get remembered.

> The __methindex proposal [4,5], would probably
> resolve that since it allows differentiating an object's method
> namespace from its field namespace so that o:__tostring() can be
> assigned different behavior from o.__tostring(o).

As it stands, this would work, but I wouldn't rule out Rima wanting to  
record method calls in the future.

> The lack of this
> difference, I suspect, is in part why Lua introduces it's own
> additional namespace (metamethods), thereby replacing o.__tostring(o)
> with a similarly fashioned getmetatable(o).__tostring(o), just without
> the same ":"-like syntactic sugar for doing that, but rather with
> operators and wrapper functions like tostring.  The metamethod concept
> is absent in many other languages, e.g. Python [6], but many of those
> languages differentiate field and method namespaces.  Is this the
> reason that metamethods exist in Lua?

It's certainly one reason why Lua's different, but I'm not sure that  
there's a good argument for which is better.  Personally, I like  
metamethods and the simplicity of how Lua handles 'objects', but  
that's just me.

> If we were to introduce a serialization API (e.g. repr), would we make
> the operation a method or metamethod?

I vote metamethod.  I'd note that Rima's repr might have the wrong  
name - it's not about serialisation (though I s'pose it could be)


Cheers,
Geoff