Re: [Feature Request?] __key

["Soni L." <fakedme@...>]

On 11/07/16 07:57 PM, Coda Highland wrote:
> On Mon, Jul 11, 2016 at 3:54 PM, Soni L. <fakedme@gmail.com> wrote:
> > After LuaConf 2016, more specifically the Lua stored procedures talk, 
> > I kept thinking about something... They had to modify the Lua VM just 
> > so they can decay userdata into keys when indexing, everything else 
> > could be done with a substitution layer on top of Lua. So, can we get 
> > a __key metamethod *for userdata only* that takes an userdatum and 
> > returns a value to be used for table indexing? For t[k] and t[k] = v 
> > this is pretty simple to understand, but Lua isn't that simple: - How 
> > should rawget() and rawset() work? - Should there be a way to invoke 
> > __index and __newindex without invoking __key? (rawkeyset() and 
> > rawkeyget()?) - Should there be a way to invoke __key without 
> > indexing? (getkey()?) (for use with rawset/rawget) - Or is this the 
> > wrong way to go about it?
> Just to verify: Are we effectively discussing something equivalent to 
> Python's __hash__? /s/ Adam
No. Keys would still be compared with rawequal() and stuff. The only 
thing is that userdata with __key would use the value returned by __key 
as the key.

E.g. if you have a "decimal" userdata (DBs have a "decimal" type), 
t[decimalObj] may decay to t[aPlainOldLuaNumber] if decimalObj is within 
some range, or to t[internedDecimalObj] (lazy interning) otherwise. With 
a "buffer" userdata, perhaps t[bufferObj] decays into 
t[aPlainOldLuaString]. Note that in this last case a mutable type 
(buffer/userdata) decays into an immutable type (string).

This is way better than __hash__; this is more like fusing __hash__ and 
__eq__ but only for userdata.

-- Disclaimer: these emails may be made public at any given time, with 
or without reason. If you don't agree with this, DO NOT REPLY.