Re: object identity/address and hashing

["Alex Davies" <alex.mania@...>]

It sounds like you're trying to reimplement something Lua already does ;) - 
I hope this is not the case.

spir wrote:
> -1- Is the object identity internally implemented as its actual memory 
> address?
Yes.

> -2- Is there a way to get this value (like python's id() func)
Unfortunately no.  Would be nice to have a topointer in lua - you can make 
your own via the C api very easily.  The odd occasion has popped up that 
I've wanted one..

> -3- I read somewhere that table key hashes (*) are obtained from key 
> adresses (or identities, as opposed to hashing data itself). Is this 
> correct? If yes, what kind of hash formula/algorithm is then used?
Yes, Lua keys are based on table/userdata/coroutine identities, not the 
values they contain.

> -4- If the above is true, then why isn't this the common way, since it 
> seems to have only advantages:
I cannot imagine why any language would do it any other way.

> -5- The simplest way to hash integers is certainly to get the modulo N, 
> where N is the number of indexes (and of "buckets" in case of collisions). 
> What about this simplissim solution?
As the number of buckets is a power of two in Lua, this'd produce a poor 
results on pointers which tend to have patterns in powers of two.  So Lua 
does a modulo of the number of buckets minus 1, which works well.

> -6- By exploring this topic, I discovered (on my computer) a minimal 
> offset of 8 bytes between addresses (I mean they are assigned modulo 8). 
> (Thus, addresses must be divided per 8 before beeing hashed.) Has this 
> something to do with Lua?
This has everything to do with the system you're compiling for.  Lua just 
uses malloc/realloc, which are supposed to return a pointer with the 
property that it's aligned for any use.  As doubles on x86 are slow when not 
8 byte aligned, malloc should always return pointers that are 8 byte 
aligned.  On another system, this might not be the case.

> (*) What's the common name for "return value of a hash func" in english?
> (In french we commonly say "empreinte", meaning _print_ like in "finger 
> print", but there are other words depending on use case.)
We tend to just call it "the hash" or "hash value".