Re: Lua strings as "atoms"???

[Chris Marrin <chris@...>]

Ben Sunshine-Hill wrote:

> It makes the C code a bit uglier, but I try to use upvalues for all my
> Lua string constants in C functions that need to be efficient. By
> pre-interning the strings, string equality is reduced to a pointer
> compare. Using the registry is another approach, and one that looks a
> lot cleaner, but isn't quite as efficient.

That's a good idea. But why is the registry less efficient? Can't you 
store a string in the registry and get back a numeric id that can be 
handed back to the registry to get pushed just as fast as an upvalue? Or 
is there some optimization in upvalues that make them faster?

The other question is still outstanding, though. If I give a string to 
Lua, will it always give me back the same address in every call to 
lua_tostring() for that same sequence of characters? Does this work 
across GC cycles, for instance?

--
chris marrin                ,""$,
cmarrin@arch.sel.sony.com b`    $                             ,,.
(408) 955-3049          mP     b'                            , 1$'
Sony    ,.`           ,b`    ,`                              :$$'
     ,|`             mP    ,`                                       ,mm
   ,b"              b"   ,`            ,mm      m$$    ,m         ,`P$$
  m$`             ,b`  .` ,mm        ,'|$P   ,|"1$`  ,b$P       ,`  :$1
 b$`             ,$: :,`` |$$      ,`   $$` ,|` ,$$,,`"$$     .`    :$|
b$|            _m$`,:`    :$1   ,`     ,$Pm|`    `    :$$,..;"'     |$:
P$b,      _;b$$b$1"       |$$ ,`      ,$$"             ``'          $$
 ```"```'"    `"`         `""`        ""`                          ,P`
"As a general rule,don't solve puzzles that open portals to Hell"'