Re: Lua strings as "atoms"???

[Rici Lake <lua@...>]

On 17-May-05, at 8:57 AM, Rici Lake wrote:

> >  And I bet there are a few more pitfalls.
>
> Possibly. But it might still be worth it.

One of the pitfalls of "short strings" is that it would require a 
change in the way the lexer recognises reserved words. (All reserved 
words are short.) Currently, the lexer uses an otherwise unused byte in 
the string structure which holds the token type for the reserved word; 
this avoids a lookup in a reserved word hash table (or direct 
recognition of the reserved words through a state machine.) On the 
other hand, the tokens read by the lexer have to be hashed into the 
string intern table, so it's not actually avoiding a hash lookup 
altogether; simply avoiding a second one. Since short strings would 
never need to be hashed into the string intern table, this would not 
necessarily be slower.

One of the big gains, on the assumption that Lua workspaces are full of 
short strings (many of mine are, for example), would be reduced load on 
the garbage collector. A short string would not even have to be marked, 
and would not require either malloc()ing or free()ing. I think this 
alone may be enough of a win to make it worthwhile. But we'd never know 
unless someone tried it.