Re: Improving gcc scalability wrt _extra_ large C files

[Jacques THOMAS <jthomas at cs dot purdue dot edu>]

Hi Neil,

I think I understand what you want to do. The only thing I don't 
understand is how you plan to deal with nested lexical scopes (i.e. 
lexical symbol overloading). If I understand correctly, cc1plus uses a 
stack of lists of identifiers, each layer in the stack corresponding to 
a lexical scope (the binding levels). Am I correct ? If so, how do you 
plan to emulate or replace this stacking mecanism ?

Jacques.

Neil Booth wrote:

> I really don't think we need yet more switches, and we definitely don't
> need more hashtables.  We've already got one for identifiers, and we
> should use it to its fullest, particularly as it's all we need here.
> Why not attach 3 pointers to them like I suggest in another mail?
>
> 1) It uses less space (3 or 4 pointers less per identifier) than even
>   the *current* scheme
> 2) It has constant time (just a pointer-derefence) name lookup.
>   Actually, that's not quite true, the tag case actually is a linked-list
>   traversal, where length(list) is <= 1 99.99% of the time, and I
>   doubt code has ever been written where length(list) > 3).
> 3) We don't pile more kludges on top of others, which is becoming
>   a real problem in our front ends IMO.
> 4) It's a simplification and it's easy to understand.
>
> It'll be more work to do since it implements the symbol table
> differently to how we currently do, but I think it's a much better
> way to go.
>
> Neil.
>