Re: Integrated hashtables, take 3

[Neil Booth <neil at daikokuya dot demon dot co dot uk>]

Fergus Henderson wrote:-

> I think that should be done in c_init() rather than in c_init_options().

Yes.  There are all kinds of nasty ordering dependencies in the C front
ends at present; I'm not sure if this is one of them.  I want to work
towards removing the (unnecessary) odering dependencies altogether.

See a future patch I'll post on creating global structures that
parameterize the front ends, rather than the current mishmash of
global variables.  Currently I think the only way someone creating a
new front end knows what needs doing is by making an attempt, and then
trying to figure out the endless link errors [I imagine, anyway].  If
we put them in a few well-chosen global strucutres, it solves all
kinds of issues we have work arounds for at the moment, as well as
being better documentation of GCC's what-it-means-to-be-a-front-end
contract.

> As a matter of style, I would recomment putting the macro
> arguments in parentheses:
> 
> 	s/c - 113/(c) - 113/

Yes, I'll fix this in a followup.  I don't think the (now committed)
patch compiles if you -DGATHER_STATISTICS either.  Conditional
compilation is bad.

> I know you only moved this code around...
> but what's the rationale for not just using sqrt()?

I can't remember - Zack and / or Joseph should know.  There was a good
reason at one time; presumably not wanting to pull in all of libm.

Neil.