Re: <iosfwd> also standard for std::string declaration?

[Nathan Myers <ncm-nospam at cantrip dot org>]

The standard way is to include <string>.  They elected not
to split it up as in <iosfwd>, and <iosfwd> is not required
to declare the string components.

Does your 300ms describe the first time it is read, or is it the 
stable, repeatable time after it is already in the buffer cache?  
We care about compile time most when we have lots of modules to 
compiler, but string is read from disk typically only once.

If you are serious about reducing compile times, you might like
to investigate "precompiled headers" (PCH), which a search of
the list archives will reveal more about.

Nathan Myers
ncm-nospam@cantrip.org

On Tue, Jun 17, 2003 at 03:07:59PM -0700, landrum Alfred wrote:
> Ok, I just did.  On the Athlon 1800 I'm using, the
> difference between including <string> & not is about
> 300 milliseconds.  Not that much, I admit.  :)
> 
> I started looking at this because I noticed how big of
> a jump in size the .i's had before & after the
> <string> inclusion - about 300k difference.
> 
> Hmm.  Well, just for completeness sake, is there a
> standard way of getting a string declaration?
> 
> --- Nathan Myers <ncm-nospam@cantrip.org> wrote:
> > On Tue, Jun 17, 2003 at 02:23:15PM -0700, landrum
> > Alfred wrote:
> > > I'm trying to reduce my compile times.  I have
> > several
> > > headers that are including <string>, even though
> > they
> > > only need a declaration.    
> > > 
> > > I noticed that <iosfwd> on my RH8 box includes
> > > bits/stringfwd.h.  
> > > 
> > > Does this imply that including <iosfwd> is a
> > standard
> > > way to get a forward declaration of std::string?  
> > 
> > 
> > No, it's a non-standard way.  BTW, have you measured
> > a difference?
> > 
> > Nathan Myers
> > ncm-nospam@cantrip.org
> 
> 
> __________________________________
> Do you Yahoo!?
> SBC Yahoo! DSL - Now only $29.95 per month!
> http://sbc.yahoo.com