Re: [boost] Preliminary submission: FC++

From: Brian McNamara (lorgon_at_[hidden])
Date: 2003-09-30 13:42:50

On Tue, Sep 30, 2003 at 11:14:54AM -0400, Neal D. Becker wrote:
> It seems that compiling on this platform sucks up too much memory, and
> would never have completed in any remotely reasonable amount of time. 

Just to be clear, the behavior you are observing is that
 - the majority of the clients compile quickly, but
 - the last few (the ones that use lambda) take forever or fail
right?

Another way to look at it; if instead of
   gmake
you do
   gmake allnl   # ALL, but No Lambda  (do not compile those clients)
the make succeeds, yes?

> I realize that this is really a flaw in gcc, not in fc++ per se, but it
> makes fc++ pretty much unusable on this platform.  I wonder if there is
> any way to improve the situation? 

This is, for the most part, currently a flaw on _every_ platform.
However this is specific to clients which make heavy use of the
"lambda" or "monad" portions of the library.  If you avoid those
portions of the library, or use them sparingly, then FC++ should be
'usable' (and hopefully also useful!).

I have done some work in the lambda implementation, and managed to speed
things up by about 20-30% since the YahooGroups posting.  But I dunno if
there's much low-hanging fruit left to go after; this is likely to
remain a long-term problem until compilers get better.

But again, I want to emphasize that this is only a problem if you make
heavy use of lambda (an FC++ feature that wasn't even available until
about two months ago in the v1.5 release).

-- 
-Brian McNamara (lorgon_at_[hidden])