Subj : Re: OO compilers and efficiency
To   : comp.programming
From : Rob Thorpe
Date : Sun Jul 31 2005 02:36 pm

Gerry Quinn wrote:
> In article <1122470945.664579.51800@z14g2000cwz.googlegroups.com>,
> robert.thorpe@antenova.com says...
> > Gerry Quinn wrote:
>
> > > Nonsense.  The C-like bits should be encapsulated in classes with
> > > standard interfaces - there is no reason for them to be exposed to the
> > > main part of the program.
>
> > Ideally yes.  But if you're writing a part of the program in the C
> > style it is probably for performance reasons.  Parts of the code like
> > this are commonly changed as new ideas come along for improving
> > performance.  This means although these parts of the code are behind
> > standard interfaces they're important and their maintainability is
> > important.
>
> That probably means they weren't thought out properly in the first
> place.
>
> I'm sure there are cases where programs are continually modified to
> improve performance.  But surely they should be an exception, given
> that computing power increases all the time, and if performance was
> merely adequate last year, it should be satisfactory this year and more
> than satisfactory next year without any further modification?
>
> Thus in most cases, the programmer should be assigned to a more useful
> task such as improving the user interface, rather than thinking of
> ideas to improve performance and introducing bugs.

I don't think you and I would really disagree with each other if we
were talking about a real program, it's an artifact of talking in the
abstract.

If you can put your C like bit behind standard interfaces, then that's
OK.  If you can't then :
> That probably means they weren't thought out properly in the first
> place.
In my experience most programs aren't well thought out and don't really
have standard interfaces anyway.

Much C-style-C++ still hanging around in programs today dates from old
requirements for performance.  The C has hung around because no-one has
had the time or inclination to replace it, or because new programmer
mistakenly believe it's there for a reason.  In situations like this
programmer should decide if they're C or C++ programmers and move in
one way or the other.

If not, and if you're writing code in C for performance reason then
it's quite likely that's because it _always_ needs to be fast.  If it's
a weak requirement that goes away as time progresses it's likely easier
to write it in C++ anyway and put up with a little slowdown until
computers get faster.

If you have a program like this I think it's often wise to isolate the
performance sensitive part in a separate library or executable.  At the
very least a clear interface must be created to it, even if it's the
only clear interface in the whole program.

.