Date: Sun, 16 Apr 2006 03:33:35 -0700 (PDT) From: Dan Strick <strick@covad.net> To: freebsd-chat@freebsd.org Subject: Re: Why is not more FreeBSD software written in C++? Message-ID: <200604161033.k3GAXZSK001785@mist.nodomain>
next in thread | raw e-mail | index | archive | help
OOPS ... I accidentally sent off that last message on the subject:
Re: Why is not more FreeBSD software written in C++?
while trying to run ispell on it. In addition to fixing a bunch of
typos that I trust everyone will just ignore, I wanted to add:
I really like C++ but I am not sure how to deal with the performance
problems. They are not so trivial that they can always be ignored.
I have done a few casual benchmarks. Code that uses C++ strings
can run up to 100 times slower than functionally equivalent code
that uses C character arrays (though such an extreme case is surely
pathological). Code that uses C++ iostreams typically runs
somewhere between 2 and 9 times slower than code that uses C stdio.
You don't believe me? Compare
cout << ' ';
to
putchar(' ');
and weep.
Of course most programs presumably spend only a small part of their
runtime on strings and i/o. I wish I could say the the advantages
of C++ are so great as to outweigh any possible performance glitch,
but I can't quite bring myself to do that.
Dan Strick
---------------------------------------------------------------------
P.S. In case you missed the message to which this one applies,
here it is:
Most software is probably written in C rather than C++ because fewer
programmers are familiar with C++. Note that this means C++ programs
are in a practical sense less portable and more difficult to maintain
even though the GNU C++ compiler is a good implementation, widely
available and pretty much standard in all unix-like OS distributions.
It may be just as well that most software is written in C. Programs
written in C probably run only a very few percent slower when compiled
as C++. (I am assuming the programs are written in the common subset
language.) Even language features peculiar to C++ are generally
implemented quite efficiently.
Programs written in a good C++ style naturally use C++ standard library
facilities (classes, private functions, templates) that can be expensive.
Since C++ programmers generally do not consider the underlying
implementations (arguably a very good thing), significant unintended
run-time overhead can result.
For example, I once rewrote a program in C++ and discovered the C++
version ran about 5.5 times slower than the original C version.
My "mistake": I used the C++ string and iostream classes. After
rewriting the C++ program to use C stdio instead of C++ iostreams
and replacing C++ string classes with C character arrays, the C++
program ran only 1.5 times slower than the C program. I must note
that the C++ version was in some ways a lot cleaner than the C version.
The problem is that a major reason for using C++ in the first place is
to take advantage of these specific C++ library features. A major
motivation for the development of C++ itself was to facilitate code
sharing by better isolating main program code from library implementation
details. A C++ program that avoids using simplifying standard library
facilities by reimplementing them is arguably bad C++.
Want to link to this message? Use this URL: <https://mail-archive.FreeBSD.org/cgi/mid.cgi?200604161033.k3GAXZSK001785>