From METCALF@crnvma.cern.ch Mon Mar 7 09:52:25 1994 Received: from vm.uni-c.dk by dkuug.dk with SMTP id AA27848 (5.65c8/IDA-1.4.4j for ); Mon, 7 Mar 1994 09:52:25 +0100 Message-Id: <199403070852.AA27848@dkuug.dk> Received: from vm.uni-c.dk by vm.uni-c.dk (IBM VM SMTP V2R2) with BSMTP id 3412; Mon, 07 Mar 94 09:50:55 DNT Received: from CERNVM.CERN.CH by vm.uni-c.dk (Mailer R2.07) with BSMTP id 1606; Mon, 07 Mar 94 09:50:51 DNT Received: from CERNVM.CERN.CH (NJE origin METCALF@CERNVM) by CERNVM.CERN.CH (LMail V1.1d/1.7f) with BSMTP id 4630; Mon, 7 Mar 1994 09:50:40 +0100 Date: Mon, 07 Mar 94 09:45:10 WET From: Michael Metcalf Subject: Stampede to C++?? To: sc22wg5@dkuug.dk X-Charset: ASCII X-Char-Esc: 29 We hear a lot about the inroads C++ is making, so I tried to assess what is really happenning by putting the following request onto comp.lang.c++: As part of our investigation of languages for future use at CERN, we are interested in discovering whether there is any significant experience in the use of C++ for scientific programming. In particular, we would like to hear (by e-mail) from anyone who has worked on a program that fulfills the following conditions: i) is written entirely in C++ (i.e. no calls to Fortran kernels); ii) uses extensively C++ features (i.e. is not simply C compiled using a C++ compiler); iii) has produced published results (i.e. is not just a prototype); iv) is at least 10,000 lines long. I received exactly two replies (below), only one of which is a solid long-term application. What conclusions can we draw? Regards, Mike ======================================================================== 52 My first project in C++ falls a little bit short of your criteria ii) and iv) but it did produce publishable results. Now that I think of it, it was quite a feat to get that large of a C++ program to work and still get my results out in time to present at a conference. The program determined the optimum pair of materials and layer spacings for multilayer x-ray mirrors for any given wavelength and angle of incidence. It ran for 49 hours on my 16MHz PC to complete one design, and it had a wild pointer in it that I didn't discover until later (my results were valid - I got lucky). You might want to contact Jack Dongarra (dongarra@cs.utk.edu) who is writing a C++ version of LAPack (LAPack++). It will use the Fortran BLAS3 library, but all higher level programming is done in C++. I believe this is the right approach to take because it is undesireable to have more than one version of source code to maintain for a library. I don't think he would be doing this unless he believed that there were important benefits to be had by using C++. There are two other languages you should probably also consider: Sather and Eiffel. Sather is described in an article in the Oct. 1993 Dr. Dobbs Journal. It is freely available. Sather is an optimized subset of Eiffel. I've had my doubts about Eiffel, but Robert Howard (rock@tower.com), the chairman of the Eiffel committee, recently said that the problems with numeric computing in Eiffel have been addressed (at least partially). Dean Schulze ======================================================================== 79 A consortium of 7 different radio astronomy observatories from around the world is building a system for processing radio astronomy data that is written in c++. This package, called aips++ (astronomical image processing system, incremented by 1) will replace an older 'classic' aips that was written in FORTRAN 66 (ugh!). There are a few low-level fortran components in this package (FFTs, BLAS, and so on) but probably more than 95 % of the package is written in c++, and the applications developer will never use FORTRAN. Advantages of c++ : once you get your object 'right', the development of applications should be much easier because your class libraries will contain classes which accurately reflect your 'world'. However there is no free lunch, because defining the right classes can be a long and complicated process. The aips++ project was started at the beginning of 1992, and we will only be releasing our first 'beta' version with working applications to astronomers at the end of this year. Astronomers and developers at the different observatories had difficulties on agreeing on uniform definitions of classes. Also, c++ is a complex language; compiler writers are having trouble producing compilers which can produce reasonable sized executables, allow you to debug, etc. Templates are the main problem; compilers seem to have trouble producing specific instantiations and compilation, in the worse case, can take hours. Most of the aips++ development has taken place with Sun workstations and we have been able to develop some custom scripts to help cut down on the compile time. However problems do remain - last week I created an executable which was 45 Mb in size! Despite some teething problems with c++, I think object-orientation is definitely the way to go. Once you have a good class library it is definitely quicker to develop applications. There do remain questions about the speed of c++ applications relative to FORTRAN, and as I indicated above, we intend to leave the necessary hooks in place in aips++ to let it call FORTRAN for serious number crunching. If you are interested you can ftp an early release of the aips++ library from aips2.cv.nrao.edu in /pub/aips++/RELEASED/libaips-3 Tony Willis