Hi, thanks for all the comments. I will do tests with gcc-4.x and, if the regression is still there, file a bug report upstream.
Heiko On Saturday 10 December 2005 20:03, Thiemo Seufer wrote: > Heiko Müller wrote: > > Dear Thiemo, > > we very much appreciate your work on the gcc-2.95 debian package. > > For us - and probably also for other users in the scientific > > community - the "old" compiler version is still of great value. > > > > We use gcc-2.95 to compile C/C++ code with very large mathematical > > expressions generated by computer algebra software. This involves > > very long (several thousand lines of code) functions to evaluate > > multi-variable polynomial expression resulting from perturbation > > theoretical solutions of physical problems. > > > > We found that gcc-2.95 -Os produces object code of acceptable quality > > within reasonable compilation times. gcc >=3 is less efficient w.r.t. > > compilation time and memory consumption and in many cases even fails > > to compile our codes due to the very long expressions. The C/C++ codes > > generated from the computer algebra software are perhaps unusual but > > not broken. > > Well, gcc 3.x was somewhat disappointing WRT, but I would expect 4.0 > to do better. If 4.x fails for your (valid and standard-conforming) > code, please consider to provide a testcase to the upstream developers. > I'm sure they are interested in it, and long-term it will help you as > well to have a more modern compiler which can handle such cases. > > > Since what we are doing is not so unusual in theoretical physics we are > > probably not alone with these kind problems. Please consider that even > > if no other debian packages would depend on gcc-2.95 many users may > > very much require it. > > Indeed, I got also one private reply which suggested gcc-2.95 is still > an interesting choice for some numerics code. > > > Thiemo