>One of the nice things about GSL is that it's being written by people with > knowledge of numerical computing; it's not trivial to write good > numerical software. It is wonderful. However I think we are speaking about two different questions: 1. Whether ROOT should provide very CERNLIB compliant C++ class methods to facilitate the legacy CERNLIB Fortran based program transition to C++ 2. Whether ROOT needs good numerical libraries. How they should be organized ? For me we should not mix two different questions above within one discussion. For me the answers are obvious. We do need both. However I think we should have two or more separate libraries (In fact there is no library called CERNLIB. There are kernlib packlib mathlib etc. On another hand I think all numerical ROOT libraries should be similar by organization. I think the priority should be for CERNLIB. This way we can enjoy the legacy FORTRAN -based HEP programs those based on very CERNLIB. If CERNLIB will be not provided quickly then people are forced to introduce the expansive custom solutions and this way CERNLIB will be lost forever. It would be pity. Practically we ROOT has several such sort libraries: TMath, TCL, TPhysics. TInverse was offered recently. TMath, TCL and TInverse has similar organization, namely all of them provide the set static methods. I think the low-level libraries )C++ implementation of CERNLIB algorithm) should have this kind of organization. TPhysics is a example of the next more high level organization. It would be nice to see instead of the next "good idea" on this issue one more real method implemented and tested ;-) Cheers, Valeri
This archive was generated by hypermail 2b29 : Tue Jan 02 2001 - 11:50:40 MET