On 19/11/17 04:00, Adriaan van Os wrote:
Mark Morgan Lloyd wrote:> I think that conventional wisdom is that if somebody's written numerical > analysis code you don't change it gratuitously, since any alterations > will change rounding errors etc. For some reason, that seems to apply > particularly to FORTRAN programs :-) The reason being explained here <https://gcc.gnu.org/onlinedocs/gcc-3.4.6/g77/Floating_002dpoint-Errors.html>
That obviously applies to all languages, I've never come across something which can represent 1/3 or pi exactly. But FORTRAN- or rather the way that people use it- has always seemed peculiarly sensitive, the classic problem being that recompiling with the optimising variant of the compiler produces significantly different results.
Elsewhere I came across discussion of (I think it was) a DEC FORTRAN compiler which produced the wrong results in a block of code only if it followed a comment. There's always been something grubby about FORTRAN compilers, and by now I find myself wondering whether people should even be attempting to design languages which don't compile easily (i.e. using recursive descent or whatever).
-- Mark Morgan Lloyd markMLl .AT. telemetry.co .DOT. uk [Opinions above are the author's, not those of his employers or colleagues] _______________________________________________ fpc-pascal maillist - fpc-pascal@lists.freepascal.org http://lists.freepascal.org/cgi-bin/mailman/listinfo/fpc-pascal