minar@reed.bitnet (Nelson Minar,(???)) (07/03/90)
Just a brief satisfied customer message.. I recently bought Turbo C++ Professional (for $99 even! - advertising mixup), and decided to test it by compiling the largest and ugliest program that I have lying around, fractint v13. Fractint is written in a combination of assembly and C, and it has been organically growing for about 2 years with 30 different people whacking at the source code. It is, to put it mildly, ugly. It only recently was written to compile with Turbo C 2.0 at all, and as far as I know Fractint v13 was written without any knowledge of the new Turbo C++. Anyway, Turbo C++ compiled the entire package (500k + of source code) with hardly any hitches. All I had to do was add in a proper prototype for the comparison function within the call to qsort(). I compiled it with every possibly compilation switch to allow for speed. The resultant executable was 6k (out of 320k or so) larger than the executable distributed as Fractint.exe (compiled, I think, with Microsoft C 5.1 with all the fast optimizations on). I tried out many of the features, and they all worked. I also did a little speed benchmarking. For fractals that took < 10 minutes to calculate (on a 12MHz 286 w/o coprocessor), there was little speed difference between the two executables. Presumably, these fractals used fixed point math that was hand written by the authors in assembly. However, with a fractal that took almost two hours to calculate, Turbo C++'s executable took 20 minutes less time to generate the fractal than the distribution executable took. Presumably this fractal was calculated with the libraries floating point routines, and so Turbo C++'s "fast floating point" (-ff) switch really is fast. This is not the most wonderful of compiler tests, and it is a specialized application. However, running this test did assure me that Turbo C++ 1.0 is a reasonably functioning program that generates good code.