Precision problem with doubles

[cane]

Hi Pelle, I was trying to develop a little program to dump (bin & hex) FPU values (IEEE double precision floating point only). I've compiled it with PellesC 4 and I've noticed a certain lack of precision compared to other compilers (I tried DMC, BCC, VC, OpenWatcom and GCC).

Compile & run the program I've attached and enter 23.85 when prompted.

You'll notice that PellesC will give you these dumps:

IEEE bin: 0 10000000011 0111110110011001100110011001100110011001100110011001
RAW hex:  40 37 D9 99 99 99 99 99

Which are wrong in the last 2 binary digits; while other compilers will give you the correct dumps:

IEEE bin: 0 10000000011 0111110110011001100110011001100110011001100110011010
RAW hex:  40 37 D9 99 99 99 99 9A

Hope this helps.

Thanks, bye.

[Pelle]

It's a known problem with low priority. I might look at it in the future.

Pelle

[cane]

I don't want to sound pedant, but a bug like this seems rather critical, and its correction should be prioritized.
And not only because it prevents me from using PellesC for my present project, but also beacuse:

• the representation is wrong per se;
• error propagation, happening during math operations, produces and amplifies totally unreliable results.

[AlexN]

It's a known problem with low priority. I might look at it in the future.

Until Pelle will patch this problem, you can try to link with msvcrt.lib. In this case your example works correct.