value of LDBL_EPSILON

[agoodeno]

In float.h, the value of LDBL_EPSILON is given as equal to the DBL_EPSILON, just to more decimal places. I believe the correct value of LDBL_EPSILON (on IEEE 754 platforms) should be 1.084202172485504434007452e-19L.

My workaround is to add in my own .h file:
#undef LDBL_EPSILON
#define LBDL_EPSILON 1.084202172485504434007452e-19L

Hope to see this easy fix in the next version of Pelles C!

Alan

[TimoVJL]

Pelles C double and long double are same size, 64-bit.
Same thing with msvc.

https://en.wikipedia.org/wiki/Machine_epsilon

https://stackoverflow.com/questions/3478743/trying-to-write-a-code-for-finding-the-machine-epsilon/37379119#37379119

Code: [Select]

FLT_EPSILON:  1.19209e-07
DBL_EPSILON:  2.22045e-16
LDBL_EPSILON: 1.0842e-19


[agoodeno]

Interesting.

Why would MSVC not support long double? It's in the C and IEE-754 standards and the FPU supports it. I suppose Pelles didn't support it either just to remain "compatible" with MSVC.

An algorithm that calculates the minimum or maximum of a mathematical function can theoretically be determined only to within the square root of the machine epsilon, so a minimum/maximum function needs all the precision it can get!

Alan

[TimoVJL]

Why would MSVC not support long double?

They did before Visual C++ and 32-bit.
Some others still supported that 'extended' like gcc, Delphi, ... (FPU 80 bit).

Quite stupid idea from C standard group folks, that allows long double to be double, who really needs that alias ?
Why not just use defines or typedefs, when retard system don't support long double, to support some ancient program codes ?

[agoodeno]

Thanks Timo for your explanation. I guess there just isn't any need to write programs using long doubles in Pelles C or MSVC, since they don't do anything more than doubles do. This bug report can be closed.

Alan