Back to files

For floating point operations, Gnumeric uses the C type gnm_float. Currently that is "double" or "long double", depending on whether Gnumeric was configured with --with-long-double.

This has some coding style consequences...

1. Don't use (say) "exp"; use "expgnum". If you simply use "exp", the argument will be cast from gnm_float to double, thus potentially losing precision and the answer will also have reduced range and precision.
2. For exact, fractional constants, use GNM_const(273.18). That will expand to either "273.18" or "273.18L", i.e., a constant with the right precision. Don't bother for fractions that are a power of 2, i.e., for 232.25. Also, don't bother for constants that aren't exact to begin with. Note: make very sure you have a decimal point or an "e" in the argument to GNM_const.

   Use GNM_const also for huge numbers, e.g., GNM_const (1e100).

3. Various <float.h> constants have "gnum" counterparts. See numbers.h. For example, GNUM_DIG.
4. Various <math.h> constants have "gnum" counterparts. See mathfunc.h. For example, M_PIgnum and M_LN2gnum.
5. To print gnm_float use one of the formats GNUM_FORMAT_f, GNUM_FORMAT_g, ... defined in numbers.h

---

Don't use... Use...

loggnum (1+x)                log1pgnum (x)
loggnum (1-x)                log1pgnum (-x)
expgnum (x * loggnum (y))    powgnum (y, x)
expgnum (x) - 1              expm1gnum (x)
1 - expgnum (x)              -expm1gnum (x)
powgnum (1+x,y)              pow1p (x,y)
powgnum (1+x,y) - 1          pow1pm1 (x,y)