Re: scm_num2float() ?

[Martin Baulig]

Michael Livshin <address@hidden> writes:

> Martin Baulig <address@hidden> writes:
> 
> > what do you think about adding something like
> > 
> >     float scm_num2float (SCM a, const char *why);
> > 
> > either as function or as macro which does the typechecking
> > to make sure that the value fits into a C float ?
> 
> why would a number _not_ fit in a C float?

For instance 1.0e40 won't fit into a C float - since MAX_FLT is normally
about 3.8e38.

> should `scm_num2float' check whether there will be a loss of precision
> compared with double?  I'm not sure such a check is very useful (and
> I'm not sure its result would be predictable, either).

No, this shouldn't be necessary.

I was thinking of something like the following:

====
#define SCM_VALIDATE_FLOAT_COPY(pos,z,cvar) \
  do { \
    double x; \
    SCM_ASSERT (SCM_INUMP (z) || SCM_BIGP (z) || SCM_REALP (z), z, pos, 
FUNC_NAME); \
    x = scm_num2dbl (z, FUNC_NAME); \
    SCM_ASSERT_RANGE (pos, z, (- G_MAXFLOAT <= x) && (x <= G_MAXFLOAT)); \
    cvar = (float) x; \
  } while (0)
====

-- 
Martin Baulig
address@hidden (private)
address@hidden (work)