A debian bug report notes for 1.6,

	(string->number "1@a")
	=> seg fault

Looks like the polar parsing code doesn't cope with #f coming back for
the angle part.  Does this change sound right?  It seems to do the
trick.

        * numbers.c (scm_istr2flo): Use SCM_INEXACTP not SCM_SLOPPY_INEXACTP,
        since value can be #f.