From 930cda9304d160cc6b37aa9ff879cbc730fb69d1 Mon Sep 17 00:00:00 2001 From: Jim Meyering Date: Fri, 11 Apr 2003 12:41:44 +0000 Subject: Remove SPC in SPC-TAB sequence. Remove trailing blanks. filter though cppi --- src/spline.c | 76 ++++++++++++++++++++++++++++++------------------------------ 1 file changed, 38 insertions(+), 38 deletions(-) (limited to 'src/spline.c') diff --git a/src/spline.c b/src/spline.c index c6296cd95..0189dec8f 100644 --- a/src/spline.c +++ b/src/spline.c @@ -29,7 +29,7 @@ The available options are: ordinates. The generated abscissas are evenly spaced with spacing given by the argument, or 1 if the next argument is not a number. --f Set the first derivative of the spline function at the left and right end +-f Set the first derivative of the spline function at the left and right end points to the first and second arguments following -f. If only one numerical argument follows -f then that value is used for both left and right end points. @@ -47,7 +47,7 @@ The available options are: -p The splines used for interpolation are forced to be periodic, i.e. y'[0] = y'[n]. The first and last ordinates should be equal. --s Set the second derivative of the spline function at the left and right end +-s Set the second derivative of the spline function at the left and right end points to the first and second arguments following -s. If only one numerical argument follows -s then that value is used for both left and right end points. @@ -75,48 +75,48 @@ Non-numeric data in the input is ignored. /* #define DOUBLE */ #ifdef DOUBLE -# define double real; /* Type used for data storage. */ -# define IFMT "%F" /* Input format. */ -# define OFMT "%18.14g %18.14g\n" /* Output format. */ +# define double real; /* Type used for data storage. */ +# define IFMT "%F" /* Input format. */ +# define OFMT "%18.14g %18.14g\n" /* Output format. */ #else -# define float real; /* Type used for data storage. */ -# define IFMT "%f" /* Input format. */ -# define OFMT "%8.5g %8.5g\n" /* Output format. */ +# define float real; /* Type used for data storage. */ +# define IFMT "%f" /* Input format. */ +# define OFMT "%8.5g %8.5g\n" /* Output format. */ #endif /* Numerical constants: These may be machine and/or precision dependent. */ -#define HUGE (1.e38) /* Largest floating point number. */ -#define EPS 1.e-5 /* Test for zero when solving linear equations. */ +#define HUGE (1.e38) /* Largest floating point number. */ +#define EPS 1.e-5 /* Test for zero when solving linear equations. */ /* Default parameters */ -#define NPTS 1000 -#define NINT 100 -#define DEFSTEP 1. /* Default step size for automatic abcissas. */ -#define DEFBVK 0. /* Default boundary value constant. */ +#define NPTS 1000 +#define NINT 100 +#define DEFSTEP 1. /* Default step size for automatic abcissas. */ +#define DEFBVK 0. /* Default boundary value constant. */ /* Boundary condition types. */ -#define EXTRAP 0 /* Extrapolate second derivative: +#define EXTRAP 0 /* Extrapolate second derivative: y''(0) = k * y''(1), y''(n) = k * y''(n-1) */ -#define FDERIV 1 /* Fixed first derivatives y'(0) and y'(n). */ -#define SDERIV 2 /* Fixed second derivatives y''(0) and y''(n). */ -#define PERIOD 3 /* Periodic: derivatives equal at end points. */ +#define FDERIV 1 /* Fixed first derivatives y'(0) and y'(n). */ +#define SDERIV 2 /* Fixed second derivatives y''(0) and y''(n). */ +#define PERIOD 3 /* Periodic: derivatives equal at end points. */ /* Token types for command line processing. */ -#define OPTION 1 -#define NUMBER 2 -#define OTHER 3 +#define OPTION 1 +#define NUMBER 2 +#define OTHER 3 /* Define error numbers. */ -#define MEMERR 1 -#define NODATA 2 -#define BADOPT 4 -#define BADFILE 5 -#define XTRAARG 6 -#define XTRAPTS 7 -#define SINGERR 8 -#define DUPX 9 -#define BADBC 10 -#define RANGERR 11 +#define MEMERR 1 +#define NODATA 2 +#define BADOPT 4 +#define BADFILE 5 +#define XTRAARG 6 +#define XTRAPTS 7 +#define SINGERR 8 +#define DUPX 9 +#define BADBC 10 +#define RANGERR 11 /* Constants and flags are global. */ int aflag = FALSE; /* Automatic abscissa flag. */ @@ -181,7 +181,7 @@ int errno; char *auxmsg; { char *msg; int fatal, usemsg; - static char *usage = + static char *usage = "usage: spline [options] [file]\noptions:\n", *options = "-a spacing\n-k const\n-n intervals\n-m points\n-p\n-x xmin xmax\n"; @@ -391,7 +391,7 @@ char *str; /* Found an option. */ *str = *++s; str[1] = '\0'; - ++s; + ++s; type = OPTION; } else if (is_number(s)) { @@ -413,7 +413,7 @@ char *str; is_number(string) char *string; { - if (isdigit(*string) || + if (isdigit(*string) || *string == '.' || *string == '+' || (*string == '-' && (isdigit(string[1]) || string[1] == '.'))) @@ -527,7 +527,7 @@ calccoef() ++ptr; /* Initialize rows 1 to n-1, sub-diagonal band is assumed to be 1. */ - for (i=1; i < nknots-1; ++i, ++ptr) { + for (i=1; i < nknots-1; ++i, ++ptr) { h = h1; if ((h1 = data[i+1].x - data[i].x) == 0.) { sprintf(str, "%8.5g", data[i].x); @@ -632,8 +632,8 @@ solvband() else { ptr->rhs = (ptr+1)->rhs - (k1 = ptr->rhs/ptr->upper)*(ptr+1)->diag; ptr->upper = (ptr+1)->upper; /* This isn't super-diagonal element - but rather one to its right. - Must watch for this when + but rather one to its right. + Must watch for this when back substituting. */ (++ptr)->diag = i-1; ++i; @@ -697,7 +697,7 @@ interpolate() struct pair *dp; struct bandm *ep; real h, xi, yi, hi, xu, xl, limit; - + h = (xmax - xmin)/nint; ep = eqns; dp = data + 1; -- cgit v1.2.3-54-g00ecf