Re: [Groff] Bug in pic? And how to do this right?

[Ralph Corderoy]

Hi Ted,

> This seems to be a bug in grops.
> ...
> Groff version 1.16.1:
> 
>   288 -322.656 720.002 101.5369 78.4631 DA
> ...
> Now, from the correct values:  101.5369 78.4631
> 
> we can get the groff 1.18.1 values:  102 78.5 by rounding to 3
> significant figures. Maybe this is what recent versions of grops do

Thanks for all the hard work.

    $ cvs diff -r groff-1_16_1 -r groff-1_18_1 grops
    ...
    --- grops/ps.cc 18 Jun 2000 10:13:56 -0000      1.4
    +++ grops/ps.cc 20 Jun 2002 05:58:36 -0000      1.20
    ...
    @@ -310,7 +321,7 @@
     ps_output &ps_output::put_float(double d)
     {
       char buf[128];
    -  sprintf(buf, "%.4f", d);
    +  sprintf(buf, "%.3g", d);
       int len = strlen(buf);
       if (col > 0 && col + len + need_space > max_line_length) {
         putc('\n', fp);
    @@ -345,6 +356,26 @@
       return *this;
     }

Version 1.13 of that file has the above change, so the CVS tags suggest
groff 1.18 was the first release with the bug.

    groff-1_18: 1.20
    groff-1_17_2: 1.12

The comment for the change includes

    revision 1.13
    ...
    * src/devices/grops/ps.cc (ps_output::put_float): Use `%g' to have
    trailing zeroes removed.

But as you say switching from %.4f to %.3g has problems.

    $ awk \
    > 'BEGIN{for (a=.0006;a<70000;a*=10){b=a+1234;printf"%.4f %.3g\n",b,b}}' \
    > </dev/null
    1234.0006 1.23e+03
    1234.0060 1.23e+03
    1234.0600 1.23e+03
    1234.6000 1.23e+03
    1240.0000 1.24e+03
    1294.0000 1.29e+03
    1834.0000 1.83e+03
    7234.0000 7.23e+03
    61234.0000 6.12e+04

If trailing zeroes particuarly add bloat then a

    s/0+$//
    s/\.$//

to the output of %.4f might be worthwhile.

Cheers,


Ralph.