New bug :-( Re: Release plan for gnuplot 4.0

[Ethan M. <merritt@u.washington.edu>]

On Friday 13 February 2004 06:03 am, Petr Mikulik wrote:
> I think there is no serious bug for 3.8k -- bug please check it.

Just found one:

gnuplot> set term pdf
gnuplot> set output 'test.pdf'
gnuplot> load 'electron.dem'
Hit return to continue
Hit return to continue
PDFlib value error: floating point value too large in pdf_ftoa
<program exits>


I have not yet tracked down the true origin of the error, but
the specific trigger of the error is the following problematic
bit of code from graphics.c:

/*{{{  map_position, wrapper, which maps double to int */
void
map_position(pos, x, y, what)
struct position *pos;
unsigned int *x, *y;
const char *what;
{
    double xx, yy;
    map_position_double(pos, &xx, &yy, what);
    *x = xx;
    *y = yy;
}

The problem is that in this case map_position_double is returning
negative numbers in xx and yy.  These are then stored as unsigned
in x and y and nothing good can happen after that.

I am pretty sure that the negative values are themselves a
result of some upstream error that I haven't found yet.
Maybe someone else can spot it?

By the way, this error is in principle not specific to pdf output.
But the other terminal types that I have tried just ignore subsequent
moveto/lineto requests as being outside the plot boundaries, wheres
the pdflib complains and exits.

-- 
Ethan A Merritt       merritt@u.washington.edu
Biomolecular Structure Center    (206)543-1421
Mailstop 357742
University of Washington, Seattle, WA 98195