gnuplot-info

[Gnuplot-info] Missing PDF pointtypes with gnuplot 4.1 CVS

[Scott L. <sl...@sl...>]

Using PDFlib Lite 6.0.1 and gnuplot 4.1 CVS 20060923, "set terminal  
pdf" output is missing pointtypes 0 (".") and 1 ("+"). Pointtype 2  
("X") and so on are present. Since test.pdf's only 6 kB, I'm  
attaching it.

This happens on a couple of my machines:

  * Fedora Core 5, x86_64
  * Fedora Core 5, i386

but a couple others are fine:

  * OS X 10.4
  * RedHat Linux 4ES

The last two are running an older version - "Sat Jul  3 00:04:32 CEST  
2004", they claim - but when I try 20060703 on one of the FC5  
machines, I still have this problem.

Also, the "X11" terminal type works fine on all machines.

Any idea what's going on?

-- 
Scott Lamb <http://www.slamb.org/>

Re: [Gnuplot-info] Missing PDF pointtypes with gnuplot 4.1 CVS

[<br...@ph...>]

Scott Lamb wrote:
> Using PDFlib Lite 6.0.1 and gnuplot 4.1 CVS 20060923, "set terminal pdf" 
> output is missing pointtypes 0 (".") and 1 ("+"). 

Pointtype 0 isn't missing --- it's just so small that you have to crank 
up the zoom quite a bit to see it (600% or above).  The only real 
problem in that plot is pt1.  That should be a + or so, but it's a dot, too.

> This happens on a couple of my machines:
> 
>  * Fedora Core 5, x86_64
>  * Fedora Core 5, i386
> 
> but a couple others are fine:
> 
>  * OS X 10.4
>  * RedHat Linux 4ES

> The last two are running an older version - "Sat Jul  3 00:04:32 CEST 
> 2004", they claim - but when I try 20060703 on one of the FC5 machines, 
> I still have this problem.

Too many variables:  3 different types of CPU, 3 OSes, 3 different CVS 
versions of gnuplot (20060923, 20060703, 20040703).   You'll have to 
reduce the variance to get a meaningful result.

For all you've shown so far, this could be a bug in either PDFlib or 
pdf.trm that only shows up under very specific conditions.

Does normal plotting work better than 'test'?  E.g. does 'plot sin(x) 
with points pt1' contain proper '+' symbols?

Re: [Gnuplot-info] Missing PDF pointtypes with gnuplot 4.1 CVS

[Scott L. <sl...@sl...>]

On Sep 25, 2006, at 1:27 PM, Hans-Bernhard Br=F6ker wrote:
> Scott Lamb wrote:
>> Using PDFlib Lite 6.0.1 and gnuplot 4.1 CVS 20060923, "set =20
>> terminal pdf" output is missing pointtypes 0 (".") and 1 ("+").
>
> Pointtype 0 isn't missing --- it's just so small that you have to =20
> crank up the zoom quite a bit to see it (600% or above).  The only =20
> real problem in that plot is pt1.  That should be a + or so, but =20
> it's a dot, too.
>
>> This happens on a couple of my machines:
>>  * Fedora Core 5, x86_64
>>  * Fedora Core 5, i386
>> but a couple others are fine:
>>  * OS X 10.4
>>  * RedHat Linux 4ES
>
>> The last two are running an older version - "Sat Jul  3 00:04:32 =20
>> CEST 2004", they claim - but when I try 20060703 on one of the FC5 =20=

>> machines, I still have this problem.
>
> Too many variables:  3 different types of CPU, 3 OSes, 3 different =20
> CVS versions of gnuplot (20060923, 20060703, 20040703).   You'll =20
> have to reduce the variance to get a meaningful result.

Oops! Yeah, my goal was to try the same version on all machines, but =20
I misread the "2004" as a "2006". Interestingly, 20040703 doesn't =20
build on these new systems; so, something's different...my PDFlib =20
Lite versions (even though both claim to be 6.0.1), my build options, =20=

or whatever. Unfortunately, I don't have a reproducible source RPM of =20=

the older version. Error below:

     term.o: In function `PDF_init':
     ../term/pdf.trm:345: undefined reference to `PDF_open_fp'

so instead I'm trying CVS 20060923 on a previously-working machine. =20
Here we go: On the RedHat Linux 4ES system (which claims to have =20
PDFlib 6.0.1),

* "Sat Jul  3 00:04:32 CEST 2004" works.
* "cvs up -D20060923" does not. (Same problem as the FC5 systems.)

> For all you've shown so far, this could be a bug in either PDFlib =20
> or pdf.trm that only shows up under very specific conditions.
>
> Does normal plotting work better than 'test'?  E.g. does 'plot sin=20
> (x) with points pt1' contain proper '+' symbols?

No. I first noticed the problem with normal plotting.

--=20
Scott Lamb <http://www.slamb.org/>

Re: [Gnuplot-info] Missing PDF pointtypes with gnuplot 4.1 CVS

[<br...@ph...>]

Scott Lamb wrote:

> Oops! Yeah, my goal was to try the same version on all machines, but  
> I misread the "2004" as a "2006". Interestingly, 20040703 doesn't  
> build on these new systems; 

Not really surprising.  PDFlib 6.0.1 is newer than that CVS version,
and made incompatible changes to the API.  One of them is that the 
PDF_open_fp() function has gone away, so gnuplot now has to use a more 
complicated technique to open its PDF output stream.

> so instead I'm trying CVS 20060923 on a previously-working machine.  
> Here we go: On the RedHat Linux 4ES system (which claims to have  
> PDFlib 6.0.1),
> 
> * "Sat Jul  3 00:04:32 CEST 2004" works.

... but that can't be using 6.0.1 (or at least it had to be patched to 
use it).

You may have to use bisection on the date axis (guided by the 'cvs log' 
of pdf.trm to find critical spots) to find the patch that
broke this (starting off the first version that works with PDFlib 6).

Re: [Gnuplot-info] Missing PDF pointtypes with gnuplot 4.1 CVS

[Scott L. <sl...@sl...>]

On Sep 25, 2006, at 3:18 PM, Hans-Bernhard Br=F6ker wrote:
>> so instead I'm trying CVS 20060923 on a previously-working =20
>> machine.  Here we go: On the RedHat Linux 4ES system (which claims =20=

>> to have  PDFlib 6.0.1),
>> * "Sat Jul  3 00:04:32 CEST 2004" works.
>
> ... but that can't be using 6.0.1 (or at least it had to be patched =20=

> to use it).
>
> You may have to use bisection on the date axis (guided by the 'cvs =20
> log' of pdf.trm to find critical spots) to find the patch that
> broke this (starting off the first version that works with PDFlib 6).

I was afraid you'd say that. Back on an FC5 system (where building's =20
much easier). After confusing myself and running "gnuplot" instead of =20=

"src/gnuplot" a bunch of times, I wrote this script:

#!/bin/sh
DATE=3D"$1"
cvs up -D$DATE && ./prepare && ./configure && make clean && make || =20
exit 1
src/gnuplot <<EOF || exit 1
set terminal pdf
set output "test-$DATE.pdf"
test
EOF
mv test-$DATE.pdf ~/public_html || exit 1
echo Check test-$DATE.pdf

20050701 works
20060101 doesn't
20051101 doesn't
20051001 works
20051017 doesn't
20051016 works

 =46rom there,

$ cvs up -r1.58 term/pdf.trm -> works
$ cvs up -r1.59 term/pdf.trm -> doesn't work

----------------------------
revision 1.59
date: 2005/10/17 06:01:02;  author: sfeam;  state: Exp;  lines: +3 -4
Treat pointsize 0 as a dot
----------------------------

--- term/pdf.trm        16 Oct 2005 19:19:35 -0000      1.58
+++ term/pdf.trm        17 Oct 2005 06:01:02 -0000      1.59
@@ -1,5 +1,5 @@
/* Hello, Emacs, this is -*-C-*-
- * $Id: pdf.trm,v 1.58 2005/10/16 19:19:35 sfeam Exp $
+ * $Id: pdf.trm,v 1.59 2005/10/17 06:01:02 sfeam Exp $
   */
/*------------------------------
@@ -659,9 +659,8 @@
      PDF_PathClose ();
      PDF_save(myPDF);
-    if (number < 0) {
-       /* Like the PostScript driver, treat all negative numbers as
-         * 'with dots'. */
+    if (number <=3D 0) {
+       /* Treat all negative point sizes as  dots */
         PDF_dot(x, y);
      } else {
         /* Change coordinate system so the point symbols themselves

--=20
Scott Lamb <http://www.slamb.org/>