Unpredictable EPS->PDF rotation (PR#4460)

On 08-Oct-03 arnima@u.washington.edu wrote:

When I create EPS files, they sometimes appear rotated in my LaTeX PDF
document and sometimes they don't. Two examples:

## x1.eps is not rotated in LaTeX
x <- seq(-1, 1, length=100)
postscript("c:/x1.eps", height=3, width=4,
           horizontal=FALSE, onefile=FALSE, paper="special")
plot(x, dnorm(x), type="l")
dev.off()

## x2.eps is not rotated in LaTeX
x <- seq(-2, 2, length=100)
postscript("c:/x2.eps", height=3, width=4,
           horizontal=FALSE, onefile=FALSE, paper="special")
plot(x, dnorm(x), type="l")
dev.off()

By LaTeX PDF, I am referring to the dvips->ghostscript pathway, but
x2.eps
also rotates when distilled into PDF outside LaTeX. I have tried
ghostscript and acrobat, placed on a page or cropped, so the problem
does not seem to be confounded with particular PDF distillation
software.

I've done an experiment or two (and I've read Peter Dalgaard's reply).

Using Red Hat 9, I imported the two files x1.eps and x2.eps as output
by R-1.7.1 into a groff document, both on one page and one above the
other, outputting the result as a PS file, temp.ps.

Using ghostview, the result displayed fine: right way up.

Then distilled this to PDF, using (a) ps2pdf from ghostscript 7.05,
then (b) epstopdf from TeX (though this too calls ghostscript).
In both cases, the entire page displayed in Acrobat Reader rotated
right through 90 degrees, i.e. the wrong way round.

Then I transferred the temp.ps file to two other machines.

[A] On one, SuSE Linux 5.2, I have an elderly copy of PStill (See Frank
Siegert's site http://www.pstill.com ) which actually is a better
distiller than ghostscript provides. The resulting PDF displayed right
way up using acroread 3.0. I then converted to PDF using ps2pdf
(ghostscript 5.10) and again it was the right way up.

[B] On the other, SuSE Linux 7.2, I converted using ps2pdf (ghostscript
5.50) and again it was the right way up when viewed with acroread 4.0.

Then I transferred both PDF files, from [A] and from [B], back to the
original Red Hat 9 machine and viewed each using aroread 5.0, and they
were still the right way up.

It would seem from this that the problem arises with the distillation
using ghostscript 7.05 (I don't have access to Acrobat Distiller).

BY THE WAY: I had a look at the PostScript code in the EPS files from R,
and noticed that the begin-page and end-page definitions use double
invocations of gsave and grestore:

  /bp  { gs gs } def
  % begin .ps.prolog
  /gs  { gsave } def
  /gr  { grestore } def
  /ep  { showpage gr gr } def

This doesn't strike me as necessary, so (in case it might have something
to do with it) I removed one of them in each case:

  /bp  { gs } def
  % begin .ps.prolog
  /gs  { gsave } def
  /gr  { grestore } def
  /ep  { showpage gr } def

to see if it made any difference (it didn't).

I hope this helps to advance the investigation!

Best wishes,
Ted.