[Xy-pic] problem with pdf version

[Michael Barr]

Please consider the following file:

\documentclass{article}
\usepackage[all]{xy}
\begin{document}

\let\*\otimes
 $$\xy
\POS(0,0)*+!!<0ex,1ex>{A\* B}\ar@{->}(50,0)*+!!<0ex,1ex>{A\* B}%
\POS(0,0)*+!!<0ex,1ex>{A\* B}\ar@{->}(40,30)*+!!<0ex,1ex>{A\* B}%
\POS(0,0)*+!!<0ex,1ex>{A\* B}\ar@{->}(30,40)*+!!<0ex,1ex>{A\* B}%
\POS(0,0)*+!!<0ex,1ex>{A\* B}\ar@{->}(0,50)*+!!<0ex,1ex>{A\* B}%
\POS(0,0)*+!!<0ex,1ex>{A\* B}\ar@{->}(-30,40)*+!!<0ex,1ex>{A\* B}%
\POS(0,0)*+!!<0ex,1ex>{A\* B}\ar@{->}(-40,30)*+!!<0ex,1ex>{A\* B}%
\POS(0,0)*+!!<0ex,1ex>{A\* B}\ar@{->}(-50,0)*+!!<0ex,1ex>{A\* B}%
\POS(0,0)*+!!<0ex,1ex>{A\* B}\ar@{->}(40,-30)*+!!<0ex,1ex>{A\* B}%
\POS(0,0)*+!!<0ex,1ex>{A\* B}\ar@{->}(30,-40)*+!!<0ex,1ex>{A\* B}%
\POS(0,0)*+!!<0ex,1ex>{A\* B}\ar@{->}(0,-50)*+!!<0ex,1ex>{A\* B}%
\POS(0,0)*+!!<0ex,1ex>{A\* B}\ar@{->}(-30,-40)*+!!<0ex,1ex>{A\* B}%
\POS(0,0)*+!!<0ex,1ex>{A\* B}\ar@{->}(-40,-30)*+!!<0ex,1ex>{A\* B}%
 \endxy$$

\end{document}

Run it through tex and view it with the dvi viewer.  Perfect.  Now run 
the dvi file through dvipdfm and look at in the Adobe viewer.  Suddenly 
the A \otimes B in the middle is much darker than all the others.  If you 
make all but one into \phantom's, this does not happen.  It is as though 
the pdf file has 12 copies of A \otimes B in 12 slightly different 
positions, kind of like Knuth's \pmb macro.  Does this arise from some 
kind of rounding error in xy-pic?  Or is there another explanation?  In 
any case, can it be cured?  My diagxy macros depend crucially on the 
ability to put these things together without problems.

I understand from Markus Schmaus, who called this to my attention, that 
the same problem arises with dvips (although I could not elicit it in 
GS 8.3) and in pdftex (I didn't try it).  This suggests that the viewer 
might be at fault, but who knows.

Michael