[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

Re: (Partial) implementation of new math encodings

Matthias Clasen <mclasen@sun2.mathematik.uni-freiburg.de> writes:

>> I was really thinking of a comparison of the graphical output: perhaps 
>> more along the lines of using pstopnm on a ps file and then comparing 
>> the resulting pbm files with pnmarith -difference. I don't know if that
>> will produce any useful information (as I haven't tried it yet).

Several months ago, I set up a test suite for ghostscript (for
which I'm an alpha tester) to do this sort of thing: produce
a bitmap representation of the output for a substantial collection
of PostScript files, and then check for differences.  It turned
out to be infeasible: even a change in optimization level on
the same machine could produce substantial differences in the
output that made the diffs useless.  A visual comparision of
the screen images from the two versions would not uncover
any differences.  I discussed this with L. Peter Deutsch, 
the ghostscript developer, but we did not find a suitable
solution.  There is one commercial test suite for PostScript,
but its cost is too high for the ghostscript developers.

Thus, I'm sceptical that diff'ing bitmaps would produce
output that a human could reasonably interpret.  Even though
TeX works internally almost exclusively in 32-bit integer
arithmetic, PostScript renderers have a LOT of floating-point,
and this may lead to more differences than would be expected.
I therefore think a comparison of the output of dvitype or
dv2dt would be a better way, since that would avoid the
introduction of machine-sensitive floating-point arithmetic.

This approach also has the advantage of not requiring software
that is not part of standard TeX distributions.

========================================================================
Nelson H. F. Beebe                  Tel: +1 801 581 5254
Center for Scientific Computing     FAX: +1 801 581 4148
Department of Mathematics, 105 JWB  Internet: beebe@math.utah.edu
University of Utah                  URL: http://www.math.utah.edu/~beebe
Salt Lake City, UT 84112, USA
========================================================================