[LRUG] Why does test invocation affect result?

[Paul Battley]

I don't know how you're generating the PDFs, so this is slightly
speculative, but it looks like it's producing unique font names by
prefixing them with a six-character incrementing ID. But I don't think
that necessarily matters.

Since you presumably care about the visual output of the PDF rather
than the source, perhaps it would make sense to render them both to
images (perhaps with gs) and compare the pixels. This would also
decouple your visual appearance tests from anything else you might
want to test, like metadata. I'm assuming that the output really is
visually identical, which might of course be a flawed assumption.

Paul.