Re: xdvipdfmx fails with some regtests (“Invalid object”)

[Han-Wen Nienhuys]

On Thu, Jun 18, 2020 at 7:45 PM Jonas Hahnfeld <hahnjo@hahnjo.de> wrote:
>
> Am Donnerstag, den 18.06.2020, 11:21 -0600 schrieb Carl Sorensen:
> > On Thu, Jun 18, 2020 at 11:04 AM Jonas Hahnfeld via Discussions on
> > LilyPond development <lilypond-devel@gnu.org> wrote:
> > > Am Donnerstag, den 18.06.2020, 00:23 +0200 schrieb Valentin Villenave:
> > > > On 6/17/20, Valentin Villenave <valentin@villenave.net> wrote:
> > > > > `make doc’ has been broken for nearly a week on my system (even with a
> > > > > clean git clone), with the following error:
> > > > >
> > > > > xdvipdfmx:fatal: typecheck: Invalid object type: -1 7 (line 2161)
> > > >
> > > > Git bisect actually tells me that xdvipdfmx started misbehaving from
> > > > the same commit that caused gs issues:
> > > >
> > > > 017927b4d63c317e1fc450be2537ccc058072538 (HEAD)
> > > >     Author: Han-Wen Nienhuys <hanwenn@gmail.com>
> > > >     Date:   Fri Jun 5 20:36:42 2020 +0200
> > > >     Unify calling convention for command-line and API GS use
> > > >
> > > > Jonas reverted some parts of that with MR !148, but that particular
> > > > issue was left unaddressed.  That affects xdvipdfmx svn/20190225,
> > > > which is shipped with Fedora 32, and with a more recent build from the
> > > > 2020 texlive release (20200315). (Now, I haven’t been able to
> > > > reproduce it on LilyDev so other libraries may be a factor as well.)
> > > >
> > > > It already took me many hours to find the eight regtests that trigger
> > > > this, but there are also many snippets and NR ly blocks, which I won’t
> > > > be able to track down :-/
> > >
> > > I can reproduce this kind of error with an empty .eps file, I'm
> > > attaching a (very minimal) example. The .ps file mimics the input that
> > > LilyPond passes to Ghostscript since the commit mentioned above.
> > >
> > > So the following fails:
> > > $ gs -dNODISPLAY -dNOSAFER -dNOPAUSE -dBATCH -dEPSCrop 
> > > -dAutoRotatePages=/None -dPrinted=false broken.ps
> > > $ xelatex include.tex
> > > but
> > > $ pdflatex include.tex
> > > succeeds.
> > >
> > > Likewise the "old" way of doing
> > > $ gs -dNOSAFER -dNOPAUSE -dBATCH -dEPSCrop -dAutoRotatePages=/None 
> > > -dPrinted=false -sDEVICE=pdfwrite -sOutputFile=broken.pdf broken.eps
> > > $ xelatex include.tex
> > > is fine, so it's likely something between Ghostscript and XeTeX (on
> > > Arch Linux: GS 9.52, TeXlive 2019, xdvipdfmx 20190503).
> > >
> >
> > is it the difference between an output .ps file and an output .eps file?
>
> No, broken.ps file is only the driver for Ghostscript:
> mark /OutputFile (broken.pdf) (pdfwrite) finddevice putdeviceprops setdevice 
> (broken.eps) run
>
> Both ways use the same broken.eps file:
> %!PS-Adobe-2.0 EPSF-2.0
>
> Yes, it's empty except for that line.

That doesn't look like an EPS file that LilyPond should be producing.

Let me do some research today where that comes from.


-- 
Han-Wen Nienhuys - hanwenn@gmail.com - http://www.xs4all.nl/~hanwen