asy from texlive-2014 now fails in make install

[ken@…]

I initially noted this at​http://lists.linuxfromscratch.org/pipermail/blfs-dev/2014-December/029088.html

Asy (asymptote) fails while running 'asy' on various files to create its docs. This is not unknown, the _separate_ upstream tarball mentions copying pre-generated docs from, I think, sourceforge if texlive is broken - but I first hit this on the texlive-2014 source which had been working in BLFS-7.6.

After building in qemu (slow!), I now think that either ghostscript-9.15 or (less likely) poppler-0.28.1 is the culprit. I rebuilt LFS-20141124 and BLFS-20141127 with gcc-4.9.1 from LFS-7.6, proved that the problem still exists with the separate upstream tarball, then went back to a copy of my build from just after firefox and repeated with those two packages reverted to 9.14 and 0.26.4 : and that worked.

I would like (if I can get both this and the i686 xindy problem resolved) to use the separate versions of both of these packages, instead of building texlive twice in the "from source" version.

So, I will note that asy links to Mesa (libOSMesa), GLU (libGLU and libGL), fftw (not in the book, but I build it anyway), libtirpc, libgsl, and libsigsegv - I think that those last four are probably optional, and perhaps libtirpc is being pulled in by another lib : I cannot build a working system without tirpc, because all my sources are on nfs.

Interestingly, asy does not link to ghostscript, but 'strings' shows matches for libgs and /usr/lib/libgs.so.

Fedora have asymptote as a separate package (but then, much of their tex stack is separated into small pieces), and they make it depend on fftw, gc, gsl, latex, ghostscript, freeglut, OSMesa.

[ken@…]

For the deps, fftw, libgsl and libsigsegv are definitely optional.

I see that libtirpc is not, in fact, linked to asy from 2.32 (I was looking at the version from 7.6 when I found that : libtirpc is intended to be used if building for mingw, so obviously it was pulled in incorrectly by the older version.

And ghostscript-9.15 is indeed the problem. Unfortunately, turning off gs seems unlikely to work (it is hardcoded in two files for 'nix builds, and one of those is the code to convert eps files to PDFs.

Unfortunately, I cannot work out what values are being passed to gs for axis3.asy, nor can I manage to run the commands listed in the docs/Makefile.

This may take some time.

[ken@…]

Actually, the reported error is deep within included asy files and I am totally lost as to what is going on.

[ken@…]

It turns out to be /bin/gs which is the problem, although I am still none the wiser about what changed.

I've now successfully installed asymptote-2.32 after installing gs-9.14 into /opt/oldgs and changing the entry for defaultGhostscript= in settings.cc. It looks as if only the gs program itself, plus the supporting files from share/ghostscript/9.14 [ certainly, files from Resource/Init/ ] are needed. That is, libgs.so is required for conversion from eps to pdf, but the version from 9.15 is ok.

I plan to take a look at building 9.14 for /usr, with its gs installed to /usr/bin/oldgs. And moving to separate asymptote instead of using whatever was shipped in this year's texlive.

Found some more test examples (using them to create PDFs without involving TeX). It looks as if freeglut might be neded to do 3D graphics with asy - on my first attempt, it told me to install libGLUT and recompile asy. I've now done that, but needed a few changes to the examples before they worked, so I'll need to test that part again. But freeglut might be on its way back from the archives.

[ken@…]

Updates:

1. Arch have a patch to make it work with gs-9.15. I've rediffed it to apply with -p1. Somebody has added what looks like the same patch to the upstream ticket.
2. Freeglut is indeed needed to do 3D graphics with asy, unless you force it to drop back to an old algorithm which sort-of works. I've now got testfiles ;-)

[ken@…]

For some reason, I thought that freeglut had been archived. My mistake.

[ken@…]

Completed in r15327.