Ken Moffat wrote:
I think I must have been a very bad person in a previous existence,
which has led to me being reincarnated for this life as someone who
would eventually care about TeX. [ yeah, I'm still reading
discworld, and I was starting to think that sort of explanation
provided the only plausible basis for what I'm seeing ;-) ]
As I said earlier this week, I'm starting to think that testing the
outer reaches of TeX is a fool's game : but I'm probably a fool :-)
Actually, I think it's not TeX per se, but all the other stuff that's
grown up around it. I have in the past read the TeXbook in quite some
detail.
There are other Knuth programs that I found interesting like tangle and
weave, but I haven't looked at them for years.
There are programs like MetaFont if you want to design your own glyphs
and of course tons of programs that are basically macros for Tex.
There are also utilities like dvips and xdvi.
I think everything else is a variation of the above.
Last week, I went back to looking at the things in TeX which I had
NOT been testing. One of those is ConTeXt. Found a straightforward
example on wikipedia, and on my compiled-from-source texlive it works
fine. I'll attach it in case anybody cares - compile it with "context
context" and you should get a PDF - not particularly useful to me,
and both slow to build and somewhat different from other TeX
invocations, but whatever floats your typesetting.
On one of my AMD machines (it is the newest, but somewhat
underpowered for compiling modern software) I decided to install the
binary texlive. Fortunately, I had already saved the x86_64 version
of that, so I can now just untar it in a bind mount if I need to retest
anywhere. But on that machine, I used the initial 2014 installer,
plus whatever downloads that now pulls down.
Then I went back to more interesting subtopics (finding a simple
use-case for ruby, and then back to trying to use more UTF-8 in
bibliographies - that is still a problem, or at least WIP).
Eventually, I wanted to test on the newest AMD machine, to make sure
I hadn't done anything silly in my test scripts.
On that machine, running the x86_64 binary, the context test fails
(some of these lines might have been wrapped when pasting):
mtx-context | warning: no (local) file './context', proceeding
mtx-context | run 1: luatex
--fmt="/home/ken/.texlive2014/texmf-var/luatex-cache/context/a86c089b384a3076dc514ba966a1fac9/formats/luatex/cont-en"
--jobname="context"
--lua="/home/ken/.texlive2014/texmf-var/luatex-cache/context/a86c089b384a3076dc514ba966a1fac9/formats/luatex/cont-en.lui"
--no-parse-first-line --c:currentrun=1 --c:fulljobname="./context"
--c:input="./context" --c:kindofrun=1 --c:maxnofruns=8
"cont-yes.mkiv"
This is LuaTeX, Version beta-0.79.1 (TeX Live 2014) (rev 4971)
\write18 enabled.
This went wrong:
...c089b384a3076dc514ba966a1fac9/formats/luatex/cont-en.lui:108: bad bytecode
register
.
<*>
cont-yes.mkiv
?
On the same machine running i686, it passed. On my older AMD
('phonon') and my i3 SandyBridge the binary passes the test. I've
now compiled from-source on the system which showed the problem (no
point running a broken version), and that too is now fine
After spending a lot of time looking at 'flags' in /proc/cpuinfo I
have now decided that this is the sort of problem you get with binary
software! The texlive binaries are _contributed_ so they use
whatever toolchain happened to be on the machine of whoever
contributed them. In this case, the machine showing the problem does
not have '3dnowext 3dnow' in its flags.
Interesting.
For me, everything else I expected to work, and which I tested from
the contributed binary, did work [ that includes my tests of fonts
and a random number using lualatex ], so in this case it is not a
major failing.
But I think it is worth noting it in the SVN book (ConTeXt appears
to be such a minor interest that I don't think it is worth an erratum
for 7.6). But would a NOTE be overkill ? I'm thinking of something
like -
NOTE:
With all contributed binary software, there may be a mismatch
between the builder's toolchain and your hardware. In most of TeX
this will probably not matter, but in the more obscure areas you
might hit problems (e.g. if your x86_64 processor does not support
3dnowext or 3dnow). In such cases, the solution is probably to
install texlive from source.
That seems pretty straightforward to me.
I don't feel totally happy about that wording, but (as with
different runtime dependencies between i386 and x86_64) it is
something which may change whenever a new version of the binary
install_tl_unx is released.
Or should I just say that anybody using TeX variants is the author
of their own misfortune ? ;-) Seriously, opinions on this would be
welcome.
With the above note, I think users should know that they are on their
own. I think the only reason we even put texlive in the book was to
build documentation for some packages. The number of users that
actually use TeX in a BLFS environment is probably pretty small. I'm
personally comfortable with only the binary installation, so what we
have for the source with appropriate caveats is quite reasonable to me.
-- Bruce
--
http://lists.linuxfromscratch.org/listinfo/blfs-dev
FAQ: http://www.linuxfromscratch.org/blfs/faq.html
Unsubscribe: See the above information page
