Without this setting, when opening a .dvi file with evince:
- If PATH does not contain /opt/texlive/2023/bin/x86_64-linux, it just
fails.
- If PATH contains /opt/texlive/2023/bin/x86_64-linux, the file can be
opened but a stupidly long time is used.
In both cases there is a warning message on the console:
warning: kpathsea: configuration file texmf.cnf not found in these
directories: ... ... ...
Setting TEXMFCNF explicitly fixes the issue (regardless PATH contains
the texlive bindir or not; though I'm not sure why libkpathsea cannot
use a reasonable default).
The texlive binaries are built with RPATH, thus they work fine w/o
ld.so.conf modification.
For other packages using kpathsea we need to symlink libkpathsea.so.6 to
/usr/lib so it can be found.
It seems install-tl-unx does not use texlive shared libraries at all (I
downloaded the tex executable from it and the executable only uses
libc.so.6 and libm.so.6).
neither firefox nor epiphany can download them, and they are not
well maintained, because rarely tested.
This is WIP because the "(HTTP)" part of "Download (HTTP)" will
need to be removed too.
But let's see what users think first...
and update the note for editors after eating my own dogfood :-(
(dependencies I had listed in my test scripts were inadequate for
the two asymptote tests).
This only applies to texlive source, at least for TL2023 the
x86_64 binaries include luametatex. For source the luametatex
sources are not included. In the unlikely case that we have a
user who both builds texlive from source and uses context, let
them know before we release 12.0 in case another luatex
vulnerability is discovered before TL2024.
In 2023 we eventually discovered that the x86_64 binaries for
asymptote had not worked for at least 2 years, and we now have
a workaround for that. My 20230807 files are intended to show
which *collections* need to be added to schme-medium to allow
the various tests to run.
Also create an ENTITY for qt5 or qt5-alternate and use that
to specify the qt5 dependenies.
The complete KF5/plasma packages should use the full qt5 build, but
the lxqt kf5/plasma components only need qt5-alternate.
Well, my script did not handle package names with dashes correctly. So
some User Note links were mistakely removed. Recheck removed links
containing dashes and add them back if applicable.
So we can simply the instruction. Also remove a stale command
explanation referring to "configure script" (we are using meson now).
Simply the explanation for libkpathsea.so symlink in texlive too as we
don't distinguish "old" and "new" evince installations anymore. The
original explanation emphasizing "old" evince installation also does not
make sense to me: we have /etc/ld.so.conf addition which should be
enough for ld.so to find libkpathsea.so.6 when evince is used; and ld.so
only searches for "libkpathsea.so.6" (which is a DT_NEEDED entry in
evince libdvidocument.so), not unversioned libkpathsea.so.
We usually put those additions in the page where we define all the
paths. So move this one to "Setting the PATH for TeX Live". This
also prevents having to edit the texlive script in jhalfs (to
remove the addition when reinstalling).
Grammar.
In explanations, options come after commands. Add comment on
other accepted options which are used by some distros but have
no real benefit in this source build.
Use scripts so that 'mkiv' (based on luatex) will work,
rather than requiring extra package 'luametatex' which
is very different from the rest of texlive.
Move it from before configuring dvisvgm to after installing
texlive. For dvisvgm there is no difference, but this ensures that
an existing version of evince linked against kpathsea can work
after texlive/2022 is replaced by texlive/2023.
