Bug 22386

Hi,
thanks for your report, I know that issue. I'm still working on reorganizing that package. But now it's documented and we won't forget before distrib goes final.

FYI: The failing the script is not very severe.

%post           -n texlive-fontsextra
cp %{texmfdistdir}/web2c/updmap-fontsextra.cfg %{texmfdistdir}/web2c/updmap.cfg
yes|%{_bindir}/updmap-sys --syncwithtrees --force > /dev/null 2>&1
yes|%{_bindir}/updmap-sys --syncwithtrees --force > /dev/null 2>&1
%{_bindir}/texhash 2> /dev/null
%{_bindir}/fmtutil-sys --all > /dev/null 2>&1

You can see the yes commands in the middle there, broken pipe leads me to believe the updmap-sys command isn't available, even though the Requires(post): texlive-dist (which itself requires texlive) should be making it available.  I also see this same scriplet as a %posttrans in texlive itself.  Maybe either texlive is broken, the command is no longer provided, or there's some kind of circular deps so texlive isn't getting installed first.

@David: that is not the problem.
yes always get's broken pipe. But still updmap-sys has some errors which is, a hard coded file is not updated. I'm working on removing this hardcoded file and generate it, but currently we use the old file which points to a few missing (tex-)packages and therefore generates this error.

I hope to have time to do this during this week. Also, I want to split up this mega package into ~6 packages. So I hope to come up with a better solution by the end of the week.

Hi guys,
Same here, and also when installing other texlive packages. But the interesting part is in dmesg. In my case I can see that the script failed because luatex is crashing due to invalid code.

traps: xprobe_avx[24592] trap invalid opcode ip:401402 sp:7ffce68ecb98 error:0 in xprobe_avx[400000+3000]
[99011.471706] traps: luatex[30086] general protection ip:609b2c sp:7ffc47a889e0 error:0 in luatex[400000+59d000]

In view of the message, it seems that a compilation flag is wrong in texlive, vector instructions AVX seems to be switched on!! Obviously, if the machine is not supported them --> segfaults. At most we should have --sse2 enable, but higher vectorization options are not generic enough.

Cheers,
Chris.

CC: (none) => eatdirt

Chris: thanks for the report. I want to reorganize the package until release. But I'll have a look on the flag. I don't rember setting any flags other than standard compile flags on all other packages we have.

Ok, I'll try to have a look, maybe it some sorts of "autooptimization" at compile time!

I had a look, this is quite a scary package indeed!

The segfault of luatex is fixed, this was due to some non-alignment issues with latest gcc version > 7 compiler, thanks to Gentoo:

https://bugs.gentoo.org/621252

But this is not enough to fix the scriplets failing. As far as I can see, this is fmtutil which returns an error because it does not find some format.

I have added the option --no-strict to allow for more flexibility, fmtutil does not return an error code in case a format is missing. Locally, both of these fixes work fine.

I have resubmitted texlive to the buildserver, and updated both texlive and texlive-texmf on the svn. I'll submit texlive-texmf once texlive has landed.

Cheers,
Chris.

Yeah, my plan is to use the official installer and make subpackages out of it. But due to the size of this package, testing takes ages ;)
Currently I've to much other work to do. So I hope I have a solution for this, before we have our release.

Both packages landed, works fine. I am closing.

There is yet some output: "yes: standard output: Broken pipe", but that's normal, redirection of "yes" stopping has been made to the screen.

Resolution: (none) => FIXED
Status: NEW => RESOLVED