[tex-live] Advice on debugging segfault in luatex (2017 source, gcc-7.1) ?
zarniwhoop at ntlworld.com
Wed May 24 05:10:45 CEST 2017
On Mon, May 22, 2017 at 10:29:16PM +0100, Ken Moffat wrote:
> On Mon, May 22, 2017 at 06:29:02PM +0900, Akira Kakuto wrote:
> > > At least ^D works there, but can somebody give me hints about how to
> > > run this test / where to run it from, please ?
> > What about
> > #
> > # change $YOUR appropriately
> > #
> > export TEXMFCNF=$YOUR/texlive-build/texk/kpathsea
> > export TEXINPUTS=$YOUR/texlive-build/texk/web2c/tests:$YOUR/texlive-build/texk/web2c/luatexdir/tests
> > export TEXFORMATS=.
> > ./luatex -ini luaimage
> Thanks. Didn't quite work out like that, but the segfault is when
> it closes - didn't notice that last night - so even after an error
> it still faults. Got the backtrace.
So, I got absolutely nowhere in working out what exactly in gcc-7.1
breaks this. I also tried to build 2016, that passes all its tests.
Looking at the trace, something in the lua area seems to be
miscompiled by gcc-7.1. The good news is that detuning CFLAGS to -O
(instead of -O2 which is the default and what I too prefer) makes
I looked at the documented changes for gcc-7.1: -fstore-merging and
-fcode-hoisting are both on for -O2 (and -Os), but turning both of
them off had no effect. No other new optimizations are mentioned in
the defaults. I then tried changing
source/libs/lua52/Makefile.in to add -O at the end of the CFLAGS but
that appeared to have no effect.
So for the moment I'm building the whole thing with -O and all my
own tests work. I already feel as if I'm too far out and not waving
but drowning, so I don't plan to try to work out where it goes wrong
(I have a hate / hate relationship with gdb). Sorry.
But if anyone else wants to build from source on gcc-7.1, this might
I live in a city. I know sparrows from starlings. After that
everything is a duck as far as I'm concerned. -- Monstrous Regiment
More information about the tex-live