On Fri, 2020-07-31 at 14:07 +0200, Florian Weimer wrote:
* Richard W. M. Jones:
> Here's another one:
>
> $ rpm -qf /usr/bin/guestfish /lib64/libreadline.so.8
> libguestfs-tools-c-1.43.1-2.fc33.x86_64
> readline-8.0-5.fc33.x86_64
> $ guestfish --version
> Segmentation fault (core dumped)
>
> (gdb) bt
> #0 0x0000000000000000 in ()
> #1 0x00007f3212b72dad in history_filename
> (filename=0x5592dd41bfa0 "/home/rjones/.guestfish") at
../histfile.c:152
> #2 0x00007f3212b75e2d in read_history_range
> (filename=<optimized out>, from=0, to=-1) at ../histfile.c:280
> #3 0x00005592dd33e646 in main ()
>
> It also caused a build failure of another package in Koji
> (search the logs for "ext2.img] Segmentation fault (core dumped)"):
>
>
https://koji.fedoraproject.org/koji/taskinfo?taskID=48245041
>
> I suspect the problem isn't in readline but is in the main package,
> mainly because I tried an older readline and that failed in the same
> way.
This looks like the ld bug producing incorrect absolute symbols:
$ eu-readelf --symbols=.dynsym usr/bin/guestfish | grep -w ABS
790: 0000000000000000 54 FUNC GLOBAL DEFAULT ABS xrealloc
791: 0000000000000000 35 FUNC GLOBAL DEFAULT ABS xmalloc
799: 0000000000000000 1294 FUNC GLOBAL DEFAULT ABS locale_charset
<
https://sourceware.org/bugzilla/show_bug.cgi?id=26314>
I believe a fix for that is on its way to rawhide.
Yup. binutils with the ld fix
is nearly done building (just waiting on s390x).
I'll have my eye on things today in case there's further fallout.
So Richard, wait for binutils-2.35.8 to hit the buildroots, respin guestfish and
we should be good.
jeff