Re: SIGSEGV in GNU bash, version 5.1.4(1)-release (sparc-sun-solaris2.10)

[Vladimir Marek]

Hi,

As a Solaris bash maintainer this interests me very much. I haven't yet
tried to upgrade bash to 5.1.x as there is no immediate need, sorry :)

I'll try to test the new bash on latest Solaris (11.4). What did you do
to observe the issue? Just execute the binary?

The two most common problems I can think of are

a) direct manipulation of the environment variables array instead of
using getenv(3C) and putenv(3C). In linux afaik you can get away with
that, on solaris you can't. But quickly looking into bash sources
that does not seem to be the case.

b) unaligned memory access. Sparc can't handle that. There is not enough
info on compile options or the core itself to judge.

Of course it can be anything else too :)

Cheers
-- 
        Vlad



On Sat, Jan 02, 2021 at 12:24:44AM +0000, Dennis Clarke wrote:
> On 12/31/20 12:31 AM, Chet Ramey wrote:
> > On 12/29/20 7:28 PM, Dennis Clarke wrote:
> >>
> >> Firstly as a minor nit that seems to re-appear yearly there are still
> >> source files in the release tarballs that are not readable to a normal
> >> user :
> >>
> >> #
> >> #  find . -ls | grep '\-\-\-'
> >> 7348095   24 -rw------- 1 286 staff 38962 Jul  8 14:19
> >> ./lib/malloc/malloc.c
> >> 7347442    1 -rw------- 1 286 staff   226 May 29  2020
> >> ./tests/varenv20.sub
> >> #
> >> # chmod 644 ./lib/malloc/malloc.c  ./tests/varenv20.sub
> >>
> >> Trivial but annoyed as it throws the compile process out the window on
> >> the first initial pass.  Otherwise all goes well until we try to use the
> >> resultant bash binary.
> > 
> > I'll change those. A minor annoyance at best.
> 
> Hey man, thank you for getting back to me on New Years Eve !
> Now that is dedication right there :-)
> 
> > 
> >>
> >> I was very surprised to see :
> >>
> >> alpha $
> >> alpha $ dbx /opt/bw/bin/bash
> >> node_alpha-host_sun4u-zone_z_003-time_1609232628-pid_10009-uid_16411-gid_20002-fid_bash.core
> >>
> >> Reading bash
> >> core file header read successfully
> >> Reading ld.so.1
> >> Reading libreadline.so.8
> >> Reading libhistory.so.8
> > 
> > I can't reproduce this with the default configure and build options, on a
> > fresh install of Solaris 10 1/13 on Virtualbox. Maybe try building without
> > linking with an external readline library. There may be some
> > incompatibility between readline-8.0 and 8.1 that I'm not aware of.
> > 
> 
> Yeah, it is not exactly easy hardware to have laying around. I suspect
> something funny is at work here in my Fujitsu SPARC world. I was going
> to rebuild readline again and give it a whirl. Tomorrow maybe. Or
> Monday.
> 
> If you want access to a build box just drop me a line with a public key.
> I still have these great hulking Oracle Fujitsu boxes running and I
> suspect I will for years. Some people really don't want to migrate code
> and that goes double when it is piles of Fortran IV stuff ported out of
> older IBM Mainframes.
> 
> 
> -- 
> Dennis Clarke
> RISC-V/SPARC/PPC/ARM/CISC
> UNIX and Linux spoken
> GreyBeard and suspenders optional
>