Re: MIPS build slave

[Simon Marchi <simon dot marchi at ericsson dot com>]

On 15-07-20 04:04 PM, Brendan Kirby wrote:
> I ran the steps below.  The build was successful.  I built with commit
> 7ac018954bd9 because the latest one was failing to build.  However, I'm
> still not seeing this file.  I've attached the output of running "make
> check RUNTESTFLAGS='gdb.mi/mi-watch.exp'" in the file
> gdb.mi-mi-watch.exp.out and I've attached the gdb/testsuite/gdb.log file
> in gdb.mi-mi-watch.exp.log.    Let me know if you want me to try
> anything else.
> 
> Brendan

Thanks for trying that.

Looking at the error message (any of them), the pattern is:

  Cannot exec $execfile -c exec $execfile

This looks awfully as if it should be

  $SHELL -c exec $execfile

but that somehow the exec-file name ended up in place of the shell.  That
command line seems to be constructed in fork-child.c:fork_inferior, but I
don't see how it could end up like that.

It's probably not related, but let's rule out obvious things.  I notice that
the SHELL of the user running the buildbot slave is be /sbin/nologin (from
an output of show env in gdb.log).  It seems to be a problem for some tests
that use the shell command at least (search for nologin in the log).  Setting
the shell of the user to a real shell will help for those.

When you ran the make check RUNTESTFLAGS="..." command that Sergio asked you
to run, was it as the same user as the buildbot slave's user?  I assume not,
since the shell is nologin.  Would it be possible to try it again, but this
time as that user?

Simon