Re: [PATCH] scripts/runtime: Adjust the premature_failure check

["Radim Krčmář" <rkrcmar@redhat.com>]

2016-05-31 20:13+0200, Lukáš Doktor:
> Dne 31.5.2016 v 17:56 Radim Krčmář napsal(a):
> > 2016-05-31 17:37+0200, Radim Krčmář:
> > > 2016-05-31 12:54+0200, Lukáš Doktor:
> > > > The premature_failure check fails on qemu-system-ppc64 as it uses
> > > > different message. This patch modifies the condition so it works
> > > > (hopefully) for all archs.
> > > > ---
> > > > diff --git a/scripts/runtime.bash b/scripts/runtime.bash
> > > > -    echo "$last_line" | grep -qi "could not load kernel" &&
> > > > +    echo "$last_line" | grep "_NO_FILE_4Uhere_" | grep -qi "no such file" &&
> > > 
> > > PPC uses "Could not open '_NO_FILE_4Uhere_': No such file [...]".
> > > The problem is that "No such file [...]" might be localized, so what
> > > about
> > >   grep -qi "could not \(load kernel\|open\) '_NO_FILE_4Uhere_'"
> > > ?
> > 
> > My tested local repo has a different line there:
> > (no idea how the posted version came around ...)
> > 
> >   echo "$last_line" | grep -qi "_NO_FILE_4Uhere_" &&
> Oups, I sent a wrong version. I didn't use the `last_line`, but full output to check for the result to be more lenient:
> 
> ```
> diff --git a/scripts/runtime.bash b/scripts/runtime.bash
> index 886814b..96c82c6 100644
> --- a/scripts/runtime.bash
> +++ b/scripts/runtime.bash
> @@ -15,9 +15,8 @@ extract_summary()
>  premature_failure()
>  {
>      local log="$(eval $(get_cmdline _NO_FILE_4Uhere_) 2>&1)"
> -    local last_line=$(tail -1 <<< "$log")
> -    echo "$last_line" | grep "_NO_FILE_4Uhere_" | grep -qi "no such file" &&
> +    echo "$log" | grep "_NO_FILE_4Uhere_" | grep -qi "no such file" &&
>          return 1
>      RUNTIME_log_stderr <<< "$log"
> ```
> 
> Which works for me on ppc64 and x86_64. Anyway it fails on aarch64 as indeed the message is localized. I'm wondering how strict this check needs to be. I'd be fine with checking the exit_code + `grep -q "_NO_FILE_4Uhere_"` anywhere in the output.

Exit code is going to be 1 pretty much every time, so I wouldn't bother
with it.

I used $last_line, because grep only checked for _NO_FILE_4Uhere_ and
qemu could have printed it for some other reasons too.
Parsing the whole log sounds good.

> Alternatively how about stronger version could check the last line in this way `echo "$last_line" | grep "_NO_FILE_4Uhere_" | grep -q -e "could not load kernel" -e "error loading" which works on my qemu-system-{x86_64,aarch64,ppc64}.

Ah, true, PPC says something different ... (too many mistakes these
days, I'll be taking another vacation soon)
QEMU's exit messages (on rawhide fedora):

  # qemu-system-aarch64 -kernel _NO_FILE_4Uhere_ -M virt
  qemu: could not load kernel '_NO_FILE_4Uhere_'

  # qemu-system-ppc64   -kernel _NO_FILE_4Uhere_
  qemu-system-ppc64: error loading _NO_FILE_4Uhere_: Failed to load ELF

  # qemu-system-x86_64  -kernel _NO_FILE_4Uhere_
  qemu: could not load kernel '_NO_FILE_4Uhere_': No such file or directory

> Any preferences?

Checking that "_NO_FILE_4Uhere_" occurs on line with/after "could not
load kernel" or "error loading" seems best to me.