Re: coreutils-8.31.90-cc4c.tar.xz on Solaris 11 OpenIndiana (1)

[Pádraig Brady]

On 01/03/2020 20:23, Bruno Haible wrote:
> On Solaris 11 OpenIndiana, I observe a test error:
>
>
> ERROR: tests/misc/env-signal-handler
> ====================================
>
> --- exp-err1    Sat Feb 29 00:36:58 2020
> +++ err1        Sat Feb 29 00:36:58 2020
> @@ -1,1 +1,0 @@
> -seq: write error:
> env-signal-handler.sh: set-up failure:
> ERROR tests/misc/env-signal-handler.sh (exit status: 99)
>
>
> On this platform, the file 'err1t' created through the command
>
>    (trap '' PIPE; $SHELL -c 'trap - PIPE; seq 999999 2>err1t | head -n1 > out1')
>
> sometimes contains the expected message "seq: write error: Broken pipe",
> sometimes it is empty. Most often it's as expected in the first 3 out of 20
> runs:
>
> $ for i in 0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19; do (trap '' PIPE; /bin/sh -c 
> 'trap - PIPE; seq 999999 2>err1t | head -n1 > out1') ; echo "Round $i:"; cat 
> err1t; done
> Round 0:
> seq: write error: Broken pipe
> Round 1:
> seq: write error: Broken pipe
> Round 2:
> seq: write error: Broken pipe
> Round 3:
> Round 4:
> Round 5:
> Round 6:
> Round 7:
> Round 8:
> Round 9:
> Round 10:
> Round 11:
> Round 12:
> Round 13:
> Round 14:
> Round 15:
> Round 16:
> Round 17:
> Round 18:
> Round 19:
>
> Sometimes it's the first 5 out of 20 runs; sometimes the first 3 and the 9th 
> run.
> So, it looks like a timing issue, as if the 'trap - PIPE' command takes a couple
> of milliseconds until it becomes effective.
>
> If I add a delay, I get mostly the expected output:
>
> $ for i in 0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19; do (trap '' PIPE; /bin/sh -c 
> 'trap - PIPE; sleep 1; seq 999999 2>err1t | head -n1 > out1') ; echo "Round 
> $i:"; cat err1t; done
> Round 0:
> seq: write error: Broken pipe
> Round 1:
> seq: write error: Broken pipe
> Round 2:
> seq: write error: Broken pipe
> Round 3:
> seq: write error: Broken pipe
> Round 4:
> seq: write error: Broken pipe
> Round 5:
> Round 6:
> seq: write error: Broken pipe
> Round 7:
> seq: write error: Broken pipe
> Round 8:
> seq: write error: Broken pipe
> Round 9:
> seq: write error: Broken pipe
> Round 10:
> seq: write error: Broken pipe
> Round 11:
> seq: write error: Broken pipe
> Round 12:
> seq: write error: Broken pipe
> Round 13:
> seq: write error: Broken pipe
> Round 14:
> seq: write error: Broken pipe
> Round 15:
> seq: write error: Broken pipe
> Round 16:
> seq: write error: Broken pipe
> Round 17:
> seq: write error: Broken pipe
> Round 18:
> seq: write error: Broken pipe
> Round 19:
> seq: write error: Broken pipe
>
> If I use bash instead of /bin/sh, the output is all right:
>
> $ for i in 0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19; do (trap '' PIPE; /usr/bin/bash 
> -c 'trap - PIPE; seq 999999 2>err1t | head -n1 > out1') ; echo "Round $i:"; cat 
> err1t; done
> Round 0:
> seq: write error: Broken pipe
> Round 1:
> seq: write error: Broken pipe
> Round 2:
> seq: write error: Broken pipe
> Round 3:
> seq: write error: Broken pipe
> Round 4:
> seq: write error: Broken pipe
> Round 5:
> seq: write error: Broken pipe
> Round 6:
> seq: write error: Broken pipe
> Round 7:
> seq: write error: Broken pipe
> Round 8:
> seq: write error: Broken pipe
> Round 9:
> seq: write error: Broken pipe
> Round 10:
> seq: write error: Broken pipe
> Round 11:
> seq: write error: Broken pipe
> Round 12:
> seq: write error: Broken pipe
> Round 13:
> seq: write error: Broken pipe
> Round 14:
> seq: write error: Broken pipe
> Round 15:
> seq: write error: Broken pipe
> Round 16:
> seq: write error: Broken pipe
> Round 17:
> seq: write error: Broken pipe
> Round 18:
> seq: write error: Broken pipe
> Round 19:
> seq: write error: Broken pipe
>
> But the value of SHELL in config.status is /bin/sh. Apparently /bin/sh is
> good enough for nearly everything, just not for this 'trap - PIPE' command.
>
> I'm not sure what is the best way forward: Use bash here instead of $SHELL
> whenever it is found in $PATH ?

Nice debugging.

It's a pity the failure is intermittent, as then one might
be able to add a check to gl_shell_test_script_ in gnulib/tests/init.sh,
to exclude the problematic shell. Though that might be a bit
draconian for only this particular issue.

I guess we should switch to bash for these cases.
bash is common enough to result in skips on few systems.
The attached should do that.

thanks!
Pádraig.

indiana-require-bash.patch
Description: Text Data