bug#23205: 25.0.92; sort-tests is flaky

[Eli Zaretskii]

> From: Philipp Stephani <phst@google.com>
> Date: Sun, 03 Apr 2016 10:30:10 +0000
> 
> In the emacs-25 branch, in the directory test/automated, run: 
> 
> declare -i failures=0 
> for ((i=0; i<1000; i++)); do 
> if ! output="$(make sort-tests 2>&1)"; then 
> ((failures++)) 
> echo "Iteration $i: $output" 
> fi 
> done 
> echo "Failures: $failures" 
> 
> There will be typically a small number of failures (< 10), caused by 
> e.g. 
> 
> Test sort-tests--lines condition: 
> (ert-test-failed 
> ((should 
> (string= 
> (substring ... 0 -1) 
> sorted-words)) 
> :form 
> (string= "Zi 
> zI" "zI 
> Zi") 
> :value nil)) 
> 
> This is presumably due to non-stable sorting. Either the sorting should 
> be made stable, or the test shouldn't assume stable sorting. 

Do you sometimes see such failures if you just run the test once?