parallel --halt soon,fail=100% </dev/null should not give division by zero.

src/parallel

@@ -216,7 +216,7 @@ sub halt {
 	    if($Global::halt_pct) {
 		$Global::halt_exitstatus =
 		    ::ceil($Global::total_failed /
-			   $Global::total_started * 100);
+			   ($Global::total_started || 1) * 100);
 	    } elsif($Global::halt_count) {
 		$Global::halt_exitstatus =
 		    ::min(undef_as_zero($Global::total_failed),101);

testsuite/tests-to-run/parallel-local-0.3s.sh

@@ -886,6 +886,12 @@ par_sem_quote() {
     sem --fg --quote -v echo
 }
 
+par_halt_on_error_division_by_zero() {
+    echo '### --halt-on-error soon,fail=100% with no input should not give division by zero'
+    stdout parallel --halt-on-error soon,fail=100% echo </dev/null
+    echo $?
+}
+
 export -f $(compgen -A function | grep par_)
 compgen -A function | grep par_ | LC_ALL=C sort |
     parallel --timeout 20 -j6 --tag -k --joblog +/tmp/jl-`basename $0` '{} 2>&1'

testsuite/tests-to-run/parallel-local-sql.sh

@@ -148,5 +148,5 @@ export -f $(compgen -A function | egrep 'p_|par_')
 # Tested that -j0 in parallel is fastest (up to 15 jobs)
 # more than 3 jobs: sqlite locks
 compgen -A function | grep par_ | LC_ALL=C sort |
-  stdout parallel --timeout 30 -vj50% -k --tag --joblog /tmp/jl-`basename $0` p_wrapper \
+  stdout parallel --timeout 60 -vj50% -k --tag --joblog /tmp/jl-`basename $0` p_wrapper \
     :::: - ::: \$MYSQL \$PG \$SQLITE

testsuite/wanted-results/parallel-local-0.3s

@@ -1326,6 +1326,8 @@ par_file_ending_in_newline	### Hans found a bug giving unitialized variable
 par_file_ending_in_newline	gzip /tmp/parallel_f1
 par_file_ending_in_newline	gzip '/tmp/parallel_f2
 par_file_ending_in_newline	'
+par_halt_on_error_division_by_zero	### --halt-on-error soon,fail=100% with no input should not give division by zero
+par_halt_on_error_division_by_zero	0
 par_halt_one_job	# Halt soon if there is a single job
 par_halt_one_job	should run 0 1 = job 1 2
 par_halt_one_job	1