Add a flag for requiring full coverage in coverage tests

Introduce the --require-full-coverage in analyze_outcomes.py so that
when analyze_outcomes.py --require-full-coverage is called, those
tests that are not executed and are not in the allowed list issue an
error instead of a warning.

Note that it is useful to run analyze_outcomes.py on incomplete test
results, so this error mode needs to remain optional in the long
term.

Signed-off-by: Tomás González <tomasagustin.gonzalezorlando@arm.com>
diff --git a/tests/scripts/analyze_outcomes.py b/tests/scripts/analyze_outcomes.py
index fde0715..24f4da7 100755
--- a/tests/scripts/analyze_outcomes.py
+++ b/tests/scripts/analyze_outcomes.py
@@ -73,15 +73,16 @@
         Results.log("Error: failed to run reference/driver components")
         sys.exit(ret_val)
 
-def analyze_coverage(results, outcomes, allow_list):
+def analyze_coverage(results, outcomes, allow_list, full_coverage):
     """Check that all available test cases are executed at least once."""
     available = check_test_cases.collect_available_test_cases()
     for key in available:
         hits = outcomes[key].hits() if key in outcomes else 0
         if hits == 0 and key not in allow_list:
-            # Make this a warning, not an error, as long as we haven't
-            # fixed this branch to have full coverage of test cases.
-            results.warning('Test case not executed: {}', key)
+            if full_coverage:
+                results.error('Test case not executed: {}', key)
+            else:
+                results.warning('Test case not executed: {}', key)
         elif hits != 0 and key in allow_list:
             # Test Case should be removed from the allow list.
             results.warning('Allow listed test case was executed: {}', key)
@@ -125,10 +126,11 @@
             result = False
     return result
 
-def analyze_outcomes(outcomes, allow_list):
+def analyze_outcomes(outcomes, args):
     """Run all analyses on the given outcome collection."""
     results = Results()
-    analyze_coverage(results, outcomes, allow_list)
+    analyze_coverage(results, outcomes, args['allow_list'],
+                     args['full_coverage'])
     return results
 
 def read_outcome_file(outcome_file):
@@ -156,7 +158,7 @@
     """Perform coverage analysis."""
     outcomes = read_outcome_file(outcome_file)
     Results.log("\n*** Analyze coverage ***\n")
-    results = analyze_outcomes(outcomes, args['allow_list'])
+    results = analyze_outcomes(outcomes, args)
     return results.error_count == 0
 
 def do_analyze_driver_vs_reference(outcome_file, args):
@@ -179,6 +181,7 @@
         'test_function': do_analyze_coverage,
         'args': {
             'allow_list': [],
+            'full_coverage': False,
         }
         },
     # There are 2 options to use analyze_driver_vs_reference_xxx locally:
@@ -430,6 +433,11 @@
                                  'comma/space-separated list of tasks. ')
         parser.add_argument('--list', action='store_true',
                             help='List all available tasks and exit.')
+        parser.add_argument('--require-full-coverage', action='store_true',
+                            dest='full_coverage', help="Require all available "
+                            "test cases to be executed and issue an error "
+                            "otherwise. This flag is ignored if 'task' is "
+                            "neither 'all' nor 'analyze_coverage'")
         options = parser.parse_args()
 
         if options.list:
@@ -449,6 +457,9 @@
                     Results.log('Error: invalid task: {}'.format(task))
                     sys.exit(1)
 
+        TASKS['analyze_coverage']['args']['full_coverage'] = \
+            options.full_coverage
+
         for task in TASKS:
             if task in tasks:
                 if not TASKS[task]['test_function'](options.outcomes, TASKS[task]['args']):