green_mode: Choose job count based on environment

Python 3.4 and 3.5 fail under our pytest configuration
when the number of jobs is the cpu count - 1.
On Python 3.4, an extra cpu needs to be reserved for the tests
to operate smoothly on CI, and on Python 3.5 and Windows,
multiprocessing needs to be disabled entirely.

Expose argument jobs throughout green_mode, so that it
can be also exposed to the user or test suite for more
controlled use and testing.

Related to #295

coala_quickstart/green_mode/green_mode.py

@@ -2,6 +2,7 @@
 import itertools
 import operator
 import os
+import sys
 from copy import deepcopy
 from pathlib import Path
 
@@ -41,6 +42,19 @@
 
 
 settings_key = 'green_mode_infinite_value_settings'
+_CI_PYTEST_ACTIVE = os.environ.get('CI') and os.environ.get('PYTEST')
+_PYTHON_VERSION_MINOR = sys.version_info[0:2]
+_RESERVE_CPUS = 1
+
+if _CI_PYTEST_ACTIVE:  # pragma: no cover
+    if os.name == 'nt':
+        # FIXME: Multiprocessing not working on windows.
+        _RESERVE_CPUS = sys.maxsize
+    elif _PYTHON_VERSION_MINOR == (3, 5):
+        # Python 3.5 fails even with _RESERVE_CPUS set to 2
+        _RESERVE_CPUS = sys.maxsize
+    elif _PYTHON_VERSION_MINOR == (3, 4):
+        _RESERVE_CPUS = 2
 
 
 def initialize_project_data(dir, ignore_globs):
@@ -272,13 +286,38 @@ def check_bear_results(ret_val, ignore_ranges):
     return True
 
 
+def _create_mp_pool(jobs: int = 0):
+    """
+    Create a multiprocessing pool.
+
+    :param jobs: Number of jobs to run concurrently.
+                 0 means auto-detect.  1 means no pool.
+    """
+    if not isinstance(jobs, int):
+        raise TypeError('jobs must be an int')
+    if jobs == 1:
+        return
+    if jobs < 0:
+        raise ValueError('jobs must be 0 or a positive integer')
+
+    import multiprocessing as mp
+    cpu_count = mp.cpu_count()
+    if cpu_count <= _RESERVE_CPUS:
+        return
+    if jobs == 0 or jobs > cpu_count - _RESERVE_CPUS:
+        jobs = cpu_count - _RESERVE_CPUS
+    pool = mp.Pool(processes=jobs)
+    return pool
+
+
 def local_bear_test(bear, file_dict, file_names, lang, kwargs,
-                    ignore_ranges):
+                    ignore_ranges,
+                    jobs: int = 0,
+                    ):
     lang_files = split_by_language(file_names)
     lang_files = {k.lower(): v for k, v in lang_files.items()}
 
-    import multiprocessing as mp
-    pool = mp.Pool(processes=mp.cpu_count()-1)
+    pool = _create_mp_pool(jobs)
 
     file_results = []
 
@@ -317,12 +356,11 @@ def local_bear_test(bear, file_dict, file_names, lang, kwargs,
             bear_obj = bear(section, None)
             ret_val = bear_obj.run(**dict(zip(kwargs, vals)))
             ret_val = [] if not ret_val else list(ret_val)
-            # FIXME: Multiprocessing not working on windows.
-            if os.name == 'nt':  # pragma posix: no cover
-                results.append(check_bear_results(ret_val, ignore_ranges))
-            else:  # pragma nt: no cover
+            if pool:  # pragma Python 3.5: no cover; pragma nt: no cover
                 results.append(pool.apply(check_bear_results,
                                           args=(ret_val, ignore_ranges)))
+            else:  # pragma Python 3.4,3.6,3.7: no cover
+                results.append(check_bear_results(ret_val, ignore_ranges))
 
         for index, result in enumerate(results):
             if result is True:
@@ -335,14 +373,15 @@ def local_bear_test(bear, file_dict, file_names, lang, kwargs,
     return {bear: file_results}
 
 
-def global_bear_test(bear, file_dict, kwargs, ignore_ranges):
-    import multiprocessing as mp
-    pool = mp.Pool(processes=mp.cpu_count()-1)
-
+def global_bear_test(bear, file_dict, kwargs, ignore_ranges,
+                     jobs: int = 0,
+                     ):
     results = []
     values = []
     file_results = []
 
+    pool = _create_mp_pool(jobs)
+
     for vals in itertools.product(*kwargs.values()):
         values.append(vals)
         section = Section('test-section-global-bear')
@@ -351,11 +390,11 @@ def global_bear_test(bear, file_dict, kwargs, ignore_ranges):
         bear_obj.file_dict = file_dict
         ret_val = bear_obj.run(**dict(zip(kwargs, vals)))
         ret_val = list(ret_val)
-        if os.name == 'nt':  # pragma posix: no cover
-            results.append(check_bear_results(ret_val, ignore_ranges))
-        else:  # pragma nt: no cover
+        if pool:  # pragma Python 3.5: no cover; pragma nt: no cover
             results.append(pool.apply(check_bear_results,
                                       args=(ret_val, ignore_ranges)))
+        else:  # pragma Python 3.4,3.6,3.7: no cover
+            results.append(check_bear_results(ret_val, ignore_ranges))
 
     for index, result in enumerate(results):
         if result is True:
@@ -367,7 +406,9 @@ def global_bear_test(bear, file_dict, kwargs, ignore_ranges):
 
 
 def run_test_on_each_bear(bear, file_dict, file_names, lang, kwargs,
-                          ignore_ranges, type_of_setting, printer=None):
+                          ignore_ranges, type_of_setting, printer=None,
+                          jobs: int = 0,
+                          ):
     if type_of_setting == 'non-op':
         printer.print('Finding suitable values to necessary '
                       'settings for ' + bear.__name__ +
@@ -389,7 +430,8 @@ def run_test_on_each_bear(bear, file_dict, file_names, lang, kwargs,
 
 def bear_test_fun(bears, bear_settings_obj, file_dict, ignore_ranges,
                   contents, file_names, op_args_limit, value_to_op_args_limit,
-                  printer=None):
+                  printer=None,
+                  jobs: int = 0):
     """
     Tests the bears with the generated file dict and list of files
     along with the values recieved for each and every type of setting
@@ -440,7 +482,9 @@ def bear_test_fun(bears, bear_settings_obj, file_dict, ignore_ranges,
             op_kwargs = get_kwargs(op_set, bear, contents)
             non_op_file_results = run_test_on_each_bear(
                 bear, file_dict, file_names, lang, non_op_kwargs,
-                ignore_ranges, 'non-op', printer)
+                ignore_ranges, 'non-op', printer,
+                jobs=jobs,
+                )
             if len(op_kwargs) < op_args_limit and not(
                     True in [len(value) > value_to_op_args_limit
                              for key, value in op_kwargs.items()]):
@@ -449,7 +493,9 @@ def bear_test_fun(bears, bear_settings_obj, file_dict, ignore_ranges,
                 unified_file_results = run_test_on_each_bear(
                     bear, file_dict, file_names, lang,
                     unified_kwargs, ignore_ranges, 'unified',
-                    printer)
+                    printer,
+                    jobs=jobs,
+                    )
             else:
                 unified_file_results = None
             final_non_op_results.append(non_op_file_results)

setup.cfg

@@ -28,6 +28,9 @@ doctest_optionflags =
   ELLIPSIS
   IGNORE_EXCEPTION_DETAIL
 
+env =
+  PYTEST=1
+
 reqsfilenamepatterns =
   requirements.txt
   test-requirements.txt

tests/green_mode/green_modeTest.py

@@ -1,5 +1,6 @@
 import operator
 import os
+import sys
 import unittest
 import yaml
 from copy import deepcopy
@@ -13,6 +14,7 @@
 from coala_quickstart.green_mode.Setting import (
     find_max_min_of_setting,
     )
+from coala_quickstart.green_mode import green_mode
 from coala_quickstart.green_mode.green_mode import (
     bear_test_fun,
     check_bear_results,
@@ -571,5 +573,58 @@ def test_write_coafile(self):
             self.assertIn(line, [i.strip('\\').replace('\\\\C', 'C')
                                  for i in contents.split('\n')])
 
-    def test_green_mode(self):
-        pass
+
+class MultiProcessingTest(unittest.TestCase):
+
+    def setUp(self):
+        self.orig_cpus = green_mode._RESERVE_CPUS
+
+    def tearDown(self):
+        green_mode._RESERVE_CPUS = self.orig_cpus
+
+    def test_no_pool(self):
+        with self.assertRaises(TypeError):
+            green_mode._create_mp_pool(None)
+        with self.assertRaises(ValueError):
+            self.assertIsNone(green_mode._create_mp_pool(-1))
+        self.assertIsNone(green_mode._create_mp_pool(1))
+
+    def test_reserve_cpus(self):
+        green_mode._RESERVE_CPUS = 1
+        with patch('multiprocessing.cpu_count', return_value=1):
+            self.assertIsNone(green_mode._create_mp_pool(0))
+        green_mode._RESERVE_CPUS = 2
+        with patch('multiprocessing.cpu_count', return_value=2):
+            self.assertIsNone(green_mode._create_mp_pool(0))
+
+        green_mode._RESERVE_CPUS = 1
+        with patch('multiprocessing.cpu_count', return_value=2):
+            self.assertIsNotNone(green_mode._create_mp_pool(0))
+
+        green_mode._RESERVE_CPUS = 1
+        with patch('multiprocessing.cpu_count', return_value=2):
+            self.assertIsNotNone(green_mode._create_mp_pool(2))
+
+        green_mode._RESERVE_CPUS = 1
+        with patch('multiprocessing.cpu_count', return_value=10):
+            self.assertIsNotNone(green_mode._create_mp_pool(2))
+
+    def test_ci_pool_min(self):
+        if not os.environ.get('CI'):
+            return
+
+        if sys.version_info[0:2] == (3, 4):
+            with patch('multiprocessing.cpu_count', return_value=2):
+                self.assertIsNone(green_mode._create_mp_pool(0))
+            with patch('multiprocessing.cpu_count', return_value=3):
+                self.assertIsNotNone(green_mode._create_mp_pool(0))
+        elif sys.version_info[0:2] == (3, 5):
+            with patch('multiprocessing.cpu_count', return_value=3):
+                self.assertIsNone(green_mode._create_mp_pool(0))
+            # Python 3.5 is forced to not be mp under CI & pytest
+            with patch('multiprocessing.cpu_count', return_value=100):
+                self.assertIsNone(green_mode._create_mp_pool(0))
+        else:
+            with patch('multiprocessing.cpu_count', return_value=2):
+                self.assertIsNotNone(green_mode._create_mp_pool(2))
+                self.assertIsNotNone(green_mode._create_mp_pool(0))