Lit: Rework r249161; Move RLIMIT_NPROC to main.py.
[oota-llvm.git] / utils / lit / lit / main.py
index 039868da78603a2d7efdfa997ee4467b00b5254b..45ec86d1c3ffa7dbe647475959c18b761438b59f 100755 (executable)
@@ -6,47 +6,24 @@ lit - LLVM Integrated Tester.
 See lit.pod for more information.
 """
 
-import math, os, platform, random, re, sys, time, threading, traceback
+from __future__ import absolute_import
+import math, os, platform, random, re, sys, time
 
-import ProgressBar
-import TestRunner
-import Util
+import lit.ProgressBar
+import lit.LitConfig
+import lit.Test
+import lit.run
+import lit.util
+import lit.discovery
 
-from TestingConfig import TestingConfig
-import LitConfig
-import Test
-
-# Configuration files to look for when discovering test suites. These can be
-# overridden with --config-prefix.
-#
-# FIXME: Rename to 'config.lit', 'site.lit', and 'local.lit' ?
-gConfigName = 'lit.cfg'
-gSiteConfigName = 'lit.site.cfg'
-
-kLocalConfigName = 'lit.local.cfg'
-
-class TestingProgressDisplay:
+class TestingProgressDisplay(object):
     def __init__(self, opts, numTests, progressBar=None):
         self.opts = opts
         self.numTests = numTests
         self.current = None
-        self.lock = threading.Lock()
         self.progressBar = progressBar
         self.completed = 0
 
-    def update(self, test):
-        # Avoid locking overhead in quiet mode
-        if self.opts.quiet and not test.result.isFailure:
-            self.completed += 1
-            return
-
-        # Output lock.
-        self.lock.acquire()
-        try:
-            self.handleUpdate(test)
-        finally:
-            self.lock.release()
-
     def finish(self):
         if self.progressBar:
             self.progressBar.clear()
@@ -55,328 +32,121 @@ class TestingProgressDisplay:
         elif self.opts.succinct:
             sys.stdout.write('\n')
 
-    def handleUpdate(self, test):
+    def update(self, test):
         self.completed += 1
+
+        if self.opts.incremental:
+            update_incremental_cache(test)
+
         if self.progressBar:
             self.progressBar.update(float(self.completed)/self.numTests,
                                     test.getFullName())
 
-        if self.opts.succinct and not test.result.isFailure:
+        shouldShow = test.result.code.isFailure or \
+            (not self.opts.quiet and not self.opts.succinct)
+        if not shouldShow:
             return
 
         if self.progressBar:
             self.progressBar.clear()
 
-        print '%s: %s (%d of %d)' % (test.result.name, test.getFullName(),
-                                     self.completed, self.numTests)
-
-        if test.result.isFailure and self.opts.showOutput:
-            print "%s TEST '%s' FAILED %s" % ('*'*20, test.getFullName(),
-                                              '*'*20)
-            print test.output
-            print "*" * 20
-
+        # Show the test result line.
+        test_name = test.getFullName()
+        print('%s: %s (%d of %d)' % (test.result.code.name, test_name,
+                                     self.completed, self.numTests))
+
+        # Show the test failure output, if requested.
+        if test.result.code.isFailure and self.opts.showOutput:
+            print("%s TEST '%s' FAILED %s" % ('*'*20, test.getFullName(),
+                                              '*'*20))
+            print(test.result.output)
+            print("*" * 20)
+
+        # Report test metrics, if present.
+        if test.result.metrics:
+            print("%s TEST '%s' RESULTS %s" % ('*'*10, test.getFullName(),
+                                               '*'*10))
+            items = sorted(test.result.metrics.items())
+            for metric_name, value in items:
+                print('%s: %s ' % (metric_name, value.format()))
+            print("*" * 10)
+
+        # Ensure the output is flushed.
         sys.stdout.flush()
 
-class TestProvider:
-    def __init__(self, tests, maxTime):
-        self.maxTime = maxTime
-        self.iter = iter(tests)
-        self.lock = threading.Lock()
-        self.startTime = time.time()
-
-    def get(self):
-        # Check if we have run out of time.
-        if self.maxTime is not None:
-            if time.time() - self.startTime > self.maxTime:
-                return None
-
-        # Otherwise take the next test.
-        self.lock.acquire()
-        try:
-            item = self.iter.next()
-        except StopIteration:
-            item = None
-        self.lock.release()
-        return item
-
-class Tester(threading.Thread):
-    def __init__(self, litConfig, provider, display):
-        threading.Thread.__init__(self)
-        self.litConfig = litConfig
-        self.provider = provider
-        self.display = display
-
-    def run(self):
-        while 1:
-            item = self.provider.get()
-            if item is None:
-                break
-            self.runTest(item)
-
-    def runTest(self, test):
-        result = None
-        startTime = time.time()
-        try:
-            result, output = test.config.test_format.execute(test,
-                                                             self.litConfig)
-        except KeyboardInterrupt:
-            # This is a sad hack. Unfortunately subprocess goes
-            # bonkers with ctrl-c and we start forking merrily.
-            print '\nCtrl-C detected, goodbye.'
-            os.kill(0,9)
-        except:
-            if self.litConfig.debug:
-                raise
-            result = Test.UNRESOLVED
-            output = 'Exception during script execution:\n'
-            output += traceback.format_exc()
-            output += '\n'
-        elapsed = time.time() - startTime
-
-        test.setResult(result, output, elapsed)
-        self.display.update(test)
-
-def dirContainsTestSuite(path):
-    cfgpath = os.path.join(path, gSiteConfigName)
-    if os.path.exists(cfgpath):
-        return cfgpath
-    cfgpath = os.path.join(path, gConfigName)
-    if os.path.exists(cfgpath):
-        return cfgpath
-
-def getTestSuite(item, litConfig, cache):
-    """getTestSuite(item, litConfig, cache) -> (suite, relative_path)
-
-    Find the test suite containing @arg item.
-
-    @retval (None, ...) - Indicates no test suite contains @arg item.
-    @retval (suite, relative_path) - The suite that @arg item is in, and its
-    relative path inside that suite.
-    """
-    def search1(path):
-        # Check for a site config or a lit config.
-        cfgpath = dirContainsTestSuite(path)
-
-        # If we didn't find a config file, keep looking.
-        if not cfgpath:
-            parent,base = os.path.split(path)
-            if parent == path:
-                return (None, ())
-
-            ts, relative = search(parent)
-            return (ts, relative + (base,))
-
-        # We found a config file, load it.
-        if litConfig.debug:
-            litConfig.note('loading suite config %r' % cfgpath)
-
-        cfg = TestingConfig.frompath(cfgpath, None, litConfig, mustExist = True)
-        source_root = os.path.realpath(cfg.test_source_root or path)
-        exec_root = os.path.realpath(cfg.test_exec_root or path)
-        return Test.TestSuite(cfg.name, source_root, exec_root, cfg), ()
-
-    def search(path):
-        # Check for an already instantiated test suite.
-        res = cache.get(path)
-        if res is None:
-            cache[path] = res = search1(path)
-        return res
-
-    # Canonicalize the path.
-    item = os.path.realpath(item)
-
-    # Skip files and virtual components.
-    components = []
-    while not os.path.isdir(item):
-        parent,base = os.path.split(item)
-        if parent == item:
-            return (None, ())
-        components.append(base)
-        item = parent
-    components.reverse()
-
-    ts, relative = search(item)
-    return ts, tuple(relative + tuple(components))
-
-def getLocalConfig(ts, path_in_suite, litConfig, cache):
-    def search1(path_in_suite):
-        # Get the parent config.
-        if not path_in_suite:
-            parent = ts.config
-        else:
-            parent = search(path_in_suite[:-1])
-
-        # Load the local configuration.
-        source_path = ts.getSourcePath(path_in_suite)
-        cfgpath = os.path.join(source_path, kLocalConfigName)
-        if litConfig.debug:
-            litConfig.note('loading local config %r' % cfgpath)
-        return TestingConfig.frompath(cfgpath, parent, litConfig,
-                                    mustExist = False,
-                                    config = parent.clone(cfgpath))
-
-    def search(path_in_suite):
-        key = (ts, path_in_suite)
-        res = cache.get(key)
-        if res is None:
-            cache[key] = res = search1(path_in_suite)
-        return res
-
-    return search(path_in_suite)
-
-def getTests(path, litConfig, testSuiteCache, localConfigCache):
-    # Find the test suite for this input and its relative path.
-    ts,path_in_suite = getTestSuite(path, litConfig, testSuiteCache)
-    if ts is None:
-        litConfig.warning('unable to find test suite for %r' % path)
-        return (),()
-
-    if litConfig.debug:
-        litConfig.note('resolved input %r to %r::%r' % (path, ts.name,
-                                                        path_in_suite))
-
-    return ts, getTestsInSuite(ts, path_in_suite, litConfig,
-                               testSuiteCache, localConfigCache)
-
-def getTestsInSuite(ts, path_in_suite, litConfig,
-                    testSuiteCache, localConfigCache):
-    # Check that the source path exists (errors here are reported by the
-    # caller).
-    source_path = ts.getSourcePath(path_in_suite)
-    if not os.path.exists(source_path):
-        return
-
-    # Check if the user named a test directly.
-    if not os.path.isdir(source_path):
-        lc = getLocalConfig(ts, path_in_suite[:-1], litConfig, localConfigCache)
-        yield Test.Test(ts, path_in_suite, lc)
-        return
-
-    # Otherwise we have a directory to search for tests, start by getting the
-    # local configuration.
-    lc = getLocalConfig(ts, path_in_suite, litConfig, localConfigCache)
-
-    # Search for tests.
-    if lc.test_format is not None:
-        for res in lc.test_format.getTestsInDirectory(ts, path_in_suite,
-                                                      litConfig, lc):
-            yield res
-
-    # Search subdirectories.
-    for filename in os.listdir(source_path):
-        # FIXME: This doesn't belong here?
-        if filename in ('Output', '.svn') or filename in lc.excludes:
-            continue
-
-        # Ignore non-directories.
-        file_sourcepath = os.path.join(source_path, filename)
-        if not os.path.isdir(file_sourcepath):
-            continue
-
-        # Check for nested test suites, first in the execpath in case there is a
-        # site configuration and then in the source path.
-        file_execpath = ts.getExecPath(path_in_suite + (filename,))
-        if dirContainsTestSuite(file_execpath):
-            sub_ts, subiter = getTests(file_execpath, litConfig,
-                                       testSuiteCache, localConfigCache)
-        elif dirContainsTestSuite(file_sourcepath):
-            sub_ts, subiter = getTests(file_sourcepath, litConfig,
-                                       testSuiteCache, localConfigCache)
-        else:
-            # Otherwise, continue loading from inside this test suite.
-            subiter = getTestsInSuite(ts, path_in_suite + (filename,),
-                                      litConfig, testSuiteCache,
-                                      localConfigCache)
-            sub_ts = None
-
-        N = 0
-        for res in subiter:
-            N += 1
-            yield res
-        if sub_ts and not N:
-            litConfig.warning('test suite %r contained no tests' % sub_ts.name)
-
-def runTests(numThreads, litConfig, provider, display):
-    # If only using one testing thread, don't use threads at all; this lets us
-    # profile, among other things.
-    if numThreads == 1:
-        t = Tester(litConfig, provider, display)
-        t.run()
-        return
-
-    # Otherwise spin up the testing threads and wait for them to finish.
-    testers = [Tester(litConfig, provider, display)
-               for i in range(numThreads)]
-    for t in testers:
-        t.start()
+def write_test_results(run, lit_config, testing_time, output_path):
     try:
-        for t in testers:
-            t.join()
-    except KeyboardInterrupt:
-        sys.exit(2)
-
-def load_test_suite(inputs):
-    import unittest
-
-    # Create the global config object.
-    litConfig = LitConfig.LitConfig(progname = 'lit',
-                                    path = [],
-                                    quiet = False,
-                                    useValgrind = False,
-                                    valgrindLeakCheck = False,
-                                    valgrindArgs = [],
-                                    useTclAsSh = False,
-                                    noExecute = False,
-                                    ignoreStdErr = False,
-                                    debug = False,
-                                    isWindows = (platform.system()=='Windows'),
-                                    params = {})
-
-    # Load the tests from the inputs.
-    tests = []
-    testSuiteCache = {}
-    localConfigCache = {}
-    for input in inputs:
-        prev = len(tests)
-        tests.extend(getTests(input, litConfig,
-                              testSuiteCache, localConfigCache)[1])
-        if prev == len(tests):
-            litConfig.warning('input %r contained no tests' % input)
-
-    # If there were any errors during test discovery, exit now.
-    if litConfig.numErrors:
-        print >>sys.stderr, '%d errors, exiting.' % litConfig.numErrors
-        sys.exit(2)
+        import json
+    except ImportError:
+        lit_config.fatal('test output unsupported with Python 2.5')
+
+    # Construct the data we will write.
+    data = {}
+    # Encode the current lit version as a schema version.
+    data['__version__'] = lit.__versioninfo__
+    data['elapsed'] = testing_time
+    # FIXME: Record some information on the lit configuration used?
+    # FIXME: Record information from the individual test suites?
+
+    # Encode the tests.
+    data['tests'] = tests_data = []
+    for test in run.tests:
+        test_data = {
+            'name' : test.getFullName(),
+            'code' : test.result.code.name,
+            'output' : test.result.output,
+            'elapsed' : test.result.elapsed }
+
+        # Add test metrics, if present.
+        if test.result.metrics:
+            test_data['metrics'] = metrics_data = {}
+            for key, value in test.result.metrics.items():
+                metrics_data[key] = value.todata()
+
+        tests_data.append(test_data)
+
+    # Write the output.
+    f = open(output_path, 'w')
+    try:
+        json.dump(data, f, indent=2, sort_keys=True)
+        f.write('\n')
+    finally:
+        f.close()
 
-    # Return a unittest test suite which just runs the tests in order.
-    def get_test_fn(test):
-        return unittest.FunctionTestCase(
-            lambda: test.config.test_format.execute(
-                test, litConfig),
-            description = test.getFullName())
+def update_incremental_cache(test):
+    if not test.result.code.isFailure:
+        return
+    fname = test.getFilePath()
+    os.utime(fname, None)
 
-    from LitTestCase import LitTestCase
-    return unittest.TestSuite([LitTestCase(test, litConfig) for test in tests])
+def sort_by_incremental_cache(run):
+    def sortIndex(test):
+        fname = test.getFilePath()
+        try:
+            return -os.path.getmtime(fname)
+        except:
+            return 0
+    run.tests.sort(key = lambda t: sortIndex(t))
 
-def main(builtinParameters = {}):    # Bump the GIL check interval, its more important to get any one thread to a
-    # blocking operation (hopefully exec) than to try and unblock other threads.
-    #
-    # FIXME: This is a hack.
-    import sys
-    sys.setcheckinterval(1000)
+def main(builtinParameters = {}):
+    # Use processes by default on Unix platforms.
+    isWindows = platform.system() == 'Windows'
+    useProcessesIsDefault = not isWindows
 
     global options
     from optparse import OptionParser, OptionGroup
     parser = OptionParser("usage: %prog [options] {file-or-path}")
 
+    parser.add_option("", "--version", dest="show_version",
+                      help="Show version and exit",
+                      action="store_true", default=False)
     parser.add_option("-j", "--threads", dest="numThreads", metavar="N",
                       help="Number of testing threads",
                       type=int, action="store", default=None)
     parser.add_option("", "--config-prefix", dest="configPrefix",
                       metavar="NAME", help="Prefix for 'lit' config files",
                       action="store", default=None)
-    parser.add_option("", "--param", dest="userParameters",
+    parser.add_option("-D", "--param", dest="userParameters",
                       metavar="NAME=VAL",
                       help="Add 'NAME' = 'VAL' to the user defined parameters",
                       type=str, action="append", default=[])
@@ -393,9 +163,18 @@ def main(builtinParameters = {}):    # Bump the GIL check interval, its more imp
     group.add_option("-v", "--verbose", dest="showOutput",
                      help="Show all test output",
                      action="store_true", default=False)
+    group.add_option("-o", "--output", dest="output_path",
+                     help="Write test results to the provided path",
+                     action="store", type=str, metavar="PATH")
     group.add_option("", "--no-progress-bar", dest="useProgressBar",
                      help="Do not use curses based progress bar",
                      action="store_false", default=True)
+    group.add_option("", "--show-unsupported", dest="show_unsupported",
+                     help="Show unsupported tests",
+                     action="store_true", default=False)
+    group.add_option("", "--show-xfail", dest="show_xfail",
+                     help="Show tests that were expected to fail",
+                     action="store_true", default=False)
     parser.add_option_group(group)
 
     group = OptionGroup(parser, "Test Execution")
@@ -417,6 +196,9 @@ def main(builtinParameters = {}):    # Bump the GIL check interval, its more imp
     group.add_option("", "--no-execute", dest="noExecute",
                      help="Don't execute any tests (assume PASS)",
                      action="store_true", default=False)
+    group.add_option("", "--xunit-xml-output", dest="xunit_output_file",
+                      help=("Write XUnit-compatible XML test reports to the"
+                            " specified file"), default=None)
     parser.add_option_group(group)
 
     group = OptionGroup(parser, "Test Selection")
@@ -429,7 +211,11 @@ def main(builtinParameters = {}):    # Bump the GIL check interval, its more imp
     group.add_option("", "--shuffle", dest="shuffle",
                      help="Run tests in random order",
                      action="store_true", default=False)
-    group.add_option("", "--filter", dest="filter", metavar="EXPRESSION",
+    group.add_option("-i", "--incremental", dest="incremental",
+                     help="Run modified and failing tests first (updates "
+                     "mtimes)",
+                     action="store_true", default=False)
+    group.add_option("", "--filter", dest="filter", metavar="REGEX",
                      help=("Only run tests with paths matching the given "
                            "regular expression"),
                      action="store", default=None)
@@ -442,32 +228,33 @@ def main(builtinParameters = {}):    # Bump the GIL check interval, its more imp
     group.add_option("", "--show-suites", dest="showSuites",
                       help="Show discovered test suites",
                       action="store_true", default=False)
-    group.add_option("", "--no-tcl-as-sh", dest="useTclAsSh",
-                      help="Don't run Tcl scripts using 'sh'",
-                      action="store_false", default=True)
-    group.add_option("", "--repeat", dest="repeatTests", metavar="N",
-                      help="Repeat tests N times (for timing)",
-                      action="store", default=None, type=int)
+    group.add_option("", "--show-tests", dest="showTests",
+                      help="Show all discovered tests",
+                      action="store_true", default=False)
+    group.add_option("", "--use-processes", dest="useProcesses",
+                      help="Run tests in parallel with processes (not threads)",
+                      action="store_true", default=useProcessesIsDefault)
+    group.add_option("", "--use-threads", dest="useProcesses",
+                      help="Run tests in parallel with threads (not processes)",
+                      action="store_false", default=useProcessesIsDefault)
     parser.add_option_group(group)
 
     (opts, args) = parser.parse_args()
 
+    if opts.show_version:
+        print("lit %s" % (lit.__version__,))
+        return
+
     if not args:
         parser.error('No inputs specified')
 
-    if opts.configPrefix is not None:
-        global gConfigName, gSiteConfigName, kLocalConfigName
-        gConfigName = '%s.cfg' % opts.configPrefix
-        gSiteConfigName = '%s.site.cfg' % opts.configPrefix
-        kLocalConfigName = '%s.local.cfg' % opts.configPrefix
-
     if opts.numThreads is None:
 # Python <2.5 has a race condition causing lit to always fail with numThreads>1
 # http://bugs.python.org/issue1731717
 # I haven't seen this bug occur with 2.5.2 and later, so only enable multiple
 # threads by default there.
        if sys.hexversion >= 0x2050200:
-               opts.numThreads = Util.detectCPUs()
+               opts.numThreads = lit.util.detectCPUs()
        else:
                opts.numThreads = 1
 
@@ -483,68 +270,54 @@ def main(builtinParameters = {}):    # Bump the GIL check interval, its more imp
         userParams[name] = val
 
     # Create the global config object.
-    litConfig = LitConfig.LitConfig(progname = os.path.basename(sys.argv[0]),
-                                    path = opts.path,
-                                    quiet = opts.quiet,
-                                    useValgrind = opts.useValgrind,
-                                    valgrindLeakCheck = opts.valgrindLeakCheck,
-                                    valgrindArgs = opts.valgrindArgs,
-                                    useTclAsSh = opts.useTclAsSh,
-                                    noExecute = opts.noExecute,
-                                    ignoreStdErr = False,
-                                    debug = opts.debug,
-                                    isWindows = (platform.system()=='Windows'),
-                                    params = userParams)
-
-    # Expand '@...' form in inputs.
-    actual_inputs = []
-    for input in inputs:
-        if os.path.exists(input) or not input.startswith('@'):
-            actual_inputs.append(input)
-        else:
-            f = open(input[1:])
-            try:
-                for ln in f:
-                    ln = ln.strip()
-                    if ln:
-                        actual_inputs.append(ln)
-            finally:
-                f.close()
-                    
-            
-    # Load the tests from the inputs.
-    tests = []
-    testSuiteCache = {}
-    localConfigCache = {}
-    for input in actual_inputs:
-        prev = len(tests)
-        tests.extend(getTests(input, litConfig,
-                              testSuiteCache, localConfigCache)[1])
-        if prev == len(tests):
-            litConfig.warning('input %r contained no tests' % input)
-
-    # If there were any errors during test discovery, exit now.
-    if litConfig.numErrors:
-        print >>sys.stderr, '%d errors, exiting.' % litConfig.numErrors
-        sys.exit(2)
-
-    if opts.showSuites:
-        suitesAndTests = dict([(ts,[])
-                               for ts,_ in testSuiteCache.values()
-                               if ts])
-        for t in tests:
-            suitesAndTests[t.suite].append(t)
-
-        print '-- Test Suites --'
-        suitesAndTests = suitesAndTests.items()
-        suitesAndTests.sort(key = lambda (ts,_): ts.name)
-        for ts,ts_tests in suitesAndTests:
-            print '  %s - %d tests' %(ts.name, len(ts_tests))
-            print '    Source Root: %s' % ts.source_root
-            print '    Exec Root  : %s' % ts.exec_root
+    litConfig = lit.LitConfig.LitConfig(
+        progname = os.path.basename(sys.argv[0]),
+        path = opts.path,
+        quiet = opts.quiet,
+        useValgrind = opts.useValgrind,
+        valgrindLeakCheck = opts.valgrindLeakCheck,
+        valgrindArgs = opts.valgrindArgs,
+        noExecute = opts.noExecute,
+        debug = opts.debug,
+        isWindows = isWindows,
+        params = userParams,
+        config_prefix = opts.configPrefix)
+
+    # Perform test discovery.
+    run = lit.run.Run(litConfig,
+                      lit.discovery.find_tests_for_inputs(litConfig, inputs))
+
+    if opts.showSuites or opts.showTests:
+        # Aggregate the tests by suite.
+        suitesAndTests = {}
+        for result_test in run.tests:
+            if result_test.suite not in suitesAndTests:
+                suitesAndTests[result_test.suite] = []
+            suitesAndTests[result_test.suite].append(result_test)
+        suitesAndTests = list(suitesAndTests.items())
+        suitesAndTests.sort(key = lambda item: item[0].name)
+
+        # Show the suites, if requested.
+        if opts.showSuites:
+            print('-- Test Suites --')
+            for ts,ts_tests in suitesAndTests:
+                print('  %s - %d tests' %(ts.name, len(ts_tests)))
+                print('    Source Root: %s' % ts.source_root)
+                print('    Exec Root  : %s' % ts.exec_root)
+
+        # Show the tests, if requested.
+        if opts.showTests:
+            print('-- Available Tests --')
+            for ts,ts_tests in suitesAndTests:
+                ts_tests.sort(key = lambda test: test.path_in_suite)
+                for test in ts_tests:
+                    print('  %s' % (test.getFullName(),))
+
+        # Exit.
+        sys.exit(0)
 
     # Select and order the tests.
-    numTotalTests = len(tests)
+    numTotalTests = len(run.tests)
 
     # First, select based on the filter expression if given.
     if opts.filter:
@@ -553,113 +326,167 @@ def main(builtinParameters = {}):    # Bump the GIL check interval, its more imp
         except:
             parser.error("invalid regular expression for --filter: %r" % (
                     opts.filter))
-        tests = [t for t in tests
-                 if rex.search(t.getFullName())]
+        run.tests = [result_test for result_test in run.tests
+                     if rex.search(result_test.getFullName())]
 
     # Then select the order.
     if opts.shuffle:
-        random.shuffle(tests)
+        random.shuffle(run.tests)
+    elif opts.incremental:
+        sort_by_incremental_cache(run)
     else:
-        tests.sort(key = lambda t: t.getFullName())
+        run.tests.sort(key = lambda result_test: result_test.getFullName())
 
     # Finally limit the number of tests, if desired.
     if opts.maxTests is not None:
-        tests = tests[:opts.maxTests]
+        run.tests = run.tests[:opts.maxTests]
+
+    # Don't create more threads than tests.
+    opts.numThreads = min(len(run.tests), opts.numThreads)
+
+    # Because some tests use threads internally, and at least on Linux each
+    # of these threads counts toward the current process limit, try to
+    # raise the (soft) process limit so that tests don't fail due to
+    # resource exhaustion.
+    try:
+        cpus = lit.util.detectCPUs()
+        desired_limit = opts.numThreads * cpus * 2 # the 2 is a safety factor
+
+        # Import the resource module here inside this try block because it
+        # will likely fail on Windows.
+        import resource
+
+        max_procs_soft, max_procs_hard = resource.getrlimit(resource.RLIMIT_NPROC)
+        desired_limit = min(desired_limit, max_procs_hard)
+
+        if max_procs_soft < desired_limit:
+            resource.setrlimit(resource.RLIMIT_NPROC, (desired_limit, max_procs_hard))
+            litConfig.note('raised the process limit from %d to %d' % \
+                               (max_procs_soft, desired_limit))
+    except:
+        pass
 
     extra = ''
-    if len(tests) != numTotalTests:
+    if len(run.tests) != numTotalTests:
         extra = ' of %d' % numTotalTests
-    header = '-- Testing: %d%s tests, %d threads --'%(len(tests),extra,
+    header = '-- Testing: %d%s tests, %d threads --'%(len(run.tests), extra,
                                                       opts.numThreads)
 
-    if opts.repeatTests:
-        tests = [t.copyWithIndex(i)
-                 for t in tests
-                 for i in range(opts.repeatTests)]
-
     progressBar = None
     if not opts.quiet:
         if opts.succinct and opts.useProgressBar:
             try:
-                tc = ProgressBar.TerminalController()
-                progressBar = ProgressBar.ProgressBar(tc, header)
+                tc = lit.ProgressBar.TerminalController()
+                progressBar = lit.ProgressBar.ProgressBar(tc, header)
             except ValueError:
-                print header
-                progressBar = ProgressBar.SimpleProgressBar('Testing: ')
+                print(header)
+                progressBar = lit.ProgressBar.SimpleProgressBar('Testing: ')
         else:
-            print header
-
-    # Don't create more threads than tests.
-    opts.numThreads = min(len(tests), opts.numThreads)
+            print(header)
 
     startTime = time.time()
-    display = TestingProgressDisplay(opts, len(tests), progressBar)
-    provider = TestProvider(tests, opts.maxTime)
-    runTests(opts.numThreads, litConfig, provider, display)
+    display = TestingProgressDisplay(opts, len(run.tests), progressBar)
+    try:
+        run.execute_tests(display, opts.numThreads, opts.maxTime,
+                          opts.useProcesses)
+    except KeyboardInterrupt:
+        sys.exit(2)
     display.finish()
 
+    testing_time = time.time() - startTime
     if not opts.quiet:
-        print 'Testing Time: %.2fs'%(time.time() - startTime)
+        print('Testing Time: %.2fs' % (testing_time,))
 
-    # Update results for any tests which weren't run.
-    for t in tests:
-        if t.result is None:
-            t.setResult(Test.UNRESOLVED, '', 0.0)
+    # Write out the test data, if requested.
+    if opts.output_path is not None:
+        write_test_results(run, litConfig, testing_time, opts.output_path)
 
     # List test results organized by kind.
     hasFailures = False
     byCode = {}
-    for t in tests:
-        if t.result not in byCode:
-            byCode[t.result] = []
-        byCode[t.result].append(t)
-        if t.result.isFailure:
+    for test in run.tests:
+        if test.result.code not in byCode:
+            byCode[test.result.code] = []
+        byCode[test.result.code].append(test)
+        if test.result.code.isFailure:
             hasFailures = True
 
-    # FIXME: Show unresolved and (optionally) unsupported tests.
-    for title,code in (('Unexpected Passing Tests', Test.XPASS),
-                       ('Failing Tests', Test.FAIL)):
+    # Print each test in any of the failing groups.
+    for title,code in (('Unexpected Passing Tests', lit.Test.XPASS),
+                       ('Failing Tests', lit.Test.FAIL),
+                       ('Unresolved Tests', lit.Test.UNRESOLVED),
+                       ('Unsupported Tests', lit.Test.UNSUPPORTED),
+                       ('Expected Failing Tests', lit.Test.XFAIL)):
+        if (lit.Test.XFAIL == code and not opts.show_xfail) or \
+           (lit.Test.UNSUPPORTED == code and not opts.show_unsupported):
+            continue
         elts = byCode.get(code)
         if not elts:
             continue
-        print '*'*20
-        print '%s (%d):' % (title, len(elts))
-        for t in elts:
-            print '    %s' % t.getFullName()
-        print
-
-    if opts.timeTests:
-        # Collate, in case we repeated tests.
-        times = {}
-        for t in tests:
-            key = t.getFullName()
-            times[key] = times.get(key, 0.) + t.elapsed
-
-        byTime = list(times.items())
-        byTime.sort(key = lambda (name,elapsed): elapsed)
-        if byTime:
-            Util.printHistogram(byTime, title='Tests')
-
-    for name,code in (('Expected Passes    ', Test.PASS),
-                      ('Expected Failures  ', Test.XFAIL),
-                      ('Unsupported Tests  ', Test.UNSUPPORTED),
-                      ('Unresolved Tests   ', Test.UNRESOLVED),
-                      ('Unexpected Passes  ', Test.XPASS),
-                      ('Unexpected Failures', Test.FAIL),):
+        print('*'*20)
+        print('%s (%d):' % (title, len(elts)))
+        for test in elts:
+            print('    %s' % test.getFullName())
+        sys.stdout.write('\n')
+
+    if opts.timeTests and run.tests:
+        # Order by time.
+        test_times = [(test.getFullName(), test.result.elapsed)
+                      for test in run.tests]
+        lit.util.printHistogram(test_times, title='Tests')
+
+    for name,code in (('Expected Passes    ', lit.Test.PASS),
+                      ('Passes With Retry  ', lit.Test.FLAKYPASS),
+                      ('Expected Failures  ', lit.Test.XFAIL),
+                      ('Unsupported Tests  ', lit.Test.UNSUPPORTED),
+                      ('Unresolved Tests   ', lit.Test.UNRESOLVED),
+                      ('Unexpected Passes  ', lit.Test.XPASS),
+                      ('Unexpected Failures', lit.Test.FAIL)):
         if opts.quiet and not code.isFailure:
             continue
         N = len(byCode.get(code,[]))
         if N:
-            print '  %s: %d' % (name,N)
+            print('  %s: %d' % (name,N))
+
+    if opts.xunit_output_file:
+        # Collect the tests, indexed by test suite
+        by_suite = {}
+        for result_test in run.tests:
+            suite = result_test.suite.config.name
+            if suite not in by_suite:
+                by_suite[suite] = {
+                                   'passes'   : 0,
+                                   'failures' : 0,
+                                   'tests'    : [] }
+            by_suite[suite]['tests'].append(result_test)
+            if result_test.result.code.isFailure:
+                by_suite[suite]['failures'] += 1
+            else:
+                by_suite[suite]['passes'] += 1
+        xunit_output_file = open(opts.xunit_output_file, "w")
+        xunit_output_file.write("<?xml version=\"1.0\" encoding=\"UTF-8\" ?>\n")
+        xunit_output_file.write("<testsuites>\n")
+        for suite_name, suite in by_suite.items():
+            safe_suite_name = suite_name.replace(".", "-")
+            xunit_output_file.write("<testsuite name='" + safe_suite_name + "'")
+            xunit_output_file.write(" tests='" + str(suite['passes'] + 
+              suite['failures']) + "'")
+            xunit_output_file.write(" failures='" + str(suite['failures']) + 
+              "'>\n")
+            for result_test in suite['tests']:
+                xunit_output_file.write(result_test.getJUnitXML() + "\n")
+            xunit_output_file.write("</testsuite>\n")
+        xunit_output_file.write("</testsuites>")
+        xunit_output_file.close()
 
     # If we encountered any additional errors, exit abnormally.
     if litConfig.numErrors:
-        print >>sys.stderr, '\n%d error(s), exiting.' % litConfig.numErrors
+        sys.stderr.write('\n%d error(s), exiting.\n' % litConfig.numErrors)
         sys.exit(2)
 
     # Warn about warnings.
     if litConfig.numWarnings:
-        print >>sys.stderr, '\n%d warning(s) in tests.' % litConfig.numWarnings
+        sys.stderr.write('\n%d warning(s) in tests.\n' % litConfig.numWarnings)
 
     if hasFailures:
         sys.exit(1)