Don't use fork to start the notebook in js tests

It can encounter a weird segfault on OS X with sqlite when Qt is present (?!)

The main reason to use the fork was to get the port number,
but this is easy now that notebooks write a server-info file.

Further advantage is that the symptom of a failed server start
is no longer silence and hanging tests, but an actual failure with the server's log output.

IPython/testing/iptestcontroller.py

@@ -1,569 +1,595 b''
 # -*- coding: utf-8 -*-
 """IPython Test Process Controller
 
 This module runs one or more subprocesses which will actually run the IPython
 test suite.
 
 """
 
 #-----------------------------------------------------------------------------
 #  Copyright (C) 2009-2011  The IPython Development Team
 #
 #  Distributed under the terms of the BSD License.  The full license is in
 #  the file COPYING, distributed as part of this software.
 #-----------------------------------------------------------------------------
 
 #-----------------------------------------------------------------------------
 # Imports
 #-----------------------------------------------------------------------------
 from __future__ import print_function
 
 import argparse
+import json
 import multiprocessing.pool
-from multiprocessing import Process, Queue
 import os
 import shutil
 import signal
 import sys
 import subprocess
 import time
 
-from .iptest import have, test_group_names as py_test_group_names, test_sections
+from .iptest import have, test_group_names as py_test_group_names, test_sections, StreamCapturer
 from IPython.utils.path import compress_user
 from IPython.utils.py3compat import bytes_to_str
 from IPython.utils.sysinfo import get_sys_info
 from IPython.utils.tempdir import TemporaryDirectory
 
 
 class TestController(object):
     """Run tests in a subprocess
     """
     #: str, IPython test suite to be executed.
     section = None
     #: list, command line arguments to be executed
     cmd = None
     #: dict, extra environment variables to set for the subprocess
     env = None
     #: list, TemporaryDirectory instances to clear up when the process finishes
     dirs = None
     #: subprocess.Popen instance
     process = None
     #: str, process stdout+stderr
     stdout = None
 
     def __init__(self):
         self.cmd = []
         self.env = {}
         self.dirs = []
 
     def setup(self):
         """Create temporary directories etc.
         
         This is only called when we know the test group will be run. Things
         created here may be cleaned up by self.cleanup().
         """
         pass
 
     def launch(self, buffer_output=False):
         # print('*** ENV:', self.env)  # dbg
         # print('*** CMD:', self.cmd)  # dbg
         env = os.environ.copy()
         env.update(self.env)
         output = subprocess.PIPE if buffer_output else None
         stdout = subprocess.STDOUT if buffer_output else None
         self.process = subprocess.Popen(self.cmd, stdout=output,
                 stderr=stdout, env=env)
 
     def wait(self):
         self.stdout, _ = self.process.communicate()
         return self.process.returncode
 
     def print_extra_info(self):
         """Print extra information about this test run.
         
         If we're running in parallel and showing the concise view, this is only
         called if the test group fails. Otherwise, it's called before the test
         group is started.
         
         The base implementation does nothing, but it can be overridden by
         subclasses.
         """
         return
 
     def cleanup_process(self):
         """Cleanup on exit by killing any leftover processes."""
         subp = self.process
         if subp is None or (subp.poll() is not None):
             return  # Process doesn't exist, or is already dead.
 
         try:
             print('Cleaning up stale PID: %d' % subp.pid)
             subp.kill()
         except: # (OSError, WindowsError) ?
             # This is just a best effort, if we fail or the process was
             # really gone, ignore it.
             pass
         else:
             for i in range(10):
                 if subp.poll() is None:
                     time.sleep(0.1)
                 else:
                     break
 
         if subp.poll() is None:
             # The process did not die...
             print('... failed. Manual cleanup may be required.')
 
     def cleanup(self):
         "Kill process if it's still alive, and clean up temporary directories"
         self.cleanup_process()
         for td in self.dirs:
             td.cleanup()
 
     __del__ = cleanup
 
 class PyTestController(TestController):
     """Run Python tests using IPython.testing.iptest"""
     #: str, Python command to execute in subprocess
     pycmd = None
 
     def __init__(self, section):
         """Create new test runner."""
         TestController.__init__(self)
         self.section = section
         # pycmd is put into cmd[2] in PyTestController.launch()
         self.cmd = [sys.executable, '-c', None, section]
         self.pycmd = "from IPython.testing.iptest import run_iptest; run_iptest()"
 
     def setup(self):
         ipydir = TemporaryDirectory()
         self.dirs.append(ipydir)
         self.env['IPYTHONDIR'] = ipydir.name
         self.workingdir = workingdir = TemporaryDirectory()
         self.dirs.append(workingdir)
         self.env['IPTEST_WORKING_DIR'] = workingdir.name
         # This means we won't get odd effects from our own matplotlib config
         self.env['MPLCONFIGDIR'] = workingdir.name
 
     @property
     def will_run(self):
         try:
             return test_sections[self.section].will_run
         except KeyError:
             return True
 
     def add_xunit(self):
         xunit_file = os.path.abspath(self.section + '.xunit.xml')
         self.cmd.extend(['--with-xunit', '--xunit-file', xunit_file])
 
     def add_coverage(self):
         try:
             sources = test_sections[self.section].includes
         except KeyError:
             sources = ['IPython']
 
         coverage_rc = ("[run]\n"
                        "data_file = {data_file}\n"
                        "source =\n"
                        "  {source}\n"
                       ).format(data_file=os.path.abspath('.coverage.'+self.section),
                                source="\n  ".join(sources))
         config_file = os.path.join(self.workingdir.name, '.coveragerc')
         with open(config_file, 'w') as f:
             f.write(coverage_rc)
 
         self.env['COVERAGE_PROCESS_START'] = config_file
         self.pycmd = "import coverage; coverage.process_startup(); " + self.pycmd
 
     def launch(self, buffer_output=False):
         self.cmd[2] = self.pycmd
         super(PyTestController, self).launch(buffer_output=buffer_output)
 
 js_prefix = 'js/'
 
 def get_js_test_dir():
     import IPython.html.tests as t
     return os.path.join(os.path.dirname(t.__file__), '')
 
 def all_js_groups():
     import glob
     test_dir = get_js_test_dir()
     all_subdirs = glob.glob(test_dir + '*/')
     return [js_prefix+os.path.relpath(x, test_dir) for x in all_subdirs if os.path.relpath(x, test_dir) != '__pycache__']
 
 class JSController(TestController):
     """Run CasperJS tests """
     def __init__(self, section):
         """Create new test runner."""
         TestController.__init__(self)
         self.section = section
         js_test_dir = get_js_test_dir()
         includes = '--includes=' + os.path.join(js_test_dir,'util.js')
         test_cases = os.path.join(js_test_dir, self.section[len(js_prefix):])
         self.cmd = ['casperjs', 'test', includes, test_cases]
 
     def setup(self):
         self.ipydir = TemporaryDirectory()
         self.nbdir = TemporaryDirectory()
         self.dirs.append(self.ipydir)
         self.dirs.append(self.nbdir)
         os.makedirs(os.path.join(self.nbdir.name, os.path.join(u'sub ∂ir1', u'sub ∂ir 1a')))
         os.makedirs(os.path.join(self.nbdir.name, os.path.join(u'sub ∂ir2', u'sub ∂ir 1b')))
         
         # start the ipython notebook, so we get the port number
+        self.server_port = 0
         self._init_server()
-        self.cmd.append('--port=%s' % self.server_port)
+        if self.server_port:
+            self.cmd.append("--port=%i" % self.server_port)
 
     def print_extra_info(self):
         print("Running tests with notebook directory %r" % self.nbdir.name)
 
     @property
     def will_run(self):
         return all(have[a] for a in ['zmq', 'tornado', 'jinja2', 'casperjs'])
 
     def _init_server(self):
         "Start the notebook server in a separate process"
-        self.queue = q = Queue()
-        self.server = Process(target=run_webapp, args=(q, self.ipydir.name, self.nbdir.name))
-        self.server.start()
-        self.server_port = q.get()
+        self.server_command = command = [sys.executable,
+            '-m', 'IPython.html',
+            '--no-browser',
+            '--ipython-dir', self.ipydir.name,
+            '--notebook-dir', self.nbdir.name,
+        ]
+        # ipc doesn't work on Windows, and darwin has crazy-long temp paths,
+        # which run afoul of ipc's maximum path length.
+        if sys.platform.startswith('linux'):
+            command.append('--KernelManager.transport=ipc')
+        self.stream_capturer = c = StreamCapturer()
+        c.start()
+        self.server = subprocess.Popen(command, stdout=c.writefd, stderr=subprocess.STDOUT)
+        self.server_info_file = os.path.join(self.ipydir.name,
+            'profile_default', 'security', 'nbserver-%i.json' % self.server.pid
+        )
+        self._wait_for_server()
+    
+    def _wait_for_server(self):
+        """Wait 30 seconds for the notebook server to start"""
+        for i in range(300):
+            if self.server.poll() is not None:
+                return self._failed_to_start()
+            if os.path.exists(self.server_info_file):
+                self._load_server_info()
+                return
+            time.sleep(0.1)
+        print("Notebook server-info file never arrived: %s" % self.server_info_file,
+            file=sys.stderr
+        )
+    
+    def _failed_to_start(self):
+        """Notebook server exited prematurely"""
+        captured = self.stream_capturer.get_buffer().decode('utf-8', 'replace')
+        print("Notebook failed to start: ", file=sys.stderr)
+        print(self.server_command)
+        print(captured, file=sys.stderr)
+    
+    def _load_server_info(self):
+        """Notebook server started, load connection info from JSON"""
+        with open(self.server_info_file) as f:
+            info = json.load(f)
+        self.server_port = info['port']
 
     def cleanup(self):
+        self.stream_capturer.halt()
+        try:
         self.server.terminate()
-        self.server.join()
+        except OSError:
+            # already dead
+            pass
+        self.server.wait()
         TestController.cleanup(self)
 
-def run_webapp(q, ipydir, nbdir, loglevel=0):
-    """start the IPython Notebook, and pass port back to the queue"""
-    import os
-    import IPython.html.notebookapp as nbapp
-    import sys
-    sys.stderr = open(os.devnull, 'w')
-    server = nbapp.NotebookApp()
-    args = ['--no-browser']
-    args.extend(['--ipython-dir', ipydir,
-                 '--notebook-dir', nbdir,
-                 '--log-level', str(loglevel),
-    ])
-    # ipc doesn't work on Windows, and darwin has crazy-long temp paths,
-    # which run afoul of ipc's maximum path length.
-    if sys.platform.startswith('linux'):
-        args.append('--KernelManager.transport=ipc')
-    server.initialize(args)
-    # communicate the port number to the parent process
-    q.put(server.port)
-    server.start()
 
 def prepare_controllers(options):
     """Returns two lists of TestController instances, those to run, and those
     not to run."""
     testgroups = options.testgroups
 
     if testgroups:
         py_testgroups = [g for g in testgroups if (g in py_test_group_names) \
                                                 or g.startswith('IPython.')]
         if 'js' in testgroups:
             js_testgroups = all_js_groups()
         else:
             js_testgroups = [g for g in testgroups if g not in py_testgroups]
     else:
         py_testgroups = py_test_group_names
         js_testgroups = all_js_groups()
         if not options.all:
             test_sections['parallel'].enabled = False
 
     c_js = [JSController(name) for name in js_testgroups]
     c_py = [PyTestController(name) for name in py_testgroups]
 
     configure_py_controllers(c_py, xunit=options.xunit,
             coverage=options.coverage, subproc_streams=options.subproc_streams,
             extra_args=options.extra_args)
 
     controllers = c_py + c_js
     to_run = [c for c in controllers if c.will_run]
     not_run = [c for c in controllers if not c.will_run]
     return to_run, not_run
 
 def configure_py_controllers(controllers, xunit=False, coverage=False,
                              subproc_streams='capture', extra_args=()):
     """Apply options for a collection of TestController objects."""
     for controller in controllers:
         if xunit:
             controller.add_xunit()
         if coverage:
             controller.add_coverage()
         controller.env['IPTEST_SUBPROC_STREAMS'] = subproc_streams
         controller.cmd.extend(extra_args)
 
 def do_run(controller, buffer_output=True):
     """Setup and run a test controller.
     
     If buffer_output is True, no output is displayed, to avoid it appearing
     interleaved. In this case, the caller is responsible for displaying test
     output on failure.
     
     Returns
     -------
     controller : TestController
       The same controller as passed in, as a convenience for using map() type
       APIs.
     exitcode : int
       The exit code of the test subprocess. Non-zero indicates failure.
     """
     try:
         try:
             controller.setup()
             if not buffer_output:
                 controller.print_extra_info()
             controller.launch(buffer_output=buffer_output)
         except Exception:
             import traceback
             traceback.print_exc()
             return controller, 1  # signal failure
 
         exitcode = controller.wait()
         return controller, exitcode
 
     except KeyboardInterrupt:
         return controller, -signal.SIGINT
     finally:
         controller.cleanup()
 
 def report():
     """Return a string with a summary report of test-related variables."""
     inf = get_sys_info()
     out = []
     def _add(name, value):
         out.append((name, value))
 
     _add('IPython version', inf['ipython_version'])
     _add('IPython commit', "{} ({})".format(inf['commit_hash'], inf['commit_source']))
     _add('IPython package', compress_user(inf['ipython_path']))
     _add('Python version', inf['sys_version'].replace('\n',''))
     _add('sys.executable', compress_user(inf['sys_executable']))
     _add('Platform', inf['platform'])
 
     width = max(len(n) for (n,v) in out)
     out = ["{:<{width}}: {}\n".format(n, v, width=width) for (n,v) in out]
 
     avail = []
     not_avail = []
 
     for k, is_avail in have.items():
         if is_avail:
             avail.append(k)
         else:
             not_avail.append(k)
 
     if avail:
         out.append('\nTools and libraries available at test time:\n')
         avail.sort()
         out.append('   ' + ' '.join(avail)+'\n')
 
     if not_avail:
         out.append('\nTools and libraries NOT available at test time:\n')
         not_avail.sort()
         out.append('   ' + ' '.join(not_avail)+'\n')
 
     return ''.join(out)
 
 def run_iptestall(options):
     """Run the entire IPython test suite by calling nose and trial.
 
     This function constructs :class:`IPTester` instances for all IPython
     modules and package and then runs each of them.  This causes the modules
     and packages of IPython to be tested each in their own subprocess using
     nose.
 
     Parameters
     ----------
 
     All parameters are passed as attributes of the options object.
 
     testgroups : list of str
       Run only these sections of the test suite. If empty, run all the available
       sections.
 
     fast : int or None
       Run the test suite in parallel, using n simultaneous processes. If None
       is passed, one process is used per CPU core. Default 1 (i.e. sequential)
 
     inc_slow : bool
       Include slow tests, like IPython.parallel. By default, these tests aren't
       run.
 
     xunit : bool
       Produce Xunit XML output. This is written to multiple foo.xunit.xml files.
 
     coverage : bool or str
       Measure code coverage from tests. True will store the raw coverage data,
       or pass 'html' or 'xml' to get reports.
 
     extra_args : list
       Extra arguments to pass to the test subprocesses, e.g. '-v'
     """
     to_run, not_run = prepare_controllers(options)
 
     def justify(ltext, rtext, width=70, fill='-'):
         ltext += ' '
         rtext = (' ' + rtext).rjust(width - len(ltext), fill)
         return ltext + rtext
 
     # Run all test runners, tracking execution time
     failed = []
     t_start = time.time()
 
     print()
     if options.fast == 1:
         # This actually means sequential, i.e. with 1 job
         for controller in to_run:
             print('Test group:', controller.section)
             sys.stdout.flush()  # Show in correct order when output is piped
             controller, res = do_run(controller, buffer_output=False)
             if res:
                 failed.append(controller)
                 if res == -signal.SIGINT:
                     print("Interrupted")
                     break
             print()
 
     else:
         # Run tests concurrently
         try:
             pool = multiprocessing.pool.ThreadPool(options.fast)
             for (controller, res) in pool.imap_unordered(do_run, to_run):
                 res_string = 'OK' if res == 0 else 'FAILED'
                 print(justify('Test group: ' + controller.section, res_string))
                 if res:
                     controller.print_extra_info()
                     print(bytes_to_str(controller.stdout))
                     failed.append(controller)
                     if res == -signal.SIGINT:
                         print("Interrupted")
                         break
         except KeyboardInterrupt:
             return
 
     for controller in not_run:
         print(justify('Test group: ' + controller.section, 'NOT RUN'))
 
     t_end = time.time()
     t_tests = t_end - t_start
     nrunners = len(to_run)
     nfail = len(failed)
     # summarize results
     print('_'*70)
     print('Test suite completed for system with the following information:')
     print(report())
     took = "Took %.3fs." % t_tests
     print('Status: ', end='')
     if not failed:
         print('OK (%d test groups).' % nrunners, took)
     else:
         # If anything went wrong, point out what command to rerun manually to
         # see the actual errors and individual summary
         failed_sections = [c.section for c in failed]
         print('ERROR - {} out of {} test groups failed ({}).'.format(nfail,
                                   nrunners, ', '.join(failed_sections)), took)
         print()
         print('You may wish to rerun these, with:')
         print('  iptest', *failed_sections)
         print()
 
     if options.coverage:
         from coverage import coverage
         cov = coverage(data_file='.coverage')
         cov.combine()
         cov.save()
 
         # Coverage HTML report
         if options.coverage == 'html':
             html_dir = 'ipy_htmlcov'
             shutil.rmtree(html_dir, ignore_errors=True)
             print("Writing HTML coverage report to %s/ ... " % html_dir, end="")
             sys.stdout.flush()
 
             # Custom HTML reporter to clean up module names.
             from coverage.html import HtmlReporter
             class CustomHtmlReporter(HtmlReporter):
                 def find_code_units(self, morfs):
                     super(CustomHtmlReporter, self).find_code_units(morfs)
                     for cu in self.code_units:
                         nameparts = cu.name.split(os.sep)
                         if 'IPython' not in nameparts:
                             continue
                         ix = nameparts.index('IPython')
                         cu.name = '.'.join(nameparts[ix:])
 
             # Reimplement the html_report method with our custom reporter
             cov._harvest_data()
             cov.config.from_args(omit='*{0}tests{0}*'.format(os.sep), html_dir=html_dir,
                                  html_title='IPython test coverage',
                                 )
             reporter = CustomHtmlReporter(cov, cov.config)
             reporter.report(None)
             print('done.')
 
         # Coverage XML report
         elif options.coverage == 'xml':
             cov.xml_report(outfile='ipy_coverage.xml')
 
     if failed:
         # Ensure that our exit code indicates failure
         sys.exit(1)
 
 argparser = argparse.ArgumentParser(description='Run IPython test suite')
 argparser.add_argument('testgroups', nargs='*',
                     help='Run specified groups of tests. If omitted, run '
                     'all tests.')
 argparser.add_argument('--all', action='store_true',
                     help='Include slow tests not run by default.')
 argparser.add_argument('-j', '--fast', nargs='?', const=None, default=1, type=int,
                     help='Run test sections in parallel. This starts as many '
                     'processes as you have cores, or you can specify a number.')
 argparser.add_argument('--xunit', action='store_true',
                     help='Produce Xunit XML results')
 argparser.add_argument('--coverage', nargs='?', const=True, default=False,
                     help="Measure test coverage. Specify 'html' or "
                     "'xml' to get reports.")
 argparser.add_argument('--subproc-streams', default='capture',
                     help="What to do with stdout/stderr from subprocesses. "
                     "'capture' (default), 'show' and 'discard' are the options.")
 
 def default_options():
     """Get an argparse Namespace object with the default arguments, to pass to
     :func:`run_iptestall`.
     """
     options = argparser.parse_args([])
     options.extra_args = []
     return options
 
 def main():
     # iptest doesn't work correctly if the working directory is the
     # root of the IPython source tree. Tell the user to avoid
     # frustration.
     if os.path.exists(os.path.join(os.getcwd(),
                                    'IPython', 'testing', '__main__.py')):
         print("Don't run iptest from the IPython source directory",
               file=sys.stderr)
         sys.exit(1)
     # Arguments after -- should be passed through to nose. Argparse treats
     # everything after -- as regular positional arguments, so we separate them
     # first.
     try:
         ix = sys.argv.index('--')
     except ValueError:
         to_parse = sys.argv[1:]
         extra_args = []
     else:
         to_parse = sys.argv[1:ix]
         extra_args = sys.argv[ix+1:]
 
     options = argparser.parse_args(to_parse)
     options.extra_args = extra_args
 
     run_iptestall(options)
 
 
 if __name__ == '__main__':
     main()