Drumroll, please! All the awesome process work gets merged into

master...
ikollipara · Jul 24, 2015 · 963e0a9 · 963e0a9
2 parents 56e710e + 57e03af
commit 963e0a9
Show file tree

Hide file tree

Showing 23 changed files with 873 additions and 254 deletions.
diff --git a/CHANGELOG.md b/CHANGELOG.md
@@ -1,3 +1,60 @@
+# Version 2.0.0
+##### ? July 2015
+
+- Major overhaul of the multiprocessing system.  Tests always run in a separate
+  worker process, even when only one process is specified.  The default number
+  of processes is now the number of logical processors detected instead of 1.
+  Entire modules are now run in the same worker process by default to avoid both
+  the overhead of multiple processes loading the same module and the overhead of
+  running module and class setUp/tearDown multiple times redundantly.  Classes
+  or methods specified individually on the command-line will still be run in
+  their own worker process.  A ton of credit for this feature needs to go to Sam
+  Spilsbury, who put in considerable time and effort to actually code up the
+  initial pull request.  Fixes issues #68, #70.
+
+- Added a gitter chatroom link to the readme.
+
+- Support for acquiring and releasing arbitrary resources via the
+  `-i/--initializer` and `-z/--finalizer` options.  Use it to setup/teardown
+  things that an individual worker process will need exclusive access to apart
+  from the other worker processes.
+
+- We're back at 100% self-test coverage again.  Yay!!!
+
+- Twisted's skip functionality is caught and recorded as skips instead of
+  failures, if your `TestCase` subclasses `twisted.trial.unittest.TestCase` and
+  sets the class attribute `.skip` to `True`, or a test raises
+  `twisted.trial.unittest.SkipTest`.
+
+- Better handling of outside-of-test exceptions that occur inside worker
+  processes.
+
+- We now capture stderr that is emitted during tests and present it after tests
+  have run, just like we do with stdout.
+
+- Capturing stdout in worker processes more consistently works (no known bugs
+  left).
+
+- The headers for stdout and stderr are now yellow, for better color scheme
+  consistency (and so they don't get confused with skip headers).
+
+- Skip report headers now display the dotted test name in bold, just like other
+  headers do.  We are so consistent!
+
+- Fixed the skip report so it goes to the stream instead of stdout.
+
+- Disabled the annoying "Coverage.py warning: No data was collected." message
+  that started happening a lot, even though coverage was working just fine.
+
+- Colors now work on AppVeyor builds, all hail the pretty colors!  (Ironically,
+  they don't support Windows ansi colors, they wrote their own interpreters for
+  posix-style color escape codes.)
+
+- We now "close" the process pool instead of "terminating" it, which results in
+  much better behavior in pypy and Windows, especially for things like tearDown
+  stuff.
+
+
 # Version 1.11.0
 ##### 18 June 2015
 
@@ -53,7 +110,7 @@
 - Switched to cyan instead of blue on Windows only.
 
 - Stubbed in the beginnings of support for designating initialization to run in
-  each subprocess to obtain whatever resources might be needed by a single
+  each process to obtain whatever resources might be needed by a single
   process (like its own database, for example).
 
 
@@ -203,7 +260,7 @@ Issue #47.
 - Fixed a crash that could occur if an exception was raised during a test
   case's setUpClass() or tearDownClass()
 - We now explicitly terminate the thread pool and join() it.  This makes self
-  unit tests much easier to clean up on Windows, where the subprocesses would
+  unit tests much easier to clean up on Windows, where the processes would
   block deletion of temporary directories.
 - Set up continuous integration for Windows using AppVeyor.  Thanks to ionelmc
   for the tip!  Issue #11.
@@ -250,7 +307,7 @@ Issue #47.
   auto-detect all versions of python (in the form of pythonX.Y) in $PATH and
   run many permutations of self tests on each version.
 - Fixed a crash that could occur if discovery did not find any tests and
-  subprocesses was set higher than one.
+  processes was set higher than one.
 
 - Fixed lots of tests so that they would succeed in all environments.
 
@@ -346,7 +403,7 @@ Issue #47.
 - Clean - Low redundancy in output. Result stats for each test is lined up in a
   vertical column.
 
-- Fast - Can run tests in independent subprocesses.
+- Fast - Can run tests in independent processes.
 
 - Powerful - Multi-target + auto-discovery.
 

diff --git a/README-pypi.rst b/README-pypi.rst
@@ -1,9 +1,7 @@
 Green
 =====
 
-Green is a colorful, clean, fast and powerful test runner for Python unit
-tests.
-
+Green is a clean, colorful, fast, multi-process test runner for Python.
 
 Documentation
 -------------
@@ -16,21 +14,22 @@ project.
 Features
 --------
 
-- **Colorful** - Terminal output makes good use of color when the terminal supports it.
 - **Clean** - Low redundancy in output. Result stats for each test is lined up in a vertical column.
-- **Fast** - Can run tests in independent subprocesses.
+- **Colorful** - Terminal output makes good use of color when the terminal supports it.
+- **Fast** - Can run tests in independent processes.  (Only 1 by default for compatibility, try "-s 0" for best results)
 - **Powerful** - Multi-target + auto-discovery.
 - **Traditional** - Use the normal ``unittest`` classes and methods for your unit tests.
 - **Descriptive** - Four verbosity levels, from just dots to full docstring output.
 - **Convenient** - Bash-completion and ZSH-completion of options and test targets.
 - **Thorough** - Built-in, optional integration with `coverage`_
-- **Modern** - Supports Python 2.7, 3.3, 3.4, and `PyPy`_
-- **Portable** - Completely supports OS X, Linux, and BSDs (and maybe Windows).
+- **Modern** - Supports Python 2.7, 3.4+, `PyPy`_ and `PyPy3`_
+- **Portable** - Completely supports OS X, Linux, and BSDs (and Windows most of the time).
 - **Living** - This project grows and changes.  See the `changelog`_
 
 
 .. _GitHub page for Green: https://github.com/CleanCut/green#green
 .. _tutorial on how to correctly layout unit tests: https://github.com/CleanCut/green#unit-test-structure-tutorial
 .. _coverage: http://nedbatchelder.com/code/coverage/
 .. _PyPy: http://pypy.org
+.. _PyPy3: http://pypy.org
 .. _changelog: https://github.com/CleanCut/green/blob/master/CHANGELOG.md
diff --git a/README.md b/README.md
@@ -1,4 +1,4 @@
-Green
+Green: A clean, colorful, fast, multi-process test runner.
 =====
 
 [![Join the chat at https://gitter.im/CleanCut/green](https://badges.gitter.im/Join%20Chat.svg)](https://gitter.im/CleanCut/green?utm_source=badge&utm_medium=badge&utm_campaign=pr-badge&utm_content=badge)
@@ -8,8 +8,6 @@ Green
 [![Windows Build Status](http://img.shields.io/appveyor/ci/CleanCut/green.svg?style=flat)](https://ci.appveyor.com/project/CleanCut/green)
 [![Coverage Status](https://img.shields.io/coveralls/CleanCut/green.svg?style=flat)](https://coveralls.io/r/CleanCut/green?branch=master)
 
-Green is a **clean, colorful test runner** for Python unit tests.
-
 #### Table of Contents
 
 1. [Features](https://github.com/CleanCut/green#features)
@@ -31,17 +29,17 @@ Green is a **clean, colorful test runner** for Python unit tests.
 Features
 --------
 
-- **Colorful** - Terminal output makes good use of color when the terminal supports it.
 - **Clean** - Low redundancy in output. Result statistics for each test is vertically aligned.
-- **Fast** - Tests can run in independent subprocesses.
+- **Colorful** - Terminal output makes good use of color when the terminal supports it.
+- **Fast** - Tests run in independent processes.  (Only 1 by default for compatibility, try `-s 0` for best results)
 - **Powerful** - Multi-target + auto-discovery.
 - **Traditional** - Use the normal `unittest` classes and methods for your unit tests.
 - **Descriptive** - Four verbosity levels, from just dots to full docstring output.
 - **Convenient** - Bash-completion and ZSH-completion of options and test targets.
 - **Thorough** - Built-in, optional integration with
   [coverage](http://nedbatchelder.com/code/coverage/).
-- **Modern** - Supports Python 2.7, 3.4+, and [PyPy](http://pypy.org).
-- **Portable** - Completely supports OS X, Linux, and BSDs (and maybe Windows).
+- **Modern** - Supports Python 2.7, 3.4+, [PyPy](http://pypy.org), and [PyPy3](http://pypy.org).
+- **Portable** - Completely supports OS X, Linux, and BSDs (and Windows most of the time).
 - **Living** - This project grows and changes.  See the
   [changelog](https://github.com/CleanCut/green/blob/master/CHANGELOG.md)
 

diff --git a/appveyor.yml b/appveyor.yml
@@ -12,6 +12,6 @@ install:
       - "%PYTHON%/python.exe C:/get-pip.py"
       - "%PYTHON%/Scripts/pip.exe install -r requirements.txt"
 test_script:
-      - "%PYTHON%/python.exe -m green.cmdline -vvv green"
+    - "%PYTHON%/python.exe -m green.cmdline -vvv -t green"
 platform:
     - AnyCPU
diff --git a/cli-options.txt b/cli-options.txt
@@ -16,16 +16,31 @@ Target Specification:
                         introspection of the object will be loaded.
 
 Concurrency Options:
-  -s NUM, --subprocesses NUM
-                        Number of subprocesses to use to run tests. Note that
+  -s NUM, --processes NUM
+                        Number of processes to use to run tests. Note that
                         your tests need to be written to avoid using the same
                         resources (temp files, sockets, ports, etc.) for the
-                        multi-process mode to work well. Default is 1, meaning
-                        disable using subprocesses. 0 means try to autodetect
-                        the number of CPUs in the system. Note that for a
-                        small number of trivial tests, running everything in a
+                        multi-process mode to work well (--initializer and
+                        --finalizer can help provision per-process resources).
+                        Default is 1, meaning just use a single process
+                        (maximizes compatibility with standard unittest and
+                        other test runners). 0 means try to autodetect the
+                        number of CPUs in the system. Note that for a small
+                        number of trivial tests, running everything in a
                         single process may be faster than the overhead of
-                        initializing all the subprocesses.
+                        initializing all the processes.
+  -i EXECUTABLE_FILE, --initializer EXECUTABLE_FILE
+                        Executable to run inside of a single worker process
+                        before it starts running tests. This is the way to
+                        provision external resources that each concurrent
+                        worker process needs to have exclusive access to. Can
+                        be a relative or absolute path.
+  -z EXECUTABLE_FILE, --finalizer EXECUTABLE_FILE
+                        Executable to run inside of a single worker process
+                        after it completes running tests and the process is
+                        about to be destroyed. This is the way to reclaim
+                        resources provisioned with --initializer. Can be a
+                        relative or absolute path.
 
 Format Options:
   -m, --html            THIS OPTION WILL BE REMOVED SOON UNLESS PEOPLE ASK FOR

diff --git a/green/VERSION b/green/VERSION
@@ -1 +1 @@
-1.11.0
+2.0.0
diff --git a/green/cmdline.py b/green/cmdline.py
@@ -70,7 +70,7 @@ def main(testing=False, coverage_testing=False):
     # Discover/Load the test suite
     if testing:
         test_suite = None
-    else:
+    else: # pragma: no cover
         test_suite = loadTargets(args.targets, file_pattern = args.file_pattern)
 
     # We didn't even load 0 tests...