WorkGroup provides way to spawn goroutines that work on a common task
and wait for their completion. It is modelled after
https://godoc.org/golang.org/x/sync/errgroup but is not equal to it.

See WorkGroup docstring for details.

Add sync.Once and sync.WaitGroup.

Add .Context, .background .with_cancel and .with_value. There is no
support for deadline/timeout yet, since Go time package is not yet
mirrored.

Contrary to Go stdlib version, we also support context.merge that takes
https://godoc.org/lab.nexedi.com/kirr/go123/xcontext#hdr-Merging_contexts
for its inspiration.

See https://blog.golang.org/context for overview of contexts.

We are going to add more packages to mirror Go stdlib (e.g. next is
context). Organize a dedicated section for what is additionally provided
besides mirroring Go language and refer to that place only once from the
top-level abstract.

Send/recv on the nil channel block forever; close panics.
If a nil channel is used in select - corresponding case is never selected.

Setting channel to nil is a usual idiom in Go to disable processing some
cases in select. Nil channel is also used as "done" for e.g.
context.Background() - for contexts that can be never canceled.

It was deprecated and scheduled to be removed in 942ee900 (golang:
Deprecate @method(cls) in favour of @func(cls)).

	pypy-gevent runtests: commands[0] | gpython -m pytest gpython/ golang/
	Traceback (most recent call last):
	  File "/home/kirr/src/tools/go/pygolang/.tox/pypy-gevent/bin/gpython", line 10, in <module>
	    sys.exit(main())
	  File "/home/kirr/src/tools/go/pygolang/.tox/pypy-gevent/site-packages/gpython/__init__.py", line 153, in main
	    _ = monkey.patch_all()      # XXX sys=True ?
	  File "/home/kirr/src/tools/go/pygolang/.tox/pypy-gevent/site-packages/gevent/monkey.py", line 976, in patch_all
	    patch_thread(Event=Event, _warnings=_warnings)
	  File "/home/kirr/src/tools/go/pygolang/.tox/pypy-gevent/site-packages/gevent/monkey.py", line 178, in ignores
	    return func(*args, **kwargs)
	  File "/home/kirr/src/tools/go/pygolang/.tox/pypy-gevent/site-packages/gevent/monkey.py", line 588, in patch_thread
	    _patch_existing_locks(threading_mod)
	  File "/home/kirr/src/tools/go/pygolang/.tox/pypy-gevent/site-packages/gevent/monkey.py", line 492, in _patch_existing_locks
	    if o._RLock__owner is not None:
	AttributeError: 'thread.RLock' object has no attribute '_RLock__owner'

It will be fixed on next Gevent release, however until then the crash is there
and we have to workaround: we can skip patching existing locks as it will be
the same behaviour of next Gevent release, since its just not possible to patch
locks created via instantiated C-level classes:

https://github.com/gevent/gevent/commit/d0e04658

We are doing monkey-patching very early, so it should be safe.

Channel tests are passing on an idle machine. However if there is
another load - e.g. in the presence of simultaneous Firefox start,
channel tests break, for example:

	    def test_select():
	        ...

	        # non-blocking try recv: ok
	        ch = chan()
	        done = chan()
	        def _():
	            for i in range(N):
	                ch.send(i)
	            done.close()
	        go(_)

	        for i in range(N):
	            tdelay()
	            if i % 2:
	                _, _rx = select(
	                        ch.recv,
	                        default,
	                )
	                assert (_, _rx) == (0, i)
	            else:
	                _, _rx = select(
	                        ch.recv_,
	                        default,
	                )
	>               assert (_, _rx) == (0, (i, True))
	E               assert (1, None) == (0, (320, True))
	E                 At index 0 diff: 1 != 0
	E                 Use -v to get the full diff

	golang_test.py:209: AssertionError

The failure here is that it was default case selected, not ch.recv_. The
default was selected because the sending goroutine was not fast to
enqueue next send before we tried to receive. We were trying to make
sure that the sender will be enqueued first via adding 1ms time delay
before trying to receive, but in the presence of concurrent load spikes
that turns out to be not enough.

We could try to fix the test by increasing the time to sleep in tdelay,
make the tests more slow and still not 100% reliable. However we can
change the tests to instead actually wait for the condition that is
semantically required: a sender enqueued on the channel.

Do that everywhere where tdelay was used.

Now tests are faster (it was ~3s total, now it is ~ 0.5s total) and pass
irregardless of whether the machine is idle or otherwise loaded.

Debian dropped Python3.6 from python3-distutils. Without it I'm no
longer being able to run tests on that python version:

	(neo) (z-dev) (g.env) kirr@deco:~/src/tools/go/pygolang$ tox -e py36-thread
	execution failed:  -- Traceback (most recent call last):
	  File "<string>", line 1, in <module>
	ModuleNotFoundError: No module named 'distutils.sysconfig'

	GLOB sdist-make: /home/kirr/src/tools/go/pygolang/setup.py
	py36-thread create: /home/kirr/src/tools/go/pygolang/.tox/py36-thread
	ERROR: invocation failed (exit code 1), logfile: /home/kirr/src/tools/go/pygolang/.tox/py36-thread/log/py36-thread-0.log
	ERROR: actionid: py36-thread
	msg: getenv
	cmdargs: '/usr/bin/python3 -m virtualenv --python /usr/bin/python3.6 py36-thread'

	Traceback (most recent call last):
	  File "/usr/lib/python3/dist-packages/virtualenv.py", line 25, in <module>
	    import distutils.sysconfig
	ModuleNotFoundError: No module named 'distutils.sysconfig'
	Running virtualenv with interpreter /usr/bin/python3.6

	ERROR: Error creating virtualenv. Note that some special characters (e.g. ':' and unicode symbols) in paths are not supported by virtualenv. Error details: InvocationError('/usr/bin/python3 -m virtualenv --python /usr/bin/python3.6 py36-thread (see /home/kirr/src/tools/go/pygolang/.tox/py36-thread/log/py36-thread-0.log)', 1)

https://bugs.debian.org/918881#19

PyPy7 always pre-imports the time module. Without the change tests were
failing:

	Traceback (most recent call last):
	  File "/home/kirr/src/tools/go/pygolang/.tox/pypy3-gevent/bin/gpython", line 10, in <module>
	    sys.exit(main())
	  File "/home/kirr/src/tools/go/pygolang/.tox/pypy3-gevent/site-packages/gpython/__init__.py", line 145, in main
	    '\n\n\t%s\n\nsys.modules:\n\n\t%s' % (bad, sysmodv))
	RuntimeError: gpython: internal error: the following modules are pre-imported, but must be not:

	        ['time']

	sys.modules:

	        ['__future__', '__main__', '__pypy__', '__pypy__._pypydatetime', '__pypy__.builders', '__pypy__.intop', '__pypy__.os', '__pypy__.thread', '__pypy__.time', '_ast', '_bootlocale', '_codecs', '_collections', '_collections_abc', '_continuation', '_csv', '_frozen_importlib', '_frozen_importlib_external', '_imp', '_io', '_locale', '_multibytecodec', '_operator', '_rawffi', '_rawffi.alt', '_signal', '_sre', '_structseq', '_thread', '_warnings', '_weakref', '_weakrefset', 'abc', 'array', 'builtins', 'codecs', 'copyreg', 'encodings', 'encodings.aliases', 'encodings.ascii', 'encodings.latin_1', 'encodings.utf_8', 'errno', 'gc', 'genericpath', 'gpython', 'marshal', 'os', 'os.path', 'posix', 'posixpath', 'pwd', 're', 'site', 'sre_compile', 'sre_constants', 'sre_parse', 'stat', 'sys', 'time', 'unicodedata']

Debian started to ship PyPy3:

	https://packages.debian.org/search?keywords=pypy3

Since we already have func (see 5146eb0b "Add support for defer &
recover") we can use @func for both plain functions and for methods.

For example instead of

	@func
	def my_function(...):
	    ...

	@method(MyClass)			<-- NOTE
	def my_method(self, ...):
	    ...

have it as

	@func
	def my_function(...):
	    ...

	@func(MyClass)				<-- NOTE
	def my_method(self, ...):
	    ...

which looks more similar to Go and exposes less golang keywords to a user.

Pymain was not adding e.g. directory of executed file to sys.path, and
as the result if there were e.g. 2 files

	dir/hello.py	# imports world
	dir/world.py

running `gpython dir/hello.py` would fail to import world.

The case for interactive console was also failing to setup sys.argv as
empty, so it was containing ['/path/to/gpython']. It was also, contrary
to standard python, unconditionally showing '>>>' prompt even when stdin
was not a tty.

Fix all that and add a test to cover pymain functionality.

Provide gpython interpreter, that sets UTF-8 as default encoding,
integrates gevent and puts go, chan, select etc into builtin namespace.
Please see details in the documentation added by this patch.

pymain was taken from go123@7a476082
(go123: tracing/pyruntraced: New tool to run Python code with tracepoints activated (draft))

We describe golang/testdata/ in MANIFEST.in, and that makes testdata to
be included into sdist. However testdata does not get installed.

Since, we want to be able to run pygolang tests, even on installed
package, we need the testdata files to be present there as well.

Fix it.

See https://setuptools.readthedocs.io/en/latest/setuptools.html#including-data-files
for setuptools-related details.

Move version definition from setup.py to golang module so that it is
available at runtime with plain golang.__version__ .

A comma was missing - this way last element was 'funcgimport', not
'func' and 'gimport', and so `from golang import *` was failing.

	.../pygolang/golang/strconv_test.py:75: PytestDeprecationWarning: raises(..., 'code(as_a_string)') is deprecated, use the context manager form or use `exec()` directly
	...

See https://docs.pytest.org/en/latest/deprecations.html#raises-warns-with-a-string-as-the-second-argument

Until now it was not included by `python setup.py sdist`.

On this topic, also include README.rst explicitly, since even if it is
currently implicitly included, that might change some day and break us.

This are functions to decode quotation that was produced by
strconv.quote().

Move quote implementation from gcompat to strconv. Make quote work on both
unicode|bytes string input and produce the same output type. Preserve qq
(still remaining in gcompat) to always produce str, since that is used
in prints.

Stop hiding qq and benchmarking bits in changelog only - add their
description into additional sections in readme.

Since py.bench description is now in readme, there is no need to keep
py.bench example in the changelog.

All tests pass.

`defer` allows to schedule a cleanup to be executed when current function
returns. It is similar to `try`/`finally` but does not force the cleanup part
to be far away in the end. For example::

   wc = wcfs.join(zurl)    │     wc = wcfs.join(zurl)
   defer(wc.close)         │     try:
                           │        ...
   ...                     │        ...
   ...                     │        ...
   ...                     │     finally:
                           │        wc.close()

For completeness there is `recover` and `panic` that allow to program with
Go-style error handling, for example::

   def _():
      r = recover()
      if r is not None:
         print("recovered. error was: %s" % (r,))
   defer(_)

   ...

   panic("aaa")

But `recover` and `panic` are probably of less utility since they can be
practically natively modelled with `try`/`except`.

If `defer` is used, the function that uses it must be wrapped with `@func` or
`@method` decorators.

The implementation is partly inspired by work of Denis Kolodin:

- https://habr.com/post/191786
- https://stackoverflow.com/a/43028386/9456786

While the second phase of select is running we queue send/recv cases to
corresponding channels. At some point - when some of the cases are
already queued - a peer goroutine might try to send/recv on that
channel. And it will succeed because a waiter was queued to the channel.

At the same time select is continuing its enqueue loop and before enqueuing
to a channel it tries to send/recv there. If that channel became just ready
(i.e. just after select poll phase) the try to send/recv will succeed. This
means that actually 2 select cases could be executed at the same time.

Fix it by carefully checking whether some case already won before trying
to send/recv on a channel.

This fixes the test failures that were demonstrated by previous 2 patches.

Previous select-select test uses 2 channels and recreats them every
iteration. However the bug we just saw in b51b8d5d (select: Run tests
more thoroughly) is of parasitic nature - where misbehaviour depends on
what state has been left there from previous select and whether the race
to get to that state fast enough succeeded.

So add a more secialized test which tries to trigger the parasitic
effects that depend on previous state left by select on the channels.

The test, similarly to b51b8d5d, currently fails.

With more iterations it triggers a bug:

	kirr@deco:~/src/tools/go/pygolang$ py.test -vsx
	============================================== test session starts ===============================================
	platform linux2 -- Python 2.7.15+, pytest-3.6.4, py-1.6.0, pluggy-0.6.0 -- /usr/bin/python2
	cachedir: .pytest_cache
	rootdir: /home/kirr/src/tools/go/pygolang, inifile:
	collected 7 items

	golang/_gopath_test.py::test_import_module PASSED
	golang/_gopath_test.py::test_import_package PASSED
	golang/gcompat_test.py::test_qq PASSED
	golang/golang_test.py::test_go PASSED
	golang/golang_test.py::test_chan PASSED
	golang/golang_test.py::test_select Exception in thread Thread-117:
	Traceback (most recent call last):
	  File "/usr/lib/python2.7/threading.py", line 801, in __bootstrap_inner
	    self.run()
	  File "/usr/lib/python2.7/threading.py", line 754, in run
	    self.__target(*self.__args, **self.__kwargs)
	  File "/home/kirr/src/tools/go/pygolang/golang/golang_test.py", line 320, in _
	    assert (_, _rx) == (1, 'b')
	AssertionError: assert (1, 'xxx1') == (1, 'b')
	  At index 1 diff: 'xxx1' != 'b'
	  Full diff:
	  - (1, 'xxx1')
	  + (1, 'b')

This test was failing when pygolang was not dev-mode installed (`pip install -e .`):

	kirr@deco:~/src/tools/go/pygolang$ py.test
	=========================================== test session starts ============================================
	platform linux2 -- Python 2.7.15+, pytest-3.6.4, py-1.6.0, pluggy-0.6.0
	rootdir: /home/kirr/src/tools/go/pygolang, inifile:
	collected 8 items

	golang/_gopath_test.py ..                                                                            [ 25%]
	golang/gcompat_test.py .                                                                             [ 37%]
	golang/golang_test.py F....                                                                          [100%]

	================================================= FAILURES =================================================
	_________________________________________________ test_go __________________________________________________

	    def test_go():
	        # leaked goroutine behaviour check: done in separate process because we need
	        # to test process termination exit there.
	        subprocess.check_call([sys.executable,
	>           dirname(__file__) + "/golang_test_goleaked.py"])

	golang/golang_test.py:38:
	_ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _

	popenargs = (['/usr/bin/python2', '/home/kirr/src/tools/go/pygolang/golang/golang_test_goleaked.py'],)
	kwargs = {}, retcode = 1
	cmd = ['/usr/bin/python2', '/home/kirr/src/tools/go/pygolang/golang/golang_test_goleaked.py']

	    def check_call(*popenargs, **kwargs):
	        """Run command with arguments.  Wait for command to complete.  If
	        the exit code was zero then return, otherwise raise
	        CalledProcessError.  The CalledProcessError object will have the
	        return code in the returncode attribute.

	        The arguments are the same as for the Popen constructor.  Example:

	        check_call(["ls", "-l"])
	        """
	        retcode = call(*popenargs, **kwargs)
	        if retcode:
	            cmd = kwargs.get("args")
	            if cmd is None:
	                cmd = popenargs[0]
	>           raise CalledProcessError(retcode, cmd)
	E           CalledProcessError: Command '['/usr/bin/python2', '/home/kirr/src/tools/go/pygolang/golang/golang_test_goleaked.py']' returned non-zero exit status 1

	/usr/lib/python2.7/subprocess.py:190: CalledProcessError
	------------------------------------------- Captured stderr call -------------------------------------------
	Traceback (most recent call last):
	  File "/home/kirr/src/tools/go/pygolang/golang/golang_test_goleaked.py", line 23, in <module>
	    from golang import go, chan
	ImportError: No module named golang
	==================================== 1 failed, 7 passed in 0.10 seconds ====================================

Fix it by injecting top-level pygolang/ into $PYTHONPATH when testing
via external processes.

Fixes 69cef96e (go: Don't allow leaked goroutines to prevent program to exit)

There was no test for @method so far and that's why it went unnoticed.
But on Python3 it breaks on f.func_name:

	In [3]: def f(): pass

	In [4]: f.func_name
	---------------------------------------------------------------------------
	AttributeError                            Traceback (most recent call last)
	<ipython-input-4-662dcbac1531> in <module>()
	----> 1 f.func_name

	AttributeError: 'function' object has no attribute 'func_name'

Fix it by using f.__name__ which works on both py2 and py3.

Add test for @method to make sure it doesn't break unnoticed.

This is the Go behaviour, as demonstratd by the following program:

---- 8< ----
package main

import (
	"fmt"
	"time"
)

func work(w int) {
	for i := 0; ; i++ {
		fmt.Printf("w%d: %d\n", w, i)
		time.Sleep(1*time.Second)
	}
}

func main() {
	for i := 0; i < 100; i++ {
		go work(i)
	}

	time.Sleep(3*time.Second)
	println("main: exit")
}
---- 8< ----

+ corresponding bits in golang.testing package.

py.bench description follows:

"""
py.bench, similarly to py.test, discovers bench_* functions and Bench* classes
and then runs each discovered benchmark via golang.testing.benchmark. Similarly
to `go test -bench`, benchmarking results are printed in Go benchmark format.

For example, running py.bench on the following code::

    def bench_add(b):
        x, y = 1, 2
        for i in xrange(b.N):
            x + y

gives something like::

    $ py.bench --count=3 x.py
    ...
    pymod: bench_add.py
    Benchmarkadd    50000000        0.020 µs/op
    Benchmarkadd    50000000        0.020 µs/op
    Benchmarkadd    50000000        0.020 µs/op
"""

The implementation is based on t/py.bench from Wendelin.core - see
following commits for its history:

lab.nexedi.com/nexedi/wendelin.core/commit/51f252d4
lab.nexedi.com/nexedi/wendelin.core/commit/074ce24d
lab.nexedi.com/nexedi/wendelin.core/commit/ed13c3f9
lab.nexedi.com/nexedi/wendelin.core/commit/fc08766d
lab.nexedi.com/nexedi/wendelin.core/commit/5a1ed45a
lab.nexedi.com/nexedi/wendelin.core/commit/bcab1246

Python3 complains that there is no .im_func & friends of a method.
Workaround it all with compat code.