This reverts commit 469f21a9.

Even though Debian stopped shipping Python3.6 I'm now building it myself
to have test coverage for two latest Python releases.

Implement deadlines / timeouts using timers added recently in 9c260fde
(time: New package that mirrors Go's time).

In the next patch we'll be adding deadlines support with another test.
Move common infrastructure that will be used in all context tests to be
test-module global.

We'll need to mark contexts that miss their deadline with
deadlineExceeded error in a follow-up patch.

In this patch: transform _BaseCtx._cancel to accept an error with which
a context has to be canceled, but for now always use canceled in places
where _cancel is called.

Same story as in Go, but we were lacking documentation notice about this.

Add time.Timer, time.Ticker and convenience functions time.tick,
time.after and time.after_func. These will be used in context to support
deadlines and timeouts.

While at time topic, also provide sleep and now from golang.time, so
that there is no need to import both golang.time and stdlib's time in a
file.

Provide time constants in the module as they are useful to have and
mirror constants provided by Go's time.

Note: timers implementation is very suboptimal for now.

Python adds dirname of run program to sys.path . This way when
golang_test_goleaked.py runs it can import modules located under golang/
just by their name. Until now this was not noticed, but in the next
patch we are going to add golang.time module and if test program is run
with golang/ in sys.path just plain `import time` won't import time from
stdlib and instead import time from golang/ .

Such behaviour can be mitigated by doing `from __future__ import
absolute_import` and we do that, including in golang_test_goleaked.py
(see 81dfefa0 "*: __future__ += absolute_imports; Use unified __future__
everywhere"). However that does not prevent modules - even modules from
stdlib - who are doing `import time` and not doing future
absolute_import to import golang's time instead of stdlib. For example
on PyPy2 threading imports time and then the test breaks:

  Traceback (most recent call last):
    File "/home/kirr/src/tools/go/pygolang/.tox/pypy-thread/site-packages/golang/golang_test_goleaked.py", line 24, in <module>
      from golang import go, chan
    File "/home/kirr/src/tools/go/pygolang/.tox/pypy-thread/site-packages/golang/__init__.py", line 38, in <module>
      import inspect, threading, collections, random, sys
    File "/usr/lib/pypy/lib-python/2.7/threading.py", line 15, in <module>
      from time import time as _time, sleep as _sleep
    File "/home/kirr/src/tools/go/pygolang/.tox/pypy-thread/site-packages/golang/time.py", line 30, in <module>
      from golang import go, chan, select, default, nilchan, panic
  ImportError: cannot import name 'go'

-> Move the test program into a directory different from golang/ to
avoid this trap.

- we are going to introduce golang.time, and from inside there without
  `from __future__ import absolute_imports` it won't be possible to import
  needed stdlib's time.

- we were already doing `from __future__ import print_function`, but
  only in some files.

-> It makes sense to apply updated __future__ usage uniformly.

- correct CHANGELOG title levels: since readme uses "----" for the
  first level and "~~~~" for the second level, "====" was interpreted as
  the third title level and "Pygolang change history" became
  sub-sub-section of "Additional packages and utilities" with section
  for each version conversely becoming first level. It was not very
  noticeable until 0c5f9d06 (readme: Push "Additional packages and
  utilities" into its own section) started to use "~~~~".
  -> Rework CHANGELOG titling to align with the one in README.

- fix minor markup bits in README.

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 kirr/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.