GitLab

Rework overview page in restructured text to fix the PDF output and add some more links to various parts of the documentation.

Repairs #85. These circular references could lead to memory leaking
under certain circumstances when using the AnyOf condition.

The repair is to remove the _check() callbacks from the Condition's
events once the condition has triggered.

Merged in jcbsnd/simpy/jcbsnd/fix-documentation-for-resourceget-resour-1465832996081 (pull request #87)

Fix documentation for resource.Get()

Added support for nested preemption

Fix Python optimize

Add performance benchmark tests

Use properties for Event.ok and Event.defused

This brings the semantic of the Event.ok property in-line with the
[recently] previous semantic of the Event.ok attribute.

Testing self._ok was unnecessary.

The benchmarks may be re-enabled on the command line with:

    py.test -m benchmark

Or to run all tests including benchmark and non-benchmark tests:

    py.test -m "benchmark or not benchmark"

The benchmark tests are also disabled by default when running the test
suite with tox, however new tox environment called "benchmark" exists
for running the benchmarks. To the benchmark tests with tox:

    tox -e benchmark

Add new PriorityStore resource

Various minor repairs for flake8 compliance.

The new `PriorityStore` class is a subclass of `Store` that maintains
its `items` list in priority order such that `get()` yields items in
priority order.

The `PriorityItem` helper class allows unorderable items to be put into
a `PriorityStore` by associating an explicit priority value with the
item.

New `Event.ok` and `Event.defused` properties wrap `Event._ok` and
`Event._defused` attributes. The key benefit is that doc strings now
can exist for `Event.ok` and `Event.defused`.

This is a performance-neutral change since all of the fast-paths still
use direct access to the `_ok` and `_defused` attributes.

Update testing recommendations

The new `tests/test_benchmark.py` adds several performance benchmarks to
the test suite. It uses the `pytest-benchmark` plugin for `pytest`.

The benchmarks cover high-frequency "micro" behaviors such as measuring
`Event()` initialization time as well as full simulations involving
resources and multiple processes.

The new benchmark tests are run automatically by `py.test`. The
benchmarks add approximately 10 seconds to the test suite runtime.
However, the time-consuming benchmarks can be disabled with:

    $ py.test --benchmark-disable

Alternatively, to only run the benchmarks:

    $ py.test --benchmark-only

These benchmarks have proven sufficient to detect performance deltas due
to rather small changes in the code. For example, adding initialization
for two attributes of `Event` caused a detectable change in performance
in most of these benchmarks.

The paths in `.coveragerc` are updated to support the new `src` and
`tests` directory structure.

Repair typo in example code: "charing" to "charging".

Update documentation in README.txt and installation.rst to recommend
running the test suite with `py.test` instead of the `python -c ...`
method. The `py.test` method is shorter and more powerful.

Plus if tests are invoked with the `py.test` method, then we no longer
need simpy.util.test(). So we also remove that function in the name of
PEP-20 (one obvious way to do it).

Also apply flake8 to entire project in tox.ini.