[TIP] CTracer vs PyTracer in coverage.py 4.2; different behavior/caching?

[R M Waters]

Hi,

I am attempting to run individual coverage reports for tests in a large
suite in order to run some analysis on the suite.  I am using nose as the
test runner, and executing the `erase()` and `start()` methods of
`coverage.Coverage` between each test invocation in order to reset the
coverage data.

Some of our test methods follow the same branch pattern, but with different
data (consider testing an encode/decode function with varying inputs).

Using PyTracer, I can retrieve the coverage data for each test via
`collector.data`.  This includes tests that have the same branch pattern
(it will return same coverage data for those tests).

However, using CTracer, I can retrieve coverage data only for the first
invocation of a particular branch pattern.  Subsequent invocations return
an empty dict.

I have even tried replacing the entire `coverage.Coverage` object for each
test, which should guarantee that all of the data and trace cache on the
python side are purged, no dice though. Is CTracer a singleton?

I can understand if this is an optimization; when running a coverage report
for a test suite, this is all useless data.

Is this expected behavior?  Am I going to have to go out of process, or is
there an easier solution?

Thank you so much,
Robert Waters
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.idyll.org/pipermail/testing-in-python/attachments/20161113/aa756642/attachment.htm>