Coveralls uses internal undocumented parts of coverage.py

[nedbat]

I'm heavily refactoring coverage.py for version 4.0. I tried to release an alpha, but it broke coveralls, and therefore many people's tests on Travis:

Submitting coverage to coveralls.io...
Traceback (most recent call last):
  File ".tox/coveralls/bin/coveralls", line 11, in <module>
    sys.exit(main())
  File "/home/travis/build/ionelmc/python-manhole/.tox/coveralls/lib/python2.7/site-packages/coveralls/cli.py", line 49, in main
    result = coverallz.wear()
  File "/home/travis/build/ionelmc/python-manhole/.tox/coveralls/lib/python2.7/site-packages/coveralls/api.py", line 73, in wear
    data = self.create_data()
  File "/home/travis/build/ionelmc/python-manhole/.tox/coveralls/lib/python2.7/site-packages/coveralls/api.py", line 122, in create_data
    self._data = {'source_files': self.get_coverage()}
  File "/home/travis/build/ionelmc/python-manhole/.tox/coveralls/lib/python2.7/site-packages/coveralls/api.py", line 132, in get_coverage
    return reporter.report()
  File "/home/travis/build/ionelmc/python-manhole/.tox/coveralls/lib/python2.7/site-packages/coveralls/reporter.py", line 27, in report
    self.parse_file(cu, self.coverage._analyze(cu))
  File "/home/travis/build/ionelmc/python-manhole/.tox/coveralls/lib/python2.7/site-packages/coveralls/reporter.py", line 58, in parse_file
    source_file = cu.source_file()
AttributeError: 'PythonCodeUnit' object has no attribute 'source_file'

We should work together on how to fix it.

[coagulant]

I guess it's not too hard to make coveralls work with coverage 4.0b. However, adding newline to the end of python source file changes results a bit, compared to previous coverage version. Why is newline necessary?

[nedbat]

Thanks for looking into updating the support. 4.0 is not in beta yet, so the interfaces still are not finished, so it may be early to write code.

The newline is only necessary in some places (compiling to bytecode), and I could adjust how it is done, but can you tell me what kind of result changes you are seeing? I don't worry if the result numbers shift up or down a little bit.

[coagulant]

I'm using a custom reporter to get both coverage data and source file at the same time.
Here is a working report function, adapted to 4.0a2

def parse_file(self, cu, analysis):
        """ Generate data for single file """
        filename = cu.file_locator.relative_filename(cu.filename)
        coverage_lines = [self.get_hits(i, analysis) for i in range(1, len(cu.parser.lines) + 1)]
        source = cu.source()

        self.source_files.append({
            'name': filename,
            'source': source,
            'coverage': coverage_lines
        })

As coverage.py adds an empty newline to the end of the file, I'm seeing this extra line in sources and in coverage output. It's a minor discrepancy, but still...

Example: running coverage example/runtests.py results (full pytest output)

Coverage 3.X:

hits per line: [None, None, None, 1, 1, None, None, 1, None, None, None, 1, 0]
number of source lines in project.py: 13
json-encoded source as string: u'# coding: utf-8\n\n\ndef hello():\n    print(\'world\')\n\n\nclass Foo(object):\n    """ Bar """\n\n\ndef baz():\n    print(\'this is not tested\')'

Coverage 4.0a2:

hits per line: [None, None, None, 1, 1, None, None, 1, None, None, None, 1, 0, None]
number of source lines in project.py: 14
json-encoded source as string: u'# coding: utf-8\n\n\ndef hello():\n    print(\'world\')\n\n\nclass Foo(object):\n    """ Bar """\n\n\ndef baz():\n    print(\'this is not tested\')\n'

I don't think that adding one extra newline changes coverage numbers much, just wanted to let you know.

[nedbat]

Thanks for the explanation.

[coagulant]

Closing this in favor of #81