coverage-7.3.3 broke pragma: no cover in match-case statement

[tdivis]

coverage-7.3.3 released few hours ago broke ignoring whole function by # pragma: no cover when the function contains match..case statement with case condition on multiple lines. My minimum example:

# code.py:
def hello():
    print("Hello world")


def hello2(a):  # pragma: no cover
    match a:
        case ("hello one"
              | "hello two"):
            print("hello1-2")
        case "hello three":
            print("hello3")

# test_code.py:
from code import hello
from unittest import TestCase


class HelloTest(TestCase):
    def test_hello(self):
        hello()

Then run coverage and you'll get report with last two match-case lines uncovered even though the whole function should be ignored.

coverage erase && coverage run --parallel-mode --source=code --branch -m unittest test_code.py && coverage combine && coverage report --show-missing
Hello world
.
----------------------------------------------------------------------
Ran 1 test in 0.000s

OK
Combined data file .coverage.glinall.600.XyruySkx
Name      Stmts   Miss Branch BrPart  Cover   Missing
-----------------------------------------------------
code.py       4      2      2      0    33%   10-11
-----------------------------------------------------
TOTAL         4      2      2      0    33%

Worked fine in coverage-7.3.2.

[nedbat]

Here is a test for the test suite that demonstrates the problem:

@ tests/test_coverage.py:1743 @ def my_func_2(super_long_input_argument_0=0, super_long_input_argument_1=1, supe
               [], "5", excludes=['my_func']
           )

  +    def test_excluding_clause_bug1713(self) -> None:
  +        self.check_coverage("""\
  +            print("1")
  +
  +            def hello_3(a):  # no thanks
  +                if ("4" or
  +                    "5"):
  +                    print("6")
  +                else:
  +                    print("8")
  +
  +            print("10")
  +            """,
  +            [1, 10], "", excludes=["no thanks"],
  +        )
  +
       def test_excluding_method(self) -> None:
           self.check_coverage("""\
               class Fooey:

@kajakaj, @maciekgyver, do you have an idea how to handle this case?

[sigma67]

Not sure if related, but I also have a new coverage miss on coverage 7.3.3

Code:

if condition_a:  # pragma: no cover; local only
  func1()
  if (
    condition_b
    == 0
  ):
    return
  func2()  # this statement is now shown as not covered

On prior versions, the whole if block was ignored correctly.

[dhuang]

We have seen this regression for a class as well, so seems like anything block level.

class Foo:  # pragma: no cover 
    def greet(self):
        print("hello world")

[nedbat]

Hi all, sorry for the regression. I'm working on a fix. The code uses tokenize to analyze the code, which isn't holding up well. I'm working on switching it to use the AST instead.

[freakboy3742]

Not sure this is exactly the same problem, but in case it helps, we're also seeing a regression in the Briefcase test suite. However, we're only seeing it on Windows. I'll try to reduce this to a simple reproduction case.

[freakboy3742]

I haven't narrowed this down to a specific reproduction case yet, but it appears to be related to an interaction with coverage-conditional-plugin. This is the coverage file for the failing run; it has 100% coverage on coverage 7.3.2, but has the 2 missing blocks of lines on coverage 7.3.3, (conditional-coverage-plugin=0.9.0), with a conditional rule of no-cover-if-is-windows = True defined.

coverage-7.3.2.zip

[ringohoffman]

I am also seeing a regression in coverage (only in CI though?) for a case like:

class MyClass:  # pragma: no cover
    def method(self):
        pass

Which was previously ignoring the lines in method(), but is now failing for missing coverage.

[freakboy3742]

FWIW: The pattern that @ringohoffman has described is very similar to what is happening in my case; however, I haven't been able to generate a simple reproduction case that fails.

In case it helps someone else: in my case, the no cover is conditional on a sys.platform check. The failure only occurs on Windows; the coverage reports missing lines on Windows, as well as when the Windows report is combined with a Linux and macOS report (Linux and macOS both do have coverage of the given lines).

The code that is being mistakenly marked as a coverage miss is:

• a method in a class where the entire class is marked no-cover
• 2 context manager blocks one after another inside a method of a class, where the method is marked no-cover.

The second one is particularly odd, because there's plenty of other lines in the method (both before and after the problematic context managers) that are marked as covered, as well as other context managers in the same method.