junit xml "tests" number seems wrong if only teardown fails

[julianheinzelmann]

Hi all,
I am using pytest 3.7.1 to run some tests and output the results using the --junitxml=file.xml argument.
Next the generated xml files are parsed by xunitparser.

I noticed, that for some testcases I run, the parsing failes, as the parser checks if the number of test cases in the report matches the tests="..." information in the header of the xml file.

The following sample test module shows the problem:
(had to rename .py to .txt)
SampletestCase2.txt

run with
python -m pytest SampletestCase2.py --junitxml=fail.xml

now the xml file looks as follows:
fail.txt

There are two <testcase>...</testcase> sections, but the header says: tests="4"
This causes an failed assertion in xunitparser

Therefore: If my asumption is correct, that the tests count in the header of the xml file should match the number of testcase sections in the xml file, then pytest outputs the wrong tests number in case ONLY the teardown failes.

I found a quick workaround (please ignore in case the above assumption is wrong and the xml format is valid, but the error is in the parser)
As i am very new to pytest internals I do not know the possible side effects, but seems to fix my problem...

In the pytest_runtest_logreport function in the file junitxml.py of pytest:

Modifying the increment of the self.cnt_double_fail_tests parameter by introducing the bool incDoubleFail, which is used to determine if there was a true double failure or not. Just at the end of the function the final decision is taken, whether the counter should be incremented.

modified_pytest_runtest_logreport.txt