`--fixtures` and `--fixtures-per-test` exit with code 0 on collection errors, internal errors and `pytest.exit()`

[lpyu001]

Description

pytest --fixtures and pytest --fixtures-per-test exit with status 0 when collection fails. A plain pytest run or pytest --collect-only on the same file exits with 2. Both options also exit with 0 when a hook raises an internal error (normally 3) and when a plugin calls pytest.exit() with a non-zero returncode.

The exit code should match what pytest returns for the same failure without these options: 2 (ExitCode.INTERRUPTED) for collection errors, or 1 (ExitCode.TESTS_FAILED) with --continue-on-collection-errors; 3 (ExitCode.INTERNAL_ERROR) for internal errors; and the returncode passed to pytest.exit(). Scripts and CI jobs that run these options currently see success after pytest has reported an error.

Exit codes observed:

Scenario	no option	--collect-only	--continue-on-collection-errors	--fixtures	--fixtures-per-test
Collection error (example 1)	2	2	1	0	0
RuntimeError in hook, INTERNALERROR (example 2)	3	—	—	0	0
pytest.exit("stop", returncode=7) in hook (example 3)	7	—	—	0	0

Two separate problems produce this.

1. The session exit status is discarded. pytest_cmdline_main() in src/_pytest/fixtures.py (lines 1675–1682) ignores the return value of showfixtures() and show_fixtures_per_test():

def pytest_cmdline_main(config: Config) -> int | ExitCode | None:
    if config.option.showfixtures:
        showfixtures(config)
        return 0
    if config.option.show_fixtures_per_test:
        show_fixtures_per_test(config)
        return 0
    return None

Both functions return the result of wrap_session() (src/_pytest/main.py, line 320), which sets session.exitstatus when an exception reaches it: INTERNAL_ERROR for unexpected exceptions, and the given returncode for pytest.exit(). This accounts for examples 2 and 3.

2. Collection errors never produce a non-zero status. A collection error does not raise. Session.perform_collect() only increments session.testsfailed. In a normal run, pytest_runtestloop() (src/_pytest/main.py, lines 400–404) turns that count into an interruption:

def pytest_runtestloop(session: Session) -> bool:
    if session.testsfailed and not session.config.option.continue_on_collection_errors:
        raise session.Interrupted(
            f"{session.testsfailed} error{'s' if session.testsfailed != 1 else ''} during collection"
        )

_showfixtures_main() (line 2505) and _show_fixtures_per_test() (line 2450) in src/_pytest/fixtures.py call session.perform_collect() directly, never reach pytest_runtestloop, and never check session.testsfailed. They return None, so wrap_session() stores ExitCode.OK via session.exitstatus = doit(config, session) or 0 (src/_pytest/main.py, line 333).

Fixing problem 1 alone is not enough. With only pytest_cmdline_main() changed to return the wrap_session() result, examples 2 and 3 exit with the correct codes, but example 1 still exits with 0 for both options.

pip list

Package   Version
--------- -----------------------
iniconfig 2.3.0
packaging 26.3
pip       26.2.1
pluggy    1.6.0
Pygments  2.21.0
pytest    9.2.0.dev335+g99ab2accc

pytest and operating system versions

• pytest 9.2.0.dev335+g99ab2accc (installed from main at commit 99ab2accc)
• Python 3.12.3
• Ubuntu 24.04.4 LTS, Linux 6.8.0-124-generic x86_64

Minimal example

Example 1: collection error

# test_bad.py
def test_broken(:
    pass

$ python -m pytest --fixtures test_bad.py; echo "exit=$?"
============================= test session starts ==============================
platform linux -- Python 3.12.3, pytest-9.2.0.dev335+g99ab2accc, pluggy-1.6.0
rootdir: /tmp/repro/bad
collected 0 items / 1 error
cache -- .../_pytest/cacheprovider.py:581
    Return a cache object that can persist state between testing sessions.

[... remaining fixture list omitted ...]

==================================== ERRORS ====================================
_________________________ ERROR collecting test_bad.py _________________________
[... traceback omitted ...]
E     File "/tmp/repro/bad/test_bad.py", line 1
E       def test_broken(:
E                       ^
E   SyntaxError: invalid syntax
=========================== short test summary info ============================
ERROR test_bad.py
=============================== 1 error in 0.09s ===============================
exit=0

The same input without --fixtures:

$ python -m pytest test_bad.py; echo "exit=$?"
...
!!!!!!!!!!!!!!!!!!!! Interrupted: 1 error during collection !!!!!!!!!!!!!!!!!!!!
=============================== 1 error in 0.16s ===============================
exit=2

python -m pytest --fixtures-per-test test_bad.py also exits with 0.

Example 2: internal error (separate empty directory)

# conftest.py
def pytest_collection_modifyitems(items):
    raise RuntimeError("boom")

# test_ok.py
def test_ok():
    pass

$ python -m pytest --fixtures; echo "exit=$?"
...
INTERNALERROR>     raise RuntimeError("boom")
INTERNALERROR> RuntimeError: boom
============================ no tests ran in 0.00s =============================
exit=0

Without --fixtures, the exit code is 3.

Example 3: pytest.exit() with a return code (separate empty directory, same test_ok.py)

# conftest.py
import pytest


def pytest_collection_modifyitems(items):
    pytest.exit("stop", returncode=7)

$ python -m pytest --fixtures; echo "exit=$?"
...
collected 1 item
============================ no tests ran in 0.08s =============================
!!!!!!!!!!!!!!!!!!!!!!!!! _pytest.outcomes.Exit: stop !!!!!!!!!!!!!!!!!!!!!!!!!!
exit=0

Without --fixtures, the exit code is 7.

[bluetech]

The request makes sense to me, I don't see a reason why these commands would return 0 if they weren't able to perform their function.

I think it'd be safest to treat this as a breaking change, so let's schedule it for pytest 10.0.