"ERROR at" test case name headings ambiguous if they use the same test case function name

[cswrd]

First of all, thanks for providing and maintaining pytest!

Given a minimal setup like this:

# conftest.py
import pytest

@pytest.fixture
def fixture():
    raise Exception()

# test_1.py and test_2.py
def test(fixture):
    pass

The "ERROR at" section being printed is ambiguous. You can't tell which section is referring to which test case. Both say ERROR at setup of test.

================================ ERRORS =================================
________________________ ERROR at setup of test _________________________

    @pytest.fixture
    def fixture():
>       raise Exception()
E       Exception

conftest.py:5: Exception
________________________ ERROR at setup of test _________________________

    @pytest.fixture
    def fixture():
>       raise Exception()
E       Exception

conftest.py:5: Exception

Given the short test summary section, sometimes it's possible to "reverse engineer" which section is referring to which test case especially if they were caused by different exception types and/or lines of code causing the issue. It requires scrolling back and forth, tough.

The "FAILURES" section uses the same headings. At least there's 1 line of code within the stack trace revealing the originally affected test case usually, though.

Afaik there is neither a necessity nor a recommendation to make test case function names unique. So I wonder, why it's not printing its full name by default for these sections.

[cfcromn]

Hi! I’d be interested in working on this.

Would using the test’s nodeid (for example, test_1.py::test) in the error/failure heading be the preferred approach here, or would you prefer only adding enough path information to disambiguate tests with the same name?

If the nodeid approach sounds good, I’d be happy to put together a PR with tests.

[siddy-stack]

I would like to resolve this issue. kindly assign

[ManasSakthivel]

Hi 'd like to pick up this issue and provide a robust fix.
Regarding the design question raised by Ronny, using an unconditional full node ID (or a clean collision-detection approach) is ideal to maintain stable CI output across sharding, -x, and xdist.
I am putting together a clean patch that addresses all the points from the previous attempt:
Fixing ANSI escape leakage so write_sep length calculations and coloring remain correct.
Preserving custom subtest headers (head_line) to prevent regressions.
Covering all three call sites (FAILURES, ERRORS, and PASSES/XPASSES) while correctly accounting for setup/teardown/collect prefixes.
Adding strict regression tests with exact string assertions and a --color=yes test case, alongside a properly formatted towncrier changelog fragment.
I'll have a clean PR up shortly!

[ManasSakthivel]

I've opened a PR with a complete fix for this issue. Here's a summary of how it addresses the bug and incorporates the feedback from Ronny's review of the previous attempt:
1]All Call Sites Covered: Disambiguation is applied uniformly across summary_failures_combined (FAILURES/XFAILURES), summary_errors (setup/teardown/collect errors), and summary_passes_combined (PASSES/XPASSES) — ensuring no section is left out.
2]No ANSI Escape Leakage: The [nodeid] suffix is appended as a clean, plain-ASCII string after any prefix is built and before write_sep() is called. TerminalWriter.sep never sees raw escape bytes, so separator fill widths remain exact (verified with a --color=yes test).
3]Subtest Regression Avoided: _getfailureheadline is left completely untouched. It continues to return rep.head_line verbatim, so SubtestReport's custom head_line (which appends subtest descriptions) is fully preserved.
4]Correct Phase Prefixes: The session-wide duplicate counter keys on the fully-prefixed heading string (e.g., distinguishing ERROR at setup of test from ERROR at teardown of test) so phases are never incorrectly conflated.
5]Unchanged for Unique Names: The [nodeid] suffix is only appended when a heading appears more than once across the session, keeping output terse in the common case.
6]Strict Tests & Changelog: Added 8 comprehensive regression tests checking exact string formatting, colors, and all report sections, alongside a properly formatted towncrier fragment.