[RFC] structured NamedTuples for code locations by RonnyPfannschmidt · Pull Request #8056 · pytest-dev/pytest

RonnyPfannschmidt · 2020-11-21T13:20:38Z

Introduces CodeLocation and ItemLocation NamedTuples to replace the unstructured strings and bare tuples currently used for code locations. Both types store a 0-based lineindex field and expose a 1-based .lineno property, making the convention explicit rather than something each call site has to remember.

`CodeLocation` (replaces `str` from `getlocation()`)

Two fields: path: Path, lineindex: int. Used for fixture locations in error messages and --fixtures / --fixtures-per-test output. __str__ produces path:lineno (1-based).

getlocation() previously stored co_firstlineno + 1 — an off-by-one since co_firstlineno is already 1-based. Now stores co_firstlineno - 1 (0-based) and lets the .lineno property add +1 back correctly.

`ItemLocation` (replaces `tuple[str, int | None, str]` from `Item.location`)

Three fields: path: str, lineindex: int | None, testname: str. Returned by Item.location, stored on TestReport. Tuple subclass — all existing indexing, unpacking, and equality comparisons continue to work unchanged.

Other changes

write_item in fixtures.py now uses item.location instead of getlocation(item.function, ...), removing the item.function assumption that fails for DoctestItem and custom Item subclasses.
BaseReport.location / TestReport.location typed as ItemLocation. TestReport.__init__ coerces plain tuples for backwards compatibility.
_report_kwargs_from_json reconstructs ItemLocation after JSON round-trip.
Hook signatures, reportinfo() contract, and JSON wire format are all unchanged.

nicoddemus

Looks good so far!

So the idea is to not change Item.location's return value or any other exposed value, just use CodeLocation internally?

nicoddemus · 2020-11-21T14:12:36Z

    lineno = function.__code__.co_firstlineno
+
+    # TODO: this cycle indicates a larger issue
+    from .pathlib import bestrelpath


Hmm indeed, perhaps getlocation can be moved to pathlib then? Makes sense there, is all about obtaining the "path" location to a function.

nicoddemus · 2020-11-21T14:23:35Z

            tw.write(" [%s scope]" % fixturedef.scope, cyan=True)
        if verbose > 0:
-            tw.write(" -- %s" % bestrel, yellow=True)
+            tw.write(" -- %s" % str(bestrel), yellow=True)


Is the str redudant?

since namedtuple isinstance tuple, %-formatting will try and unpack it. if this converts to f-string you won't need str(...)

Good execuse to switch to an fstring :)

asottile

looks good!

asottile · 2020-11-21T17:03:47Z

            tw.write(" [%s scope]" % fixturedef.scope, cyan=True)
        if verbose > 0:
-            tw.write(" -- %s" % bestrel, yellow=True)
+            tw.write(" -- %s" % str(bestrel), yellow=True)


since namedtuple isinstance tuple, %-formatting will try and unpack it. if this converts to f-string you won't need str(...)

bluetech

Nice cleanup!

bluetech · 2020-11-21T18:49:50Z


 import py

-from _pytest.compat import assert_never


What is the reason for this change?

import cycle maangement

I thought

from _pytest.compat import assert_never

and

import _pytest.compat

are equivalent from an import-cycle perspective. Maybe not...

the latter delays the access of the module, allowing ~some cycles to occur (as long as there's no attribute access the cycle will import a "partially initialized module" which gets finalized later)

Interesting, didn't know about this partially initialized state.

bluetech · 2020-11-21T18:57:09Z

    tw = _pytest.config.create_terminal_writer(config)
    verbose = config.getvalue("verbose")

    def get_best_relpath(func):


IMO we can just inline this function now.

bluetech · 2020-11-21T18:58:27Z

            tw.write(" [%s scope]" % fixturedef.scope, cyan=True)
        if verbose > 0:
-            tw.write(" -- %s" % bestrel, yellow=True)
+            tw.write(" -- %s" % str(bestrel), yellow=True)


Good execuse to switch to an fstring :)

RonnyPfannschmidt · 2020-11-22T19:16:38Z

@nicoddemus for practical reasons we have to deprecate node.location and reportinfo() - its a hard to fix api

adding a qualified name to the code location is no longer needed/sensible on modern python, where the qualified name is available

pytest itself computes it by a listchain from hell

i pushed my current state as im currently error blind on location determination wrt allowing/disallowing escape, the testsuite should be made to pass

bluetech · 2020-11-28T19:14:09Z


 import py

-from _pytest.compat import assert_never


Interesting, didn't know about this partially initialized state.

Co-authored-by: Cursor <cursoragent@cursor.com>

…0-based - CodeLocation: rename field `lineno` to `lineindex` (0-based stored), add `.lineno` property (1-based) and `__str__` using 1-based display. `getlocation()` now stores `co_firstlineno - 1` as lineindex. - ItemLocation: new NamedTuple (path, lineindex, testname) that replaces the bare `tuple[str, int | None, str]` returned by `Item.location`. Has `.lineno` property (1-based) and `__str__` for display. - Item.location returns ItemLocation instead of a plain tuple. Runtime-compatible: ItemLocation is a tuple subclass. - TestReport/BaseReport: location typed as ItemLocation, with coercion in __init__ for backwards compat. JSON deserialization in _report_kwargs_from_json reconstructs ItemLocation. - write_item in fixtures.py now uses item.location instead of getlocation(item.function, ...), eliminating the item.function assumption that broke for DoctestItem and custom Item subclasses. - Fixes off-by-one in getlocation display (was co_firstlineno + 1, now correctly co_firstlineno via 0-based storage + 1-based property). Co-authored-by: Cursor AI <ai@cursor.sh> Co-authored-by: Anthropic Claude Opus 4 <claude@anthropic.com>

pytest-xdist (<= 3.x) sends the location parameter from pytest_runtest_logstart/logfinish directly over execnet, which cannot serialize NamedTuple subclasses. Convert to plain tuple at the hook call site and in _report_to_json serialization. Co-authored-by: Cursor AI <ai@cursor.sh> Co-authored-by: Anthropic Claude Opus 4 <claude@anthropic.com>

- Remove unused allow_escape parameter from getlocation() (added in this branch, never called with True) - Add ItemLocation.__str__ coverage tests for None lineindex - Add changelog entry for the feature Co-authored-by: Cursor AI <ai@cursor.sh> Co-authored-by: Anthropic Claude Opus 4 <claude@anthropic.com>

RonnyPfannschmidt requested review from asottile, bluetech and nicoddemus November 21, 2020 13:20

nicoddemus reviewed Nov 21, 2020

View reviewed changes

asottile reviewed Nov 21, 2020

View reviewed changes

bluetech reviewed Nov 21, 2020

View reviewed changes

bluetech reviewed Nov 28, 2020

View reviewed changes

Base automatically changed from master to main March 9, 2021 20:40

RonnyPfannschmidt force-pushed the location-as-tuple branch from 356cbba to 4056e07 Compare June 6, 2021 13:51

RonnyPfannschmidt marked this pull request as draft November 16, 2021 12:51

RonnyPfannschmidt force-pushed the location-as-tuple branch from 304995d to 02ad46a Compare March 17, 2023 08:58

[RFC] code location as a namedtuple

7b826c9

Co-authored-by: Cursor <cursoragent@cursor.com>

RonnyPfannschmidt force-pushed the location-as-tuple branch from 02ad46a to 7b826c9 Compare May 24, 2026 09:16

RonnyPfannschmidt changed the title ~~[RFC] code location as a namedtuple~~ feat: structured CodeLocation and ItemLocation NamedTuples for code locations May 24, 2026

RonnyPfannschmidt changed the title ~~feat: structured CodeLocation and ItemLocation NamedTuples for code locations~~ [RFC] structured NamedTuples for code locations May 24, 2026

psf-chronographer Bot added the bot:chronographer:provided (automation) changelog entry is part of PR label May 24, 2026

RonnyPfannschmidt added this to the 10.0 milestone May 24, 2026

RonnyPfannschmidt force-pushed the location-as-tuple branch from be17a37 to 0705197 Compare May 24, 2026 17:25

Uh oh!

Conversation

RonnyPfannschmidt commented Nov 21, 2020 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

CodeLocation (replaces str from getlocation())

ItemLocation (replaces tuple[str, int | None, str] from Item.location)

Other changes

Uh oh!

nicoddemus left a comment

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

asottile left a comment

Choose a reason for hiding this comment

Uh oh!

Uh oh!

Choose a reason for hiding this comment

Uh oh!

bluetech left a comment

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

RonnyPfannschmidt commented Nov 22, 2020

Uh oh!

Uh oh!

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Uh oh!

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

4 participants

RonnyPfannschmidt commented Nov 21, 2020 •

edited

Loading

`CodeLocation` (replaces `str` from `getlocation()`)

`ItemLocation` (replaces `tuple[str, int | None, str]` from `Item.location`)