[Bug]: the required Windows lane fails on the long-path CLI budget fixture because the fixture pins the workload path length but not its JSON-escaped cost

Offen Anfängerfreundlich
#6,842 1 Kommentar 0 Reaktionen 0 zugewiesene Personen Auf GitHub ansehen

Maintainer antworten meist innerhalb von 1 Tag

Dieses Issue hat noch niemand übernommen.

Bewertung

Schwierigkeit
2/5
Geschätzter Aufwand
1-3 Stunden
Anfängerfreundlichkeit
84/100
Issue-Typ
Bug
Klarheit
Klar beschrieben
Aktivitätsstatus
Aktiv
Tech-Stack
json, powershell, python
Bereich
ci-cd, testing

Rechercherichtung

Beginne mit tests/control_plane/test_cli_output_budget.py und der Long-Path-Fixture, die die Root auf 128 Zeichen festlegt. Reproduziere das Problem mit dem dokumentierten pytest-Befehl und einem tieferen temporären Namespace. Erledigt ist die Aufgabe, wenn alle drei Long-Path-Budgettests dieselbe gemessene Arbeitslast erzeugen und unabhängig von der Tiefe des temporären Windows-Verzeichnisses bestehen.

Vom Indexierungsmodell aus dem Issue-Text verfasst.

Beschreibung

LoopX version or commit: main dc9fdbb51

Host or runtime surface: Not host-specific (fails in the required windows-powershell lane; the defect is in the budget fixture, not in the CLI)

Area: Build, packaging, installer, or CI (required-check regression in Python Tests)

Problem

The required windows-powershell job of the Python Tests workflow is red on the current main head, and it fails on exactly three tests:

FAILED tests/control_plane/test_cli_output_budget.py::test_long_path_budget_is_independent_of_runner_temp_prefix
FAILED tests/control_plane/test_cli_output_budget.py::test_long_path_budget_uses_native_temp_without_a_drive_root_directory[False]
FAILED tests/control_plane/test_cli_output_budget.py::test_long_path_budget_uses_native_temp_without_a_drive_root_directory[True]
AssertionError: heartbeat_prompt_brief/json emitted 10522 chars; variant ceiling is 10500

merge-gate lists windows-powershell in needs, so the whole required qualification is red for that lane. Runner: job 117808033252 in run 39172287050.

The long-path budget fixtures build a fixture root whose path length is pinned to 128 characters so that the measured packet is host-independent:

parent = Path(tempfile.gettempdir() if os.name == "nt" else "/tmp").resolve()
...
while len(str(parent / prefix)) + 8 > 128:
    parent = parent.parent
prefix += "p" * (128 - len(str(parent / prefix)) - 8)
with tempfile.TemporaryDirectory(prefix=prefix, dir=parent) as directory:
    root = Path(directory).resolve()
    assert len(str(root)) == 128

Path length is not the quantity the budget measures. The emitted heartbeat_prompt_brief JSON quotes the fixture root and the paths derived from it about twenty times, and JSON escapes every Windows separator in each quoted copy, so emitted size grows with both the length and the separator count. Varying the fixture root on this checkout gives:

emitted_chars(root) = 7782 + 20 * len(root) + 20 * separators(root)

At the pinned 128-character length this reduces to 10342 + 20 * separators, the rows below:

A 128-character root therefore measures differently on every host, depending only on how deep its temporary namespace is:

Fixture root separators measured chars verdict (ceiling 10500)
D:\Temp\<128-char root> 2 10382 passes
D:\Temp\a\b\c\pytest-1\<128-char root> 5 10442 passes
hosted Windows runner C:\Users\<user>\AppData\Local\Temp\pytest-of-<user>\pytest-1\test_...\<128-char root> 9 10522 fails
local deep TEMP (five nested segments) 10 10542 fails

So the guard reports a failure for a deep temporary namespace rather than for a change in the emitted packet, and the size of the measured workload is not the one the ceilings were pinned against. #6833 (merged in dc9fdbb51) exposed this by adding long_path_budget to the Windows lane selection and by holding only the length constant; the runner's temp namespace is four segments deeper than the local one, which is exactly the observed 10522 - 10442 = 80 = 4 x 20 difference.

Impact

  • windows-powershell is a required lane (merge-gate needs it): the required qualification is red on main, and every pull request based on current main inherits a failing required check it did not cause.
  • The same three tests are part of the scheduled Python Tests qualification, so the daily run is red for this cause as well.
  • The guard cannot be trusted on Windows in either direction: on a deeper namespace it fails without any packet change, and on a shallower one it measures a smaller workload than the pinned ceiling describes.

Minimal reproduction

git checkout dc9fdbb51
uv sync --extra test
# point the host temp namespace deeper than the workload, as the hosted runner does
# PowerShell: $env:TEMP = $env:TMP = "<deep-drive>/a/b/c/d/e"
TEMP=<deep-drive>/a/b/c/d/e uv run --extra test python -m pytest -q \
  tests/control_plane/test_cli_output_budget.py -k long_path_budget

Observed on dc9fdbb51: the same three failures, with the emitted count 10542 instead of 10522 because the namespace is one segment deeper.

Expected behavior

The three tests qualify the same workload on every host: a fixture root that the emitted-text budget sees with a constant cost (or a comparison that accounts for the root the fixture actually created), so the verdict does not depend on the depth of the host's temporary namespace.

Actual behavior

The verdict depends on the host's temporary-namespace depth, and the required Windows lane stays red on main until the fixture holds the measured quantity constant.

Vorherrschende Sprache
Python
Sterne
6.2k
Forks
596
Ø Merge
5 Std. 40 Min.
Gemergte PRs (30 T.)
927

Entwicklungsumgebung

Erste Schritte

  1. Lesen Sie das ganze Issue und danach den Beitragsleitfaden des Projekts.
  2. Schreiben Sie ins Issue, dass Sie es übernehmen — das erspart doppelte Arbeit.
  3. Forken Sie das Repository und arbeiten Sie in einem Branch.
  4. Öffnen Sie einen Pull Request, der die Issue-Nummer nennt.

Mehr aus loopx-project/loopx

Alle Issues in loopx-project/loopx

Ähnliche Issues

Weitere Issues zu Python

Neue Issues direkt in Ihr Postfach

Eine kurze Übersicht über anfängerfreundliche GitHub-Issues.