refactor: use resolve_results_dir instead of hardcoded /tmp/results

[eltorrero]

Replace hardcoded /tmp/results paths with the resolve_results_dir helper in tests/shared/gnome_shell_steps.py and tests/smoke/features/environment.py (and its duplicate in tests/vanilla-gnome/features/environment.py).

This completes Step 2 of the recommendation in issue #779.

— hive: backend=goose

[kubestellar-hive]

Approved by @castrojo for Hive auto-merge on green CI.

[hanthor]

import os was deleted but never replaced — this fails the Lint gate

In tests/vanilla-gnome/features/environment.py the diff removes the function-local import:

     try:
-        import os
-        if os.path.exists("/tmp/results/atspi_tree.txt"):
+        results_dir = resolve_results_dir(context)
+        if os.path.exists(os.path.join(results_dir, "atspi_tree.txt")):

but no module-level import os is added. The only thing that could bind os in that module is from qecore.common_steps import *.

Merging this branch onto main and running the exact command from pr-validate.yml:

$ ruff check tests/ scripts/ --select E,F,W --ignore E501
F405 `os` may be undefined, or defined from star imports
   --> tests/vanilla-gnome/features/environment.py:195:12
F405 ... :195:27
F405 ... :203:9
F405 ... :204:14
Found 4 errors.

On main the same command reports All checks passed!, so these four are introduced by this PR and Lint & syntax will fail.

At runtime it is worse than a lint nit: if qecore.common_steps does not re-export os, after_all raises NameError, and the surrounding except Exception: pass swallows it — atspi_tree.txt would silently stop being written with no failure anywhere.

Fix: add import os to the module-level imports of tests/vanilla-gnome/features/environment.py.

Minor

The new import is inserted between import sys and import traceback:

import sys
from tests.shared.results_dir import resolve_results_dir
import traceback

That puts a first-party import inside the stdlib block. It belongs with the other from tests.shared... imports below.

Note on CI status

This PR currently shows no check runs at all (total_count: 0) — the fork's workflows appear never to have been approved, so nothing has actually validated it. The three other files in the diff look correct to me; it is only tests/vanilla-gnome that is broken.

---

Generated by Claude Code

[castrojo]

Closing as a duplicate of #825, which implements the same change without a regression.

#791, #823 and #825 all route the AT-SPI tree dump through resolve_results_dir at the same three call sites; your tests/shared/gnome_shell_steps.py post-image is in fact byte-identical to #825's. None of the three is superseded by main.

The blocker here is the one @hanthor already identified, and it still stands: tests/vanilla-gnome/features/environment.py drops the function-local import os from after_all and adds no module-level replacement. That module's header is import sys / import traceback — there is no import os. The four os. uses then resolve only through from qecore.common_steps import *, which is F405 under the required ruff check tests/ scripts/ --select E,F,W gate, and at runtime a NameError swallowed by the surrounding except Exception: pass — atspi_tree.txt would silently stop being written with nothing failing.

Secondary: the new unit test patches tests.shared.gnome_shell_steps.resolve_results_dir to return "/tmp/results", which mocks out the precedence chain it is meant to cover.

#825 carries the same production change with both imports intact, plus tests that exercise the real resolver. Track it there.