tests: PASTE_ENV in tests/test_doctor_legs_members_next.py is an import-time snapshot that overwrites GIT_CONFIG_COUNT, seven tests hand-build the same append beside appended_to_command_scope, and six git clone --no-local calls build clone_just_pushed's options by hand #758

Goal: every test that gives git a setting through GIT_CONFIG_COUNT appends to the entries the environment holds, built at the call by one helper, so a setting set by the session fixture, the Windows job or another fixture reaches every git the test starts.

Filed by lane openRepoShape-5, from rv662's review of #662's PR (#748, "PASTE_ENV ... is a snapshot of os.environ taken at import") and from that PR's own "Left out" (five tests that build the same arithmetic by hand; six git clone --no-local calls, "not read closely"). Tests only.

Lands after: lane 5's #645 (PR #713, fix/645-fixture-copy-maintenance-race, head 979bb394), which adds appended_to_command_scope and the autouse no_auto_maintenance fixture, and #662 (PR #748, fix/662-insteadof-helper, head deb029c9), which adds insteadof_entries, insteadof_args and submodule_update_through_git and makes read_through_git append. Neither is on main.

Lines (path:line at main 66b230ed; the helpers are at deb029c9:tests/conftest.py: appended_to_command_scope :279, insteadof_entries :107, submodule_update_through_git :192, read_through_git :1595)

• PASTE_ENV is built once at import and sets GIT_CONFIG_COUNT to 1 (tests/test_doctor_legs_members_next.py:63-68; used at :101, :442, :514, :609, :785 and :1018). Read from source, not measured. It copies os.environ when the module is imported, which pytest does at collection; tests/conftest.py: the session turns git's automatic maintenance off, in command scope and in a global config that includes the person's own, so a copytree of a fixture no longer meets a detached repack deleting its loose objects (#645) #713's session fixture no_auto_maintenance sets GIT_CONFIG_COUNT, its two keys and GIT_CONFIG_GLOBAL later, at session start. So the copy holds neither scope, and its own "GIT_CONFIG_COUNT": "1" would overwrite a count the environment did hold (the Windows job holds two). A line pasted through paste() (sh -c row["next"]) that clones a local remote, such as the git submodule update --init --recursive --checkout the rows hand over, therefore runs with git's automatic maintenance on, which is the race tests: the holder fixtures copy the family fixture with shutil.copytree while git's background gc or maintenance can still be packing it, so a test errors at setup when a loose object vanishes mid-copy (#440's race, through a file copy) #645 closes for every other clone the suite makes. tests/conftest.py: one helper builds the url.<file URI>.insteadOf rewrite that clone_just_pushed, read_through_git and the adopt tests' pasted each built, and the empty-mount submodule update --init sites use it (#662) #748 fixed the same overwrite in read_through_git (its body: "it used to overwrite the count, which dropped tests/conftest.py: the session turns git's automatic maintenance off, in command scope and in a global config that includes the person's own, so a copytree of a fixture no longer meets a detached repack deleting its loose objects (#645) #713's two settings") and moved three submodule update calls of this file onto its wrapper, and left PASTE_ENV. Direction: build the environment per call: {**os.environ minus PYTHONPATH, **GIT_IDENTITY, PYTHONDONTWRITEBYTECODE, **appended_to_command_scope(os.environ, [("protocol.file.allow", "always")])}, in a function the six uses call, as rv662 asks ("Build it per test from read_through_git").
• Seven tests build held = int(os.environ.get("GIT_CONFIG_COUNT") or 0) and append by hand. Read from source: tests/test_adopt_git_reads.py:114, tests/test_adopt_nested_attributes.py:386, tests/test_adopt_reencoded_files.py:105, tests/test_adopt_root_symlinks.py:226, tests/test_adopt_ignored_shape_files.py:585, tests/test_bootstrap_after_bump.py:151 and tests/test_family_bump_remove_rollback.py:680. tests/conftest.py: one helper builds the url.<file URI>.insteadOf rewrite that clone_just_pushed, read_through_git and the adopt tests' pasted each built, and the empty-mount submodule update --init sites use it (#662) #748's body counts five, from its base; main holds seven by grep. tests/test_clone_just_pushed.py:84 is the helper's own test and stays. Direction: one call to appended_to_command_scope each, with the entries it adds. The other GIT_CONFIG_COUNT literals in tests (tests/test_adopt_empty_symlinks.py:73, tests/test_non_utf8_git_reads.py:81 and :423, tests/test_family.py:77, tests/test_family_dry_run.py:81 and others) set the count to a fixed number and ride in the same sweep only where the environment they replace can hold entries; read each before changing it.
• Six git clone --no-local calls build the options clone_just_pushed builds. Read from source, not read closely (tests/conftest.py: one helper builds the url.<file URI>.insteadOf rewrite that clone_just_pushed, read_through_git and the adopt tests' pasted each built, and the empty-mount submodule update --init sites use it (#662) #748's own words): tests/test_bump_leg_dry_run_truth.py:98 and :110, tests/test_doctor_legs_members_next.py:1890 and :1991, tests/test_doctor_pin_restore_merges.py:714 and tests/test_family_bump_to_checks.py:235. Each passes protocol.file.allow and --no-local by hand; the helper adds the url.<file URI>.insteadOf rewrite that makes the clone read the remote through git. Whether a given call clones one of the session's remotes (and so wants the helper) or a repository the test made itself (and needs nothing) is the question tests/conftest.py: one helper builds the url.<file URI>.insteadOf rewrite that clone_just_pushed, read_through_git and the adopt tests' pasted each built, and the empty-mount submodule update --init sites use it (#662) #748 left open; answer it per call, as tests/conftest.py: one helper builds the url.<file URI>.insteadOf rewrite that clone_just_pushed, read_through_git and the adopt tests' pasted each built, and the empty-mount submodule update --init sites use it (#662) #748 did for the 55 clone sites.

Verify: tests/test_doctor_legs_members_next.py, the seven files above, tests/test_clone_just_pushed.py and tests/test_no_auto_maintenance.py pass on Linux. A test beside tests/test_no_auto_maintenance.py that starts paste()'s shell with GIT_CONFIG_COUNT=2 in os.environ and asserts git config --show-origin through it lists the two maintenance keys and protocol.file.allow. A hygiene test that fails a new "GIT_CONFIG_COUNT": "<digit>" literal in tests/ outside conftest.

Overlaps. #748 (fix/662-insteadof-helper) edits tests/conftest.py, tests/test_doctor_legs_members_next.py (the submodule update sites, :631, :987 and :1054 at its base), tests/test_family_bump_remove_rollback.py and tests/test_clone_just_pushed.py, so this lands after it. fix/648-bootstrap-and-gate-order (#682), fix/681-validate-family-remedies (#747) and fix/704-bootstrap-leg-moves (#745) edit tests/test_doctor_legs_members_next.py too (one hunk each, at :1587, far from :63). fix/704-bootstrap-leg-moves edits tests/test_bootstrap_after_bump.py. l3-687-pin-restore-other-side edits tests/test_doctor_pin_restore_merges.py. No pinned file changes.

Parent: #461.

Milestone: v1.1.1. Tests only, and no test fails today.

🤖 Generated with Claude Code