tests: the quoting-sentence inventory misses a quoter passed as a value and pins functions where it means calls, and the launcher guard has no row for src/openreposhape/cli.py or setup-project.py #759

Goal: the two guards #665 tightens hold what their comments promise for the forms they still cannot see: a host-dialect quoter reached any way but a direct call, a function with one call switched to POSIX and one not, and a tool that starts a Python child through another spelling of subprocess.

Filed by lane openRepoShape-5, from the checklist candidates in #665's PR (#724, head 151a0861). Tests only. Each line is read from source; the second also has a measured count (below).

Lands after: lane 5's #665 (PR #724, fix/665-four-guards). The inventory (host_calls, host_quoting, NAMED) and LAUNCHERS' scripts/family.py row are in that PR; main has the older shape of both. Line numbers below are at 151a0861 for the inventory and at main 66b230ed for LAUNCHERS.

Lines

• A quoter handed on as a value is not a call, and the inventory does not see it (tests/test_readme_windows_quoting_sentence.py:141-163, host_calls; its docstring says so: "A quoter handed on as a value (map(shell_quoted, words)) is not a call and is not seen"). functools.partial(shell_line) is the same. Read from source: nothing in the shipped tools does it today (grep of the repository for map(shell_quoted, map(shell_word, map(quote_arg, partial(shell_, partial(quote_arg outside tests/ finds none), so no host-dialect line escapes the sentence now. Direction: treat a Name or Attribute that is a host quoter and is an argument of any call, or the value of an assignment the alias walk (quoter_names, :118) already follows, as a use of it, and name its function in host_quoting. Test it with the two forms above, as test_a_quoter_imported_under_another_name_is_a_host_call (:247) does for aliases.
• The function pin counts functions, not calls. test_the_functions_that_quote_for_the_host_are_the_ones_the_sentence_names (:234-246) compares set(host_quoting(tool_sources()).get(tool, {})) with NAMED[tool], a set of function names. A function with two host-dialect calls where only one is switched to 'posix' still has a call that follows the host, so it still passes: the test cannot tell. Measured on main 66b230ed with render-cli-reference.py and tests/: the CLI reference's children write no bytecode under templates/, the launcher guard reads family.py's siblings child, the quoting-sentence inventory follows an aliased quoter, a suffixless tool and each tool's calling functions, and an S5778 pytest.raises builds its argument first (#665) #724's host_calls over scripts/family.py and setup-project.py: _switch_command 1, _print_next 1, _checked_mount_path 1, _add_command 3, _print_next_steps 2, and family_next_command 2. Only the first three are single-call functions, which is why render-cli-reference.py and tests/: the CLI reference's children write no bytecode under templates/, the launcher guard reads family.py's siblings child, the quoting-sentence inventory follows an aliased quoter, a suffixless tool and each tool's calling functions, and an S5778 pytest.raises builds its argument first (#665) #724's body says "Today the one-call functions are _switch_command, _print_next, _checked_mount_path": for the other three, one call switched to POSIX leaves the function in the set, so it is seen only if every call of it is switched. Direction: make NAMED map each function to its number of host-dialect calls and compare the counts, so one call switched in _add_command (3 to 2) fails until the sentence and the table change together.
• LAUNCHERS has no row for src/openreposhape/cli.py or setup-project.py (tests/test_validator_child_encoding.py:610-613, two rows; interpreter_runs at :616-631 finds only subprocess.run([sys.executable, ...]) by ast.unparse(argv.elts[0]) == "sys.executable"). src/openreposhape/cli.py starts its Python children with run_child([sys.executable, str(script), ...], standard.child_environment()) at :436, :462, :474 and :564, which is neither subprocess.run nor the matched spelling; setup-project.py starts its with its own run([PYTHON, ...]) (:696, :2129, :2260, :2326, :2501; PYTHON = sys.executable at :116), whose subprocess.run at :288 takes no env=. Read from source, not measured: whether setup-project.py's children need inherited_child_environment() is the guard's own question ("a bare subprocess.run([sys.executable, ...]) takes whatever the operator's shell holds, which is the defect", :636-639), and nothing here has asked it of these two. Direction: teach interpreter_runs the two spellings (a call to run_child or to a name bound to sys.executable as the first element), add one row each, and see which fails: a failing setup-project.py row is a finding for its run, and a passing one pins what is true.

Verify: tests/test_readme_windows_quoting_sentence.py and tests/test_validator_child_encoding.py pass on Linux. Mutants, each killed by a new case: map(shell_quoted, ...) added to update-shape.py; one of _add_command's three calls switched to 'posix'; env= dropped from one run_child call in cli.py.

Overlaps. fix/665-four-guards (#724) edits both test files, so this lands after it. #711 (the Git Bash dialect, v1.1.1) will move NAMED when it changes the host question; whichever lands second re-gates against the other. scripts/family.py is edited by seven open branches (#646, #684, #686, #688, #689, #692, #693) but this issue edits no tool. No pinned file changes.

Parent: #461.

Milestone: v1.1.1. Tests only, and no shipped line is unguarded today.

🤖 Generated with Claude Code