setup-project.py's run and the assembly root's bootstrap.py _run start Python children with no environment, so a cp1252 stdout gets two encodings from setup and a name it cannot hold ends make bootstrap in validate-manifest.py #836

Goal: every Python child that setup-project.py and the assembly root's bootstrap.py start writes in the encoding the parent's own stdout has, so a stdout that cannot hold a character ends in neither a stream of two encodings nor a traceback in a validator.

Filed by lane openRepoShape-5, from the checklist candidates of #759's PR (#808, writer w759: candidates 1 to 3, and the finding its body measures), and from lane 2's review of that PR (rv808, SHOULD 1: "there are six Python children, not five"). #759 itself (the guards) is not reopened: it holds the setup-project.py row as a strict xfail (NOT_YET) and changes no tool. The first two lines below were measured by w759 and are not re-run here; the third is read from source.

Lands after: lane 5's #759 (PR #808, fix/759-quoting-launcher-guards, head 84ce67c9), which adds the LAUNCHERS rows and NOT_YET (tests/test_validator_child_encoding.py:616 and :628 at that head). The assembly bootstrap.py is also edited by #764 (PR #815) and #710 (PR #824); whichever lands second re-gates against the others. path:line is at main 6626a3c5.

Lines

• setup-project.py reconfigures its own stdout to UTF-8 and gives each Python child an environment of its own, so the run writes two encodings (w759, measured). main reconfigures both streams to UTF-8 (setup-project.py:2788-2791), and run (:269-300) calls subprocess.run with no env= (:288-291), so a child inherits the stdout and writes in the code page its environment gives. Six children start through run: the re-run of itself in a checkout (:696), the naming validator (:2198), the plan (:2330) and its re-run for the diagnosis (:2312, captured), the scaffold (:2395) and scripts/bootstrap.py (:2570). tests/: the quoting-sentence inventory sees a quoter handed on as a value and pins each function's host-dialect calls, and the launcher guard reads run_child and run([PYTHON, ...]), with rows for cli.py and setup-project.py (#759) #808's body measured setup-project.py --project Sample --yes --org demoorg --local-remote-dir <dir> --into <dir> with a gh stub: under PYTHONIOENCODING=cp1252:strict and under cp1252:surrogateescape, exit 0 and 9 of 129 stdout lines that are not valid UTF-8 (the naming validator's three family ... rows with their em dash, the scaffold's four shape ..., spec and code lines with their ellipsis, and two lines of the validators bootstrap.py runs); under utf-8:strict, none. Nothing raises, since scaffold-project.py and scripts/bootstrap.py escape what their stdout cannot hold; the stream is two encodings. rv808 found the sixth child (:2312 at its base, again = run(plan, ...)), which tests/: the quoting-sentence inventory sees a quoter handed on as a value and pins each function's host-dialect calls, and the launcher guard reads run_child and run([PYTHON, ...]), with rows for cli.py and setup-project.py (#759) #808's guard cannot see, because plan is a parameter. Direction: run takes the environment: inherited_child_environment for a child that inherits stdout (so PYTHONIOENCODING is the stream's own encoding, after the reconfigure, and a handler that never raises) and captured_child_environment for capture=True, which run decodes as UTF-8 (scripts/shape_materialize.py:266 and :286; bump-leg.py, update-shape.py and family.py already pass them). setup-project.py cannot import them: it runs before there is a checkout to import from (w759), so it holds a local copy of each. Then take the setup-project.py row out of NOT_YET; rv808 asked that the guard read the env= of the function a tool defines as its starter (def run), so a fix inside run flips the row and :2312 is covered.
• The assembly root's bootstrap.py starts the validators with no environment, so a project name the code page cannot hold ends make bootstrap (w759, measured: it is bump-leg.py (a real bump) and update-shape.py apply run the project's own validators as children that cannot print its display name on a cp1252 stdout, so a project named 日本 is refused and rolled back as bump-leg-validators-red or update-validators-red under PYTHONIOENCODING=cp1252:surrogateescape, and the refusal blames the validators #254's defect in a pinned copy). _run (templates/assembly-root/scripts/bootstrap.py:143-150, its one caller at :1088) calls subprocess.run([sys.executable, *args], cwd=str(root), check=False); the parent has called escape_unencodable_stdout() (:1096), the child has not. w759 scaffolded a project locally, set project.yaml name: to two CJK characters, and ran PYTHONIOENCODING=cp1252:strict python3 scripts/bootstrap.py: validate-manifest.py raised UnicodeEncodeError at print(f"manifest ok: {manifest.get('name')} ...") (templates/assembly-root/scripts/validate-manifest.py:1037 at main) and bootstrap ended bootstrap FAILED: manifest (exit 1), exit 1. With utf-8:strict it printed manifest ok: <name> (sample), 3 legs and exited 0. make bootstrap and setup-project.py step (7) both run it. Direction: _run passes env= built as the line above says. scripts/shape_materialize.py is not in COPIED_VERBATIM, so a project does not carry it; both roots' bootstrap.py import their shared code from scripts/repo_shape.py (templates/family-root/scripts/siblings.py:183-189), so the helper moves there, or the script holds a copy. Either is a shape bump: the copy of bootstrap.py is re-pinned in every project, and repo_shape.py is too if the helper moves. Bundle it with bootstrap.py (the assembly root's): a SHAPE_UPSTREAM_PATH directory nobody may enter ends in a traceback, a validator's refusal is reported as FAILED, exit 1, a reftable leg's cherry-pick goes unseen, and docs/cli.md says bootstrap places each leg AT the pin with no condition #764's and bootstrap.py and repo_shape.py: a leg on a branch whose name is not UTF-8 makes every later make bootstrap end in a UnicodeDecodeError, and a character the filesystem encoding lacks still ends step (a) in a traceback #710's edits to the same file.
• The guard has no row for the two pinned scripts, and the family root's siblings.py starts a child the same way, with no defect found. LAUNCHERS (tests/test_validator_child_encoding.py:611 at main) lists tools by file; the survey tests/: the quoting-sentence inventory sees a quoter handed on as a value and pins each function's host-dialect calls, and the launcher guard reads run_child and run([PYTHON, ...]), with rows for cli.py and setup-project.py (#759) #808 made with its new interpreter_runs over every shipped file found the assembly bootstrap.py and templates/family-root/scripts/siblings.py (place, :567-568: subprocess.run([sys.executable or PYTHON, str(script)], cwd=str(path), check=False), the member's own bootstrap.py) with no environment, and no row for either, because both are pinned copies. For siblings.py the parent calls escape_unencodable_stdout() (:1351) and the child is the member's own bootstrap, which does too, so the two agree on the code page; nothing was measured to fail there. The survey also listed shape-doctor.py:1697 (the doctor sets PYTHONIOENCODING for its own children) and src/openreposhape/standard.py:560 (a digest program with captured output and an env handed in), which need nothing. Direction: once the second line lands, add the assembly bootstrap.py row; add siblings.py's row only if the guard can read a pinned copy without making the next edit to it a re-pin of the test. If it can, say what the row pins (that the child inherits an environment the code page agrees with), as the cli.py row does.

Verify: tests/test_validator_child_encoding.py, tests/test_setup_project_py.py, tests/test_setup_sh.py, tests/test_bootstrap_without_git.py, tests/test_bootstrap_after_bump.py and tests/test_repo_hygiene.py pass on Linux. For the first line, #808's probe as a test: setup-project.py --yes --local-remote-dir under PYTHONIOENCODING=cp1252:strict and cp1252:surrogateescape, a gh stub first on PATH, asserting exit 0 and that every stdout line decodes as the encoding the parent wrote (none is "not valid UTF-8" where the parent says UTF-8), and the NOT_YET row removed. For the second, a scaffolded project whose name: the code page cannot hold, run under cp1252:strict: exit 0 and manifest ok, with the validator's name printed through the escape. One mutant per env= put back to nothing, each caught. No Windows run (CI is Linux only, #597): the code page is emulated with PYTHONIOENCODING, as #254's and #808's tests do.

Overlaps. #808 edits tests/test_validator_child_encoding.py. #756, #811 and #821 edit setup-project.py (#811 the name checks near :350-470, #821 the Git Bash dialect), far from run and the children. #815 and #824 edit the assembly bootstrap.py; #772, #775, #788 and #794 (lane 3's siblings.py and doctor branches) edit templates/family-root/scripts/siblings.py. #817, #821, #823 and #827 edit scripts/repo_shape.py, which a moved helper would also edit. The assembly bootstrap.py and repo_shape.py are pinned copies in every project.

Parent: #459.

Milestone: v1.1.1. Each case needs a stdout whose code page cannot hold a character the run prints (Windows' ANSI page on a redirected stdout): the stream is mixed, or make bootstrap refuses a project that is fine. Nothing is lost, #254's fix covers bump-leg.py and update-shape.py apply, and v1.0.0 shipped with it as v1.1.0 will.

🤖 Generated with Claude Code