feat!: Bump openjd-* Rust crates to the openjd-model 0.11.0 release

[leongdl]

Fixes: n/a (dependency bump for OpenJobDescription/openjd-rs#416 and #420)

BREAKING CHANGE: openjd.model._v1 SimpleAction.script is an openjd.expr.FormatString instead of str, and SimpleAction(script=...) requires one. Read the template text with .raw(). LIST[PATH] defaults must be relative, stay inside the template directory, and are joined with it; relative submitted PATH values are normalized.

What was the problem/requirement? (What/Why)

openjd-rs released openjd-expr 0.10.1, openjd-model 0.10.1 and openjd-sessions 0.7.2 on 2026-09-30 (openjd-rs#416), then openjd-model 0.11.0 and openjd-sessions 0.7.3 on 2026-10-03 (openjd-rs#420). This package pinned 0.10.0 / 0.10.0 / 0.7.1. openjd-sessions has no source change in either release.

One public Rust API change breaks compilation: openjd-rs#419 makes template::SimpleAction::script a FormatString. The PySimpleAction constructor and getter were the two errors.

What was the solution? (How)

Bump the three pins and adapt the two call sites. Following #373's capability-name precedent, the Python-facing SimpleAction.script is now an openjd.expr.FormatString in both directions, matching its args / timeout siblings and the EmbeddedFile.data it desugars to.

The new Rust-side SimpleActionKind / simple_actions / simple_action helpers are not exposed: nothing here consumes them, and step.bash .. step.node already cover introspection. resolve_syntax_sugar changed signature, but the bindings never call it.

Cargo.lock moves only the three crates. THIRD-PARTY-LICENSES.txt was regenerated with scripts/check_third_party_licenses.sh --update and changed only the three version lines.

pyo3 0.29.2 → 0.29.3 (second commit). Rust 1.99.0, stable since 2026-10-01, makes clippy flag clone_on_copy inside pyo3 0.29.2's #[pyclass(from_py_object)] expansion on our Copy enums. That produced 13 errors and failed Rust Quality's clippy -D warnings step on every PR, including unrelated dependabot runs. pyo3 0.29.3 fixes the generated code (PyO3/pyo3#6309). The floor in rust-bindings/Cargo.toml is raised to 0.29.3 so a fresh resolve cannot pick the affected release. Only the four pyo3 lines in THIRD-PARTY-LICENSES.txt changed.

What is the impact of this change?

Every behaviour change below was measured through this package's Python API on both the old and new pins, with expectations copied from the upstream tests' assertions rather than their prose.

Upstream	Python-visible effect	Measured
openjd-rs#419	bash / python / cmd / powershell / node steps are validated at decode, with errors at the authored path (steps[0] -> bash -> args[1]). A malformed script now fails at parse (DecodeValidationError). Non-ASCII step names no longer raise PanicException.	28/28 match upstream. On 0.10.0, 17 of 21 upstream rejection cases decoded: 9 failed later at create_job at the desugared path, 8 were never rejected.
openjd-rs#419 (named only in the squash body)	The comprehension-shadows-let check now covers a step script's authored embeddedFiles[].data.	Decoded on 0.10.0; now rejected at steps[0] -> script -> embeddedFiles[0] -> data. v0 agrees 4/4.
openjd-rs#421	LIST[PATH] defaults must be relative and stay inside the template directory, and are joined and normalized element by element. Relative submitted PATH / LIST[PATH] values are joined and normalized. Defaults are constraint-checked after the join, including environment-template parameters. Absolute submitted values are returned as written.	27/27 POSIX rows match. create_job without a preceding preprocess is unchanged (7 cases identical on both pins).
openjd-rs#410	A job environment's let failure reads like a step script's (script let binding 'q': ...). The missing-extension message lists extensions sorted.	2/2
openjd-rs#410, #417, #418	openjd.expr memory accounting. Some expressions that fit now exceed the limit (coercion to the target type, string slices, attribute base lookups) and some that failed now fit.	Every upstream byte figure reproduces, 21/21. No existing Python test pinned a moved figure.

v0 divergence. v0 preprocess_job_parameters returns LIST[PATH] defaults unjoined and does not normalize a submitted ... openjd-specifications#191 (Template Schemas §2.2, §2.12) makes v1's behaviour normative, so v0 is the non-conformant side. Recorded as two strict xfails in test_known_gaps.py and in reports/model-bindings-quality-evaluation-report.md; fixing v0 is a separate change. On the #419 decode cases v0 agrees with v1 on outcome in 22/24 and on field path in 16/24. The two outcome differences are v0 rejecting Task.File.<step>_script, which predates this bump.

How was this change tested?

• Have you run the unit tests? Yes.

hatch run test: 6323 passed, 26 skipped, 5 xfailed, coverage 94.22%. The +2 skipped are the Windows-only #421 tests; the +2 xfailed are the v0 gaps above. No strict xfail flipped. hatch run lint and hatch run typing clean. cargo build --all-targets, cargo clippy --all-targets -- -D warnings, cargo test and cargo test --doc against rust-bindings/ clean. License verify mode passes.

The pyo3 fix was checked against the toolchain CI uses: rustup run 1.99.0 cargo clippy --manifest-path rust-bindings/Cargo.toml --all-targets -- -D warnings reports the same 13 clone_on_copy errors as CI with pyo3 0.29.2, and 0 with 0.29.3.

New tests:

• test/openjd/model_v1/test_parse.py: TestSimpleActionValidation (all 21 upstream rejection rows, 3 caller-limit rows, the brace-mismatch parse failure, controls) and the embedded-file data comprehension cases. Full-message assertions at the authored path.
• test/openjd/model_v1/test_create_job.py: SimpleAction caps, LIST[PATH] defaults, submitted-path normalization, post-join constraints, Windows rows, let-failure messages, sorted missing extensions.
• test/openjd/model_v1/test_template_types.py: script type, str refusal, pickle.
• test/openjd/model_v1/test_known_gaps.py: two v0 strict xfails.
• test/openjd/expr/test_memory.py, test_slicing.py: memory-accounting figures, extreme slice step.

Mutation check, each mutant rebuilt from source, bytecode cleared, restored by checksum:

Mutant	Result
pins and the adaptation reverted to 0.10.0 / 0.10.0 / 0.7.1	62 of 79 new or edited cases fail; the 17 survivors are controls whose behaviour 0.10.0 shares
script getter returns str	4 failed
constructor accepts str	4 failed

What could NOT be verified

• The three-auditor review of the new tests was not run. The mutation results above are the author's own. A separate semantic review read every added test for vacuous patterns and found none, but it did not run independent mutants.
• The Windows #421 rows were not run locally; they run on the windows CI lane only. The Python 3.9 lane was not run locally either.
• URI-allowed LIST[PATH] rows are unreachable from Python: the binding hard-codes allow_uri_path_values=false at rust-bindings/src/model/create_job_fns.rs:158. Upstream test_create_job.rs covers them.
• TestExtremeSliceSteps pins the openjd-rs#418 overflow fix only in a build with overflow checks. 'hello'[1::9223372036854775807] panicked on 0.10.0 only in a dev build; a hatch editable install of 0.10.0 already returned e. Its docstring says so.
• src/openjd/_openjd_rs.pyi was hand-edited, because scripts/generate_stubs.sh does not run on macOS (precedent ffd3f22). A stub regeneration on Linux should reproduce the edit, but nobody has checked.

Was this change documented?

• Are relevant docstrings in the code base updated? Yes.

• specs/python-model-interface.md: SimpleAction.script as FormatString, the SimpleAction decode-validation rules, and the PATH / LIST[PATH] preprocessing rules (including that walk-up also permits absolute defaults).
• src/openjd/_openjd_rs.pyi: the script type on SimpleAction.
• reports/model-bindings-quality-evaluation-report.md: the v0 PATH divergence.

Is this a breaking change?

Yes. SimpleAction.script is openjd.expr.FormatString instead of str. A caller reading the template text uses .raw(); a caller constructing one wraps in FormatString(...).

Templates that decoded before may now be rejected: SimpleAction steps that only failed at create_job (or never), embedded-file data with a comprehension variable shadowing a let binding, and LIST[PATH] defaults that are absolute or escape the template directory.

Does this change impact security?

In the direction of enforcement. LIST[PATH] defaults can no longer be absolute or escape the template directory unless allow_job_template_dir_walk_up=True, and defaults are constraint-checked after the join. SimpleAction scripts are validated at decode rather than at job creation.

---

By submitting this pull request, I confirm that you can use, modify, copy, and redistribute this contribution, under the terms of your choice.