Align builtin reduction callback controls - #9718

Summary

Group CPython 3.15 min/max streaming, sum and all/any iteration, argument binding and callback ownership controls in 274 changed lines across three production Rust files.

• Consume a single iterable one item at a time, interleaving key and comparison callbacks without materialization or length-hint calls. Multiple positional candidates use borrowed caller slots.

• Keep raw caller arguments alive until internal cleanup finishes. Share one min/max engine with explicit winner/current item and key ownership and correct success/error release order.

• Reuse ArgSpec through a borrowed keyword-only entry point. Validate arity, keyword count/names and default restrictions before obtaining an iterator.

• Scope CPython-style typo suggestions and lossless WTF-8 keyword diagnostics to the new entry point. Reuse the existing suggestion algorithm with UTF-8 byte-length thresholds and suppress suggestions for surrogate-bearing names. Existing parser and suggestion consumers retain their behavior.

• Bind sum inputs before iterability checks, acquire and validate its iterator before rejected-start errors, and retain the original start through all callbacks.

• Preserve generic sum accumulator-before-item-before-iterator error cleanup without changing compensated arithmetic or numeric fast paths.

• Share one all/any engine with canonical iterator validation and one cached next slot. Python slot wrappers continue to resolve a replaced/deleted next dynamically, including mutation from truth callbacks and item finalizers.

• Preserve raw inputs, item-before-next cleanup, short circuiting and length-hint avoidance without changing generic iteration or min/max/sum traversal policy.

Validation

Pinned CPython reference: d220409a13b77de7a6c30abe16ea90db01528edc. The focused oracle passed 22 grouped assertions and recorded 30 outcomes/traces, covering streaming, mutation, length-hint avoidance, binding, defaults, comparisons, exception propagation and internal finalizer order. These are reference/source checks, not RustPython runtime tests.

Independent source review passed, including 344 additional CPython diagnostic assertions over 172 keyword names. The initial compiler run caught a Result-versus-Option guard method mismatch; the corrected final source passed threaded cargo check in 1m05s and all-targets Clippy in 2m58s. All 12 unchanged baseline diagnostics match exactly, without suppression. Ordinary commit hooks passed. No tests, Lib files, expected failures or workflows are changed.

Limits and integration

The original validation used reference and source checks. Current native validation is recorded in the borrowed-source follow-up below. Full runtime differential tests and separate platform/feature graphs remain outside that validation; hosted CI is pending.

Caller-frame argument destruction order can depend on the original call syntax, which FuncArgs no longer records. This patch retains raw owners through internal cleanup; it does not impose one universal caller-tail ordering or claim to solve that wider lifetime issue. Duplicate-keyword detection belongs to VM call binding before FuncArgs' unique-key map.

PR #9668 makes ArgSpec fields and parse_with public. Read-only three-way source composition succeeds while retaining those public interfaces and the new shared parser internals; the combined graph was not compiled. #9699 does not implement min/max behavior, and the examined older builtin changes do not duplicate this group. Existing non-opted-in parser consumers' surrogate diagnostics and suggestion behavior remain unchanged.

AI-assisted contribution by OpenAI Codex; assistance is disclosed in the commit trailer.

Sum follow-up validation

Independent exact-source review passed 30 pinned CPython oracle cases, 14 additional numeric error controls and the existing arithmetic snippet on CPython. Final threaded cargo check passed in 1m05s and all-targets Clippy in 2m17s; all 12 baseline diagnostics match unchanged. An initial redundant-clone diagnostic was fixed by making the existing iterator-then-raw-start release order explicit, without suppression or removing retention. Normal commit hooks passed. These earlier checks were reference/source and compilation checks; the current native results are recorded below.

The sum wrapper continues to use shared FromArgs binding. Exact total-arity/duplicate-argument diagnostic precedence and keyword typo suggestions are deferred; this does not alter shared binder behavior for other functions.

All/any follow-up validation

Independent exact-source review passed bounded CPython controls for truth and item-finalizer next-slot mutation (eight cases), invalid iterator results, short circuiting, errors and length-hint avoidance. The existing test_all and test_any methods were extracted unchanged and passed on pinned CPython. Final threaded host-env/VM/stdlib cargo check passed in 1m11s, all-targets Clippy in 2m47s, and all 12 baseline diagnostics match unchanged. Normal commit hooks passed. A newly unused adapter import was removed before final validation. No native RustPython runtime test was included in that earlier validation; current results are recorded below.

Terminal StopIteration conversion still uses the shared wrapper, which discards the original exception before returning its payload. This follow-up preserves the previous payload-before-iterator release order; full traceback-dependent terminal ownership parity remains deferred. Existing generic binder diagnostic precedence is unchanged.

Borrowed-source follow-up

Rebased independently onto main at fb57e68, preserving all seven prior commits patch-for-patch. The follow-up changes only the shared all/any helper: borrow &PyObject and obtain its iterator through get_iter, while each wrapper retains the original iterable through callbacks and cleanup. This removes coupling to an owned-reference type; it does not reduce reference-count clones, because get_iter still acquires an owned reference internally.

Plain PyIter conversion would release the original source too early. The existing ArgIterable adapter retains its source but does not validate the object returned by __iter__ when acquiring it. Restoring that adapter to sum would move invalid-iterator rejection after invalid-start checks. A common source-retaining, once-validated iterator argument remains a separate proposal; no new conversion framework is introduced here.

Independent review, formatting and ordinary commit hooks passed. This follow-up changes no tests, expected-failure markers or unrelated production files.

Validation on exact commit 29ef984:

• Full workspace tests with the documented WASM, venvlauncher and C-API exclusions: 1,513 passed, 20 ignored across unit and documentation suites.
• Full workspace all-target Clippy and separate C-API all-target Clippy: passed with zero warnings. Both used ordinary Clippy without -D warnings.
• Separate C-API tests: 149 passed, 2 ignored.
• Fresh release RustPython -m test test_builtin: SUCCESS, 157 tests run and 14 skipped.

Development and Rust test profiles used CARGO_PROFILE_DEV_DEBUG=0 and CARGO_PROFILE_TEST_DEBUG=0; release validation used the normal release profile. The safe existing CPython 3.12.14 host interpreter was verified in the fresh PyO3 build configuration. C-API checks used the repository's RustPython 3.15 configuration without an external interpreter. The CPython 3.15 comparison for this follow-up was source-only; no CPython 3.15 executable was run. Hosted CI for the updated head remains pending.