[rmorozov]

Requested by Ruslan · project thread

Problem

Since 0357 S3 the Python module goes through the public host-object contract. That contract iterates every IsSequence object that has a Len by index (HostSequenceList: Len + ItemAtTo). A list subclass that overrides __iter__ therefore looped over its stored items:

class L(list):
    def __iter__(self): return iter(["x", "y"])
Template("{% for i in l %}{{ i }}{% endfor %}").render(l=L([1, 2, 3]))  # Jinja2: xy, module: 123

The verifier found a second case. A subclass that overrides only __getitem__ looped through that override (GGG), but Python's list iterator ignores it and gives 123.

Change

• Public API. New public virtual IHostObject::IteratesByIndex(), true by default. EngineBridge::Iterate uses HostSequenceList only when it returns true. Otherwise it goes through Iterate. This adds a vtable slot before the G1 freeze, as the backlog report recommended, and is recorded in docs/api-2.0.md section 7 item 10.
• Python module. PyHost::IteratesByIndex is true only for an exact list or tuple. A subclass is iterated through iter(): its own __iter__ if it has one, otherwise the base type's iterator, which is also live. The exact-type fast path is unchanged.
• Codex follow-up: list operators. +, *, comparisons and str % x keep reading the stored items, as Python's list operations do. host::AsSequence now reads a new private IEngineHostObject::SequenceItems(), which the bridge answers by index.
• Codex follow-up: loop.length. A loop over any host object takes loop.length (and revindex) from the object's Len when the template first reads it, as Jinja2 takes len(iterable) (SetLengthOfValue, src/statements.cpp). It falls back to counting the items when the object has no length. loop.last still comes from peeking.

Validation

• PublicHostObjectTest.SequenceThatIteratesByItsOwnIterate covers a C++ host with its own Iterate, Len 3 and 2 iterated items: the loop, loop.length, last, + and *, narrow and wide.
• python/tests/test_values.py::test_list_and_tuple_subclasses_iterate_as_python_does has 8 rows. They cover __iter__ and __getitem__ overrides, a plain subclass, a non-sequence with __len__, the loop attributes and the operators.
  • The contexts are built fresh for each engine, because render_both's deepcopy rebuilds a list subclass from its __iter__ and would hide the case (verifier F1).
  • With IteratesByIndex forced true, 4 of the first 5 rows fail.
• Debug build: ctest 8/8 passes. Python module (3.12, limited API): full python/tests gives 1629 passed, 17 skipped, 2 xfailed.
• Verifier (round 1, before the Codex fixes): ASan+UBSan clang build, PublicHostObject* 47/47 with no reports. The module under ASan had no sanitizer reports across a 90-template edge matrix.
• scripts/preflight.sh --perf passes. The instruction gate shows at most +0.06%, on mitsuhiko_table.

Remaining limits

Task 0453: l|pprint of such a subclass prints its stored items, while Jinja2's pprint iterates it. It printed the same way before this PR. The task was first numbered 0452, but another thread took that number at the same time.

Roles: verifier, 1 round (F1 and F3 fixed, F2 and F4 first filed, then fixed after Codex). Red CI pushes: none so far.
Codex: 2 findings (P2), 2 fixed, 0 declined. Security review: not needed. The diff touches the host-object bridge and the loop's length source, not the lexer, parsers, loaders, escaping or the sandbox.

🤖 Generated with Claude Code

https://claude.ai/code/session_01AWNL7mqnaXzqZnVND5kuSG

[rmorozov]

@codex review

---

Generated by Claude Code

[chatgpt-codex-connector]

Codex Review Summary: ✅ Completed

This comment shows the latest Codex review activity on this pull request.

Review	Status	Commit	Review trigger
📝 Code Review	✅ Completed 2026-11-10T22:31:24.498763Z	c455b69	Manual request

ℹ️ About Codex in GitHub

Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you

• Open a pull request for review
• Mark a draft as ready
• Comment "@codex review" or "@codex security review".

Codex reacts with 👀 while any review is running, comments if it has suggestions, and reacts with 👍 once all reviews finish with no findings.

[chatgpt-codex-connector]

💡 Codex Review

Here are some automated review suggestions for this pull request.

Reviewed commit: 330371990d

ℹ️ About Codex in GitHub

Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you

• Open a pull request for review
• Mark a draft as ready
• Comment "@codex review".

If Codex has suggestions, it will comment; otherwise it will react with 👍.

Codex can also answer questions or update the PR. Try commenting "@codex address that feedback".

[rmorozov]

@codex review

---

Generated by Claude Code

[chatgpt-codex-connector]

Codex Review: Didn't find any major issues. Swish!

Reviewed commit: c455b69b2c

ℹ️ About Codex in GitHub

Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you

• Open a pull request for review
• Mark a draft as ready
• Comment "@codex review".

If Codex has suggestions, it will comment; otherwise it will react with 👍.

Codex can also answer questions or update the PR. Try commenting "@codex address that feedback".