[rmorozov]

Requested by Ruslan · project thread

Parity batch 3, group E. Closes 0370.

Problem

A JSON object passed with jinja2::ToValue is a dict to the template. Passed with jinja2::Reflect, it became a GenericMapAdapter with the struct policy (KeysFirst, HasAttributes() == true), so the same document rendered differently depending on how the host passed it. Examples: d.items with a key items, attr('name'), map('attr', ...), is sequence. The ParityReflect pass (0322) listed six such lines.

Change

• include/jinja2cpp/json_adapter.h adds an exported marker, jinja2::IJsonNode. It is empty for now. json::ObjectAccessor and json::ArrayAccessor derive from it. This is the marker from the approved 0049 plan (D7 option c). Public header change for G1: it is additive and changes no existing vtable.
• GenericMapAdapter asks IsJsonNode() once, when it is made. For a JSON object it then answers MethodsFirst and HasAttributes() == false. Any other host map, such as a reflected struct, keeps keys-first lookup.
• IsJsonNode() caches its answer by the accessor's most-derived type, in two relaxed atomic slots (one for the last JSON type, one for the last other type). Without the cache, the cross-cast on every lent map item cost up to +19% on reflect renders (verifier round 1).
• Everything that asks those two questions now follows for JSON views. That includes tuple unpacking: {% set x, y = j %} unpacks the keys, which was the third bullet of 0212.

Tests

• The six reflect.* lines naming 0370 are deleted from test/parity/divergences/reflect.txt.
• New DictMethods group in test/binding/conformance.h. It runs on every binding, in all four modes, narrow and wide. Its rows are shadowed_method, attr_filter, tests and unpack_keys, with expectations pasted from conformance_expect.py --cpp under Python 3.12 and Jinja2 3.1.6.
• New ExpressionTest.ReflectedStructFieldComesBeforeADictMethod: a reflected struct keeps its fields before dict methods.

Validation

• Debug build: full ctest 6/6. ParityReflect/* plus *JsonConformance* with JINJA2CPP_PARITY_CONTEXT=reflect: 3115/3115.
• The verifier also ran:
  • Clang address+undefined (RelWithDebInfo): ctest 6/6, reflect pass 3115/3115;
  • a shared-library build (GCC): ctest 6/6; the cast across the .so works;
  • a 76-template driver against Python. No regression; 17 templates became exact Python matches. Host data was never written, and mutating methods still go through the copy-on-write cell.
• scripts/preflight.sh --perf: all PASS, including the instruction gate on the default data.
• bench/count.py --data reflect, head against master: worst case Render/for_filter_if +0.45%. Everything else is at or below +0.38%. Allocation counts are unchanged; bytes grow slightly (one bool per adapter).
• Not run here: the MSVC and Apple Clang builds. The dynamic_cast goes to an exported empty struct, the same pattern as detail::OwnedAccessor. CI checks these platforms.
• Not run: the boost_json binding (the CI bindings job runs the conformance suite on it).

Open

• Decision D6 of the 0212 plan (/mnt/project-files/release-2.0/0212-plan.md): should IJsonNode get the NodeIdentity() virtual before this merges, so the marker never ships empty? This waits for Ruslan.
• A JSON module export now takes the dict path in template_module.cpp (CopyMap) instead of keeping the view. The verifier flagged it for a reviewer's look; the same size caps apply.

Roles: architect (0212 plan), verifier, 1 round. Round 1 had two blocking findings, both fixed: the instruction cost of the cast and an include-cleaner hit. Red CI pushes: none yet. Codex: pending.

---

Generated by Claude Code

[rmorozov]

@codex review

---

Generated by Claude Code

[chatgpt-codex-connector]

Codex Review Summary

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

Review	Status	Commit	Review trigger
📝 Code Review	✅ Completed 2026-11-03T15:23:38.738215Z	c6ead2a	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: c6ead2a26c

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".