The expression choice offered binary_expr last, so a block heading a && chain could not parse

[briansrls]

An ordering fact, not a missing form

dag_grammar_expr_expr is an ordered Choice that offered match_expr, if_expr and loop_expr before binary_expr. But dag_grammar_primary_expr_core's final three alternatives already are those same productions — a block-shaped expression is a legal binary operand by construction.

The ordering denied it. Given match a { ... } && b, the choice committed to match_expr, which succeeded having consumed the match alone, and && b reached no production: the parse ended with tokens remaining. Nothing was unrepresentable; the alternative that represents it was never offered.

A discriminating pair, against the live parser

New witness v2.test.parse.block_expr_as_binary_operand_parse runs source text through dag_prepared_grammar — the grammar under judgment, not a model of it.

arm	before	after
match a {…} && true	FAIL	PASS
if a {…} else {…} && true	FAIL	PASS
bare match · bare if · a && true	PASS	PASS

The three controls are what make this a repair rather than a trade: a change that satisfied the reds by making block forms parse only as operands would take them red.

The population

Whole-corpus census, same native binary both sides:

	before	after
total file refusals	829	650
at parse	673	480
at later stages	156	170

193 files left the parse stage; 14 of those now fail deeper. Net −179. That signature — the wall moving back rather than vanishing — is what a correct parse repair looks like, and the net is the honest headline, not the parse figure.

What is NOT verified

Tree shape. The controls assert that a bare match/if still parses, not that it yields the same node. A bare match now routes binary_expr → primary_expr_core → match_expr rather than straight to match_expr; if binary_expr with zero operators wraps instead of collapsing to its operand, every bare block form in the corpus changes shape. The census is indirect evidence against it — normalize runs, no new refusals — but it is not a structural assertion. The idiom to close it exists: the sibling expression_bodied_fn_decl_parse_test asserts on the parsed node via find_arrow_body_child. I'd take that before calling this done, and will add it on request.

Deadness of the trailing alternatives. match_expr/if_expr/loop_expr now sit below binary_expr and are unreachable exactly if binary_expr never fails where they would succeed. That argument is not made here, so they are retained rather than deleted, and the annotation says so.

fn_literal, block_expr and arrow_lambda stay above binary_expr, where they already were: primary_expr_core admits a braced field-initializer list, so a block offered below could be read as a record literal.

🤖 Generated with 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 2026-09-20T22:49:27.160489Z	ebef802	PR opened

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.