[ScriptedAlchemy]

Merge #4478 first, followed by #4526, #4510, #4524, #4489, #4485, #4527, #4528, #4530, #4531, #4529, #4534, and #4541.

This PR is number 10 of 13, based on #4530. Merge #4529 next.

Use merge commits. After merging a parent PR, change its child's base to master before merging the child. Auto-merge is disabled.

Fixes #4420. Stack parent: #4530. Merge after the existing ordered stack using a merge commit.

Problem

insert_imports advanced one physical line per use and treated inner and outer attributes alike. A destination with a multi-line use std::{ ... }; got the new import spliced inside the group, and a destination starting with #![allow(dead_code)] got the import ahead of the inner attribute. Both produce invalid Rust.

Root cause

Insertion position came from a per-line prefix classifier (classify_leading_line), not from the Rust parse the dependency analyzer already uses.

Fix

• The insertion byte boundary uses the existing canonical Rust parser: after module inner attributes, inner docs, leading comments and existing use/extern crate declarations, never inside one. Outer docs and attributes stay attached to the item they annotate.
• A destination that has syntax errors (or no grammar) refuses the move with a typed error instead of guessing. Empty and missing destinations behave as before.
• The line classifier and LeadingKind are deleted; the only remaining caller (inner-doc skip) uses is_inner_doc_line.

Verification

Red first (commit 1): move_symbol_inserts_imports_after_a_multiline_use_group, ..._after_module_inner_attributes and move_symbol_refuses_a_destination_that_does_not_parse failed (the first two with rustc errors on the generated destination). With the fix, bazel test //crates/tracedecay-source-edit:unit_test passes (all tests) and bazel build --config=clippy //crates/tracedecay-source-edit/... is clean.

🤖 Generated with Claude Code

Independent boundary regression and correction

An existing import and an outer attribute may share a physical line. Rounding the import's syntax end to the next line detached #[cfg(any())] from the following item and applied it to the newly inserted import. The real MoveSymbol fixture then failed Rust compilation because the moved function's Widget import was disabled.

The insertion uses the parsed node's byte boundary, preserves the destination text around it, and inserts a newline where necessary. The new regression executes the existing preview/publication journey and compiles the resulting fixture.

• BEFORE: 0 passed / 1 failed; invocation 6fe80e3b-a163-429d-b651-e981eddd13c0, missing Widget after the move.
• AFTER exact: 1 passed / 114 filtered; invocation eca65863-6a1e-4bff-9e79-7cb0d30b880d; affected formatting passed.
• Full source-edit unit suite: 115 passed / 0 failed; invocation 7e338640-af8b-4361-926b-50d59a944904.

All runs used canonical Bazel targets with fresh test results. No timeout, budget, or host setting changed.

Scoped Clippy and final formatting passed in 8420704d-e70a-481c-b928-2d1a34aac5ef. The first Clippy run caught the now-unused splice_lines import; removing that import was the only change after the 115-test pass.

Stack integration

The parent stack already contains canonical byte-based destination rendering, typed source-edit-destination-syntax-unsupported refusal, validation for dependency-free Rust moves, and newline preservation. Integration retains that implementation rather than duplicating the parser. This layer keeps the real execution regressions and removes the obsolete line classifier in favor of the remaining inner-doc predicate. No destination validation or existing regression is dropped.

Integrated parent validation: 121 source-edit unit tests passed, with formatting, in c759f31b-2115-407e-aa85-1ffe9eb18855. This includes the parent destination tests and every new execution regression.

Integrated scoped Clippy passed in 2e8d67bd-73a9-4027-a613-d1ccadf75a10. Parent integration uses a merge commit and preserves the canonical renderer byte-for-byte.

Import binding and destination-scope follow-up

Physical-line matching treated import text in comments and nested scopes as a module binding, while the text alias splitter missed valid multiline/commented as clauses. Import discovery and destination matching now use the maintained Rust grammar's path and alias nodes. Equivalent grouped or multiline imports are reused, and aliases keep their original meaning and bytes. Existing public exports and relative-source import safeguards remain intact.

Automatic insertion now returns the existing typed refusal before publication when it would overlap a conditional or unknown attributed import, a different known binding or item name, or an unresolved wildcard binding. Exact unconditional matches and legitimate self-import suppression remain supported; anonymous as _ imports remain distinct from wildcards. These checks use syntax-visible declarations and do not perform macro expansion or full namespace resolution.

The existing preview/apply fixture was extended for commented/nested imports, alias trivia, equivalent grouped paths, alias self-imports, manual dependency hints, and conservative refusal with unchanged files and no edit journal. A focused native check used the exact committed import-binding helper and pinned Rust grammar: the original helper failed the multiline-alias case (0 passed / 1 failed), while the repaired helper passed both alias and grouped/unnamed-binding checks (2 passed / 0 failed / 0 ignored). This helper check does not execute the full source-edit application journeys. Pinned formatting, diff checks, and normal hooks passed for both commits; the complete native suite remains a hosted CI requirement.

Unconditional lint-attribute follow-up

The destination availability check now recognizes direct, unqualified built-in lint attributes (allow, expect, warn, deny, and forbid) with parsed parenthesized arguments. These attributes do not disable the existing import, so a matching binding is reused with its original bytes and lint policy intact. Every attached attribute must qualify. Conditional, qualified, procedural, unknown, and documentation attributes retain the conservative refusal; source relocation still treats all attached attributes as owned syntax.

Actual preview/apply fixtures now cover a matching allow(unused_imports) import and a grouped public alias under multiple lint attributes. Additional refusal cases combine lint attributes with cfg_attr, a qualified attribute, or a procedural attribute. Existing assertions, compiler warning expectations, and mutation boundaries are preserved.

Pinned formatting, whitespace checks, normal pre-commit/commit-msg hooks, and two independent source reviews passed. A focused native check compiled the exact committed attribute helpers with the pinned Rust grammar and nightly compiler. The original rule reproduced the allow(unused_imports) failure (0 passed / 1 failed); the repaired helpers passed both checks (2 passed / 0 failed / 0 ignored), including all five lint names, trivia, strict source attachment ownership, and refusal of active, unknown, documentation, or malformed attribute forms. Full preview/apply application journeys remain a hosted CI requirement.

[changeset-bot]

⚠️ No Changeset found

Latest commit: 0b6e8b5

Merging this PR will not cause a version bump for any packages. If these changes should not result in a new version, you're good to go. If these changes should result in a version bump, you need to add a changeset.

Click here to learn what changesets are, and how to add one.

Click here if you're a maintainer who wants to add a changeset to this PR

[devin-ai-integration]

Note

Newer findings are available below. Devin Review posted a newer report on this PR, in addition to the findings presented here.

🔍 Devin Review: 1 flag

Not posted on this PR by your GitHub settings — view it in Devin Review. (Configure)

[ScriptedAlchemy]

Included in the integration PR #4589, which runs CI once for all open PRs. This PR will be marked merged automatically when #4589 lands.