fix(gc): classify boxed string concat as poll-capable

[proggeramlug]

Closes #7872.

Summary

• add js_string_concat_box to POLL_CAPABLE_RUNTIME, so every --moving-only checker arm sees the helper's path through js_dynamic_string_or_number_add to coercion/user code
• extend the planted poll-reach self-test with that exact two-hop production edge

Reproduction

On the pre-fix main checkout:

$ python3 scripts/gc_root_dominance_check.py --audit-poll-reach
error: ALLOC_RE symbols that CALL a POLL_CAPABLE_RUNTIME symbol but are not in POLL_CAPABLE_RUNTIME:
  js_string_concat_box                             -> js_dynamic_string_or_number_add
AUDIT_EXIT=2

After this change, the same audit reports 3,791 exported symbols, 1,616 intra-runtime call edges, 384 ALLOC_RE matches, and no unlisted poll-reaching allocation wrapper.

Validation

• python3 scripts/gc_root_dominance_check.py --self-test
• all four static audits (--audit-alloc-re, --audit-poll-capable, --audit-poll-reach, --audit-immovable-sources)
• python3 scripts/gc_gate_wiring_check.py
• python3 scripts/check_test_registration.py
• shadow corpus: 139/139 sources, 161 IR modules, 0 skips
  • root-store dominance: 0 violations; 40/40 planted violations caught
  • unrooted allocas: 0
  • stale registers: 18, within the pinned 39 budget
• native corpus: 139/139 sources, 161 IR modules, 0 skips, 35,246 emitted statepoints / 21,290 non-empty live bundles
  • checked 35,085 safepoints and 50,900 relocations
  • 0 unrooted, 0 stale; 40/40 planted statepoint violations caught
• exact before/after classification over the curated corpus found 27 newly poll-reaching compiled functions but 0 newly exposed stale windows
• bash scripts/check_file_size.sh
• git diff --check

Additional observation, not changed here: the dependency-scale Zod corpus emitted all 81 expected IR modules but its final link failed on unresolved intra-Zod symbols, identically with Perry's cache disabled. Running the checker directly over those emitted modules still met the normal floors (12,903 functions), reported 0 dominance violations, 0 unrooted allocas, and 89 stale uses within the 118 budget; the exact classification delta for this list entry was 0 newly exposed stale windows.

Summary by CodeRabbit

• Bug Fixes

  • Improved garbage-collection safety checks for boxed string concatenation when coercion may execute user code.
  • Corrected poll-reachability analysis across indirect numeric coercion paths.
  • Added coverage for exact two-step coercion scenarios to improve audit accuracy and validation.

• Documentation

  • Added a changelog entry describing the corrected moving-only audit behavior and verification results.

[coderabbitai]

No actionable comments were generated in the recent review. 🎉

ℹ️ Recent review info

⚙️ Run configuration

Configuration used: defaults

Review profile: CHILL

Plan: Pro Plus

Run ID: 59d513e9-7843-4e19-a25f-48667ab6f86f

📥 Commits

Reviewing files that changed from the base of the PR and between 23f09a0 and 6e55f06.

📒 Files selected for processing (1)

• changelog.d/7896-boxed-concat-poll-reach.md

🚧 Files skipped from review as they are similar to previous changes (1)

• changelog.d/7896-boxed-concat-poll-reach.md

---

📝 Walkthrough

Walkthrough

The GC root-dominance checker classifies js_string_concat_box as poll-capable, models its coercion path through js_number_coerce, and verifies the result in its self-test. A changelog entry documents the correction.

Changes

GC poll-reach classification

Layer / File(s)	Summary
Poll-capable classification and reach fixture scripts/gc_root_dominance_check.py	Adds js_string_concat_box to POLL_CAPABLE_RUNTIME and models its two-hop path through js_dynamic_string_or_number_add to js_number_coerce.
Self-test and changelog scripts/gc_root_dominance_check.py, changelog.d/7896-boxed-concat-poll-reach.md	Updates expected poll-reach results and documents the corrected boxed-concatenation audit classification.

Estimated code review effort: 2 (Simple) | ~10 minutes

Possibly related PRs

• PerryTS/perry#7835: Changed js_string_concat_box to use dynamic coercion for non-string operands.
• PerryTS/perry#7679: Expanded and validated POLL_CAPABLE_RUNTIME poll-reach classification.
• PerryTS/perry#7284: Updated POLL_CAPABLE_RUNTIME and related checker self-tests.

🚥 Pre-merge checks | ✅ 4 | ❌ 1

❌ Failed checks (1 warning)

Check name	Status	Explanation	Resolution
Docstring Coverage	⚠️ Warning	Docstring coverage is 0.00% which is insufficient. The required threshold is 80.00%.	Write docstrings for the functions missing them to satisfy the coverage threshold.

✅ Passed checks (4 passed)

Check name	Status	Explanation
Title check	✅ Passed	The title clearly identifies the primary change: classifying boxed string concatenation as poll-capable for GC checks.
Description check	✅ Passed	The description explains the change, linked issue, reproduction, validation commands, audit results, and corpus findings; missing optional template sections are non-critical.
Linked Issues check	✅ Passed	The PR satisfies issue #7872 by updating poll-capable classification, adding the focused self-test, passing the audit, and reviewing corpus results.
Out of Scope Changes check	✅ Passed	The changes are limited to the poll-capable runtime classification, its focused self-test, and related changelog documentation.

✨ Finishing Touches

🧪 Generate unit tests (beta)

• Create PR with unit tests
• Commit unit tests in branch fix/7872-poll-capable-concat

---

Thanks for using CodeRabbit! It's free for OSS, and your support helps us grow. If you like it, consider giving us a shout-out.

❤️ Share

• X
• Mastodon
• Reddit
• LinkedIn

_{Comment @coderabbitai help to get the list of available commands.}

[proggeramlug]

@coderabbitai review

[coderabbitai]

⚠️ Action not completed

Already reviewed.

Note: CodeRabbit is an incremental review system and does not re-review already reviewed commits. This command is applicable only when automatic reviews are paused.

[coderabbitai]

Actionable comments posted: 1

🤖 Prompt for all review comments with AI agents

Verify each finding against current code. Fix only still-valid issues, skip the
rest with a brief reason, keep changes minimal, and validate.

Inline comments:
In `@changelog.d/7896-boxed-concat-poll-reach.md`:
- Around line 1-3: Expand the changelog fragment to include concise context
referencing Issue `#7872`, plus the completed GC root-dominance moving-only audits
and corpus checks from the PR objective. Preserve the existing behavior summary
and record the validation results as history and validation notes.

🪄 Autofix

Fix all unresolved CodeRabbit comments on this PR:

• Push a commit to this branch (recommended)
• Create a new PR with the fixes

---

ℹ️ Review info

⚙️ Run configuration

Configuration used: defaults

Review profile: CHILL

Plan: Pro Plus

Run ID: 5306b7a8-ee26-4944-9e5d-3ac47e3e2d27

📥 Commits

Reviewing files that changed from the base of the PR and between c5a0341 and 23f09a0.

📒 Files selected for processing (2)

• changelog.d/7896-boxed-concat-poll-reach.md
• scripts/gc_root_dominance_check.py