Code Analysis Default References

Stable default function references

Draft validated, not Labs: #1260, runtime e6778b531, evidence/package head 80875e5e75d3a9e1a91a50a51a56a0315550a19f. Now the lower layer of actual native stack #1262, followed by configured-alias #1261, based on preserved #1252; original team commits/authors/PR ownership retained. Java #1259 and Go/Rust #1256/#1258 remain separate. This is this JS branch's package, not an integrated all-language release. No approval, merge, release or live installation.

One more evidenced source step

Before, Prettier main/comments/print.js:126 and attach.js:139 had resolved imports but no function targets. After, the normal CLI follows isNonEmptyArray into utils/is-non-empty-array.js:5, where a function is exposed by export default isNonEmptyArray at line9. The external Array.isArray implementation and caller-supplied filter(comment) remain unknown source targets.

A real supplemental change investigation follows Prettier 3.3.3 → 3.4.2: loadExternalConfig gained actual required.__esModule/default-return handling. The unchanged production caller reaches its changed definition at load-config.js:40 →load-external-config.js:9 before, and :39 →:10 after. Parent captures zero references; the candidate finds the single independently source-verified caller on both sides with valid navigation. Both paired results remain partial / exit 1 because package/config inputs differ; this is not equivalent-build/runtime/complete-impact evidence.

Fixed sources, replay and finite score

• Prettier 3.4.2 cca946176c3ec04ae46bf7bcb08c5bba8c041682, analyzer root src. Original comments-path source answers/Node controls were committed before implementation.
• Prettier 3.3.3 52829385bcc4d785e58ae2602c0b098a643523c9, same root; supplemental pair selected source-first after initial head implementation, before its base/paired capture. Not the original freeze or blind review.
• Chalk v5.3.0 72c742d4716b1f94bb24bbda86d96fbb247ca646, analyzer root repository. Reserved negative source review after initial code but before its first capture, not blind. test/chalk.js:12 default callable is factory-created (source/index.js:204/225), not a direct ordinary function declaration; no rule tuning from this slice.

Full report · source-first manifest · public CLI replay · source/ZIP parity · wider signature audit.

Original fixed observations	Correct	Wrong	Missed known targets	Expected unknowns
Parent normal CLI	0	0	2	2
Candidate source / ZIP	2	0	0	2

The supplemental real pair is reported separately: 0→1 captured caller per side. These finite observations do not measure general accuracy or developer productivity. The first replay reused head line39 on the base; frozen per-side source evidence already said base40/head39. Final replay derives positions from those records. The correction is retained; no analyzer change for it.

Model, coverage and unknowns

Producer js-ts-lexical-v3, adapter revision 20. It follows a captured relative default import through same-module default identifier expressions/local default export specifiers into ordinary function declarations with one body, rejecting observed binding writes or direct eval. Compiler symbol identity preserves unrelated shadows/property writes. Current readers retain v1/v2 coverage without relabeling old captures.

Factories, variable/const-initializer defaults, arbitrary expressions, re-export chains, dynamic/member dispatch and complete impact remain unsupported. Any observed source write may conservatively suppress a valid snapshot even if later/unexecuted. A supplemental JS expando-property control corrected an overly conservative rejection; independent Node ESM execution confirms the function remains original. Removing the write/eval guard makes the overwrite regression fail. Authored controls execute only tiny fixtures, never upstream applications.

Actual source profile: 351 selected files for each Prettier/src version; 15 for Chalk, zero parse errors. No root tsconfig at those analyzer roots, so allowJs/NodeNext/ESNext fallback applies; ancestor Prettier tsconfig is intentionally not discovered. Resolver reads ../package.json for Prettier and package.json for Chalk, with hashes/missing probes retained. Product program uses captured roots/noResolve/noLib/types[], not the upstream configured build. Source scope does not equal project build/test inclusion. No new runtime dependency/toolchain/language server or project/config execution.

Supplemental full-resolution/checkJs TypeScript 5.9.3 signature audit agrees with all 418 newly emitted Prettier targets, zero disagree/unavailable. It reports 236 diagnostics (missing dependencies and JSDoc/type errors), not successful compilation or runtime identity. All 6,926 / 539 call ranges and all 1,103 / 44 previously bound Prettier/Chalk records are preserved; functions/imports unchanged. This checks additions/preservation, not complete recall; the Redux non-relative alias gap remains unresolved.

Verification layers

Layer	Evidence
Local	348 pass / one inherited TS installed-configuration assertion / zero skipped. Exact parent reproduces same assertion (10/1). Fifty focused checks pass.
Source / ZIP CLI	Two checks and two saved questions exit 0; real paired question partial/exit 1. Two facts files byte-identical; three questions agree after installation/output/workflow path normalization.
Package	Two canonical builds equal; 156 tracked module files match. Packaged imported/question/change selection 41 pass. ZIP SHA256 c42bbe3ca39ebb48fe772aa41b144674ef2dc8e350daf965a874185308350488.
Remote CI	Final head 80875e5e7 run 42397027606 completed: Code Analysis 349/349 pass, full regression shard 1 passes. Overall fails only on inherited README Classic artifact drift in shard 2 (same expected ca273484… / actual 7f68d730… hashes) and test(22) aggregate; all other executed jobs pass, deployments skipped.
Integration	Draft only; no approval/merge/release/live installation.

Reproduce from the linked report with fixed checkouts prettier, prettier-base and chalk. run.py supports preserved parent93378351d via --expect baseline and either source/extracted-package CLI. The Node control replay preserves its original frozen oracle, writing only a requested output file. Next use actual wrong targets/missed references/investigation interruptions across languages; retain all unknown/build-scope boundaries.