[johnnyreilly]

First small step in trimming devDependencies.

Removed

• babel: old v6 placeholder package (it only tells you to install babel-cli). Nothing at the root requires it, and the babel execution tests that use it list it in their own package.json
• lint-staged: it was configured in package.json, but nothing installs a git hook to run it (no husky or similar), so it never ran. CI already runs npm run format:check and npm run lint
• semver: only used by test/execution-tests/run-tests.js to skip tests whose folder name starts with a minimum TypeScript/webpack version (e.g. 2.1.4_, 4.7.0_5.0.0_). It's replaced by a small isLowerVersion helper that compares major.minor.patch. One deliberate difference: prerelease tags are ignored, so 7.1.0-beta counts as equal to 7.1.0 (semver calls it lower). No test prefix is anywhere near 7.x, so this changes nothing today
• glob: only used by test/comparison-tests/create-and-execute-test.js to list every file in a directory. It's replaced by a documented listFiles(dir, { includeDotFiles }) helper that walks directories with fs.readdirSync(dir, { withFileTypes: true }). It returns /-separated paths (also on Windows), keeps glob's sort order, returns an empty list for a missing directory like glob did, and keeps the one call that skipped dotfiles (cleanHashFromOutput) skipping them (without walking into dot-folders). Deliberate differences: symlinked folders aren't followed (Node's recursive readdirSync follows them, so a link back to a parent listed the same files over and over), and a broken symlink is skipped; any other error, such as permissions, is thrown so --save-output can't delete a snapshot it couldn't read. glob is still installed indirectly via karma-webpack; we just no longer depend on it directly
• markdown-toc: only used by the generate-toc script, which is also removed. The README table of contents was already maintained by hand: Prettier has rewritten its bullets, so re-running the generator would fight Prettier. Also fixed the one stale TOC entry ("Declarations (.d.ts)" -> "Declaration Files (.d.ts)", matching its heading). This removes 573 lines from package-lock.json
• webpack-cli: only used by test/execution-tests/run-tests.js to run webpack --bail for the one execution test without a karma config (5.6.2_5.0.0_optionsCaching). A new test/execution-tests/run-webpack.js script does the same through webpack's Node API: like webpack-cli, it accepts a config that exports an object, a function or a Promise, or an array of them; config functions get webpack-cli's env and argv (including bail: true), and one that throws is reported through the script's own error handling; it builds once with bail: true, prints the stats using each config's own stats setting (colors only when stdout is a terminal, unless the config says otherwise), and exits with 1 on errors. run-tests.js runs it with execFileSync, so there's no shell quoting. --debug still works via node --inspect-brk. test/comparison-tests/README.md now suggests npx webpack-cli for building a test by hand. This removes 39 packages from node_modules

Also removed their entries from knip.json (ignoreDependencies), the lint-staged config block and the generate-toc script from package.json.

Upgraded

• mocha 6 → 12. This removes about 1,080 lines from package-lock.json. Only the comparison test runner uses mocha, and it needed three changes:
  • test/comparison-tests/run-tests.js passes --color instead of -c. Mocha 6 restarted itself with --color in process.argv, which is how picocolors decides to enable colors; mocha 12 passes -c through unchanged, so the colored output snapshots stopped matching
  • It also passes --test-to-run=<name> instead of --test-to-run <name>. Mocha 12 parses arguments with util.parseArgs and treated the separate test name as a spec file, printing a warning for every test
  • test/comparison-tests/create-and-execute-test.js now reads its arguments with util.parseArgs (non-strict, since mocha's own arguments are mixed in). It accepts both forms of --test-to-run, and fails with a clear error if the test name is missing, empty, or looks like a flag (meaning parseArgs took the next flag as the value) (before, it silently used the node binary's path)
• .claude/settings.json: the allowed mocha commands now point at node_modules/mocha/bin/mocha.js, as mocha 12 removed bin/_mocha, and pass the test name as --test-to-run=*

Other cleanup

• Removed the root Dockerfile and the docker:build / postdocker:build scripts. The image was built on node:18 (ts-loader needs Node 22+, mocha 12 needs 20.19+) with outdated Chrome install steps (apt-key, libgconf-2-4), and nothing else used it. The test-probe workflow and CI already cover running the tests on Linux
• Also removed .dockerignore (only used by that Dockerfile) and the Dockerfile / .dockerignore entries in .npmignore. The published package is unchanged (npm pack --dry-run still lists 44 files)
• Removed .devcontainer/ (Dockerfile and devcontainer.json). It was based on an old Node 14 dev container template, well below the Node 22+ that ts-loader needs. Its .npmignore entry is gone too
• Removed the unused --extra-option argument from test/comparison-tests/create-and-execute-test.js. Nothing passed it: run-tests.js only sends --test-to-run and --save-output
• var → const/let in test/execution-tests/run-tests.js and run-webpack.js, and dropped a commented-out Karma require
• Callbacks in test/comparison-tests/create-and-execute-test.js are now arrow functions. Named functions stay as declarations, and the test function createTest returns stays a function because it uses mocha's this.timeout
• New npm run check:comparison-tests and npm run check:execution-tests scripts type-check the test harness scripts with tsc. The comparison tests' tsconfig.json now also covers link-test-lib.js, and the execution tests get their own tsconfig.json (run-tests.js, run-webpack.js). To pass, run-webpack.js types the webpack callback's stats as Stats or MultiStats and treats missing stats as a failed build. The lint workflow (.github/workflows/lint.yml) now runs both checks after npm run lint, so type errors in the harness scripts fail CI
• The two harness tsconfigs had identical compiler options, so they now extend a shared test/tsconfig.base.json (with lib: es2022, matching src). The execution tests check now also covers test/run-tests.js and test/karmaConfig.js, which nothing type-checked before
• Removed the @ts-ignore comments from the comparison test harness. A new test/comparison-tests/globals.d.ts declares the mocha globals it uses (describe, it, and the test's this.timeout and done), since mocha 12 ships no types; this avoids adding @types/mocha. create-and-execute-test.js now checks for webpack() returning null (as its types allow) and fails the test with a clear error, and newline.loader.js types its this as webpack's LoaderContext
• AGENTS.md lists the three check: scripts, and no longer says lint has no separate type-check script
• New npm run check:benchmark-tests, also run by the lint workflow, as nothing type-checked the benchmark .mts files. Dropped esModuleInterop from their tsconfig, as TypeScript 7 always enables it
• src/tsconfig.json now uses "module": "node20" (which implies moduleResolution: node16) instead of module: commonjs + moduleResolution: bundler, so imports are checked with Node's resolution rules, which is where ts-loader runs. Also dropped declarationDir, which matched outDir. The build output is byte-identical: building the old and new configs side by side gives the same 36 files, including .d.ts and source maps

Kept

The first version of this PR also removed babel-polyfill and babel-preset-es2016, which broke CI. The babel execution tests do run (the isNotBabelTest filter is commented out), and some of them load these packages from the root install:

• 2.1.4_babel-es2016 and 2.1.4_babel-react use the es2016 preset
• 2.4.1_importCodeSplitting imports babel-polyfill

Both are restored with their original version ranges. babel-core, babel-loader, babel-preset-es2015 and babel-preset-react stay too, because the comparison and execution tests use them.

Testing

• npm run knip, npm run format:check, npm run check:benchmark-tests, npm run check:comparison-tests and npm run check:execution-tests pass
• npm run execution-tests: all 27 suites pass locally (webpack 5), including after the var → const change
• npm run comparison-tests: all 57 suites pass locally (with mocha 12)
• --save-output on basic runs cleanly and leaves no changed snapshots
• With mocha 12: the full comparison suite, --save-output, the old --test-to-run <name> form (used by the allowed Claude commands) and a missing test name (clear error) all behave as expected
• run-webpack.js: 5.6.2_5.0.0_optionsCaching passes. For each config shape (object, array, function, Promise), a working build writes its bundles and exits 0, and a broken build exits 1. The config's stats setting is respected: 'errors-only' prints nothing on success and still shows errors, an array uses each config's own setting, and colors are off when output isn't a terminal unless the config forces them. That test is skipped on webpack 4, so the webpack 4 CI jobs don't run the script
• run-webpack.js config functions: an array of functions (including async) builds every entry, a function that throws exits 1 through the script's error handling, and argv.bail is true
• An empty --test-to-run= and --test-to-run --save-output both fail with the clear error message
• listFiles returns exactly what glob.sync did (same files, same order) for all 729 folders under test/comparison-tests, with and without dotfiles, plus the missing-folder case. A folder with a loop -> . symlink gives each file once; a broken symlink is skipped; a symlink into an unreadable folder throws EACCES. The Windows path handling is only covered by CI's Windows comparison-test job
• isLowerVersion gives the same answers as semver.lt for every test prefix against the TypeScript and webpack versions CI uses, apart from the prerelease case above
• Simulated webpack 4: exactly the two _5.0.0_ tests are skipped (4.7.0_5.0.0_node16-file-suffix, 5.6.2_5.0.0_optionsCaching). On webpack 5, none are skipped

🤖 Generated with Claude Code

[github-actions]

Benchmark (Ubuntu)

Scenario	transpileOnly	PR branch median (ms)	base branch median (ms)	Δ vs base branch
Cold build	false	921.7	914.3	+0.8%
Incremental rebuild (leaf touch)	false	46.1	47.1	-2.0%
Incremental rebuild (hub touch)	false	357.7	355.7	+0.6%
Cold build	true	306.9	302.4	+1.5%
Incremental rebuild (leaf touch)	true	28.3	27.6	+2.4%
Incremental rebuild (hub touch)	true	28.4	28.8	-1.4%
Cold build (all orphan files)	false	979.2	968.3	+1.1%
Incremental rebuild (hub touch, all orphan files)	false	393.5	382.6	+2.8%

PR branch = /home/runner/work/ts-loader/ts-loader, base branch = /home/runner/work/ts-loader/ts-loader-main. 2 warmup + 10 measured iterations per scenario, median reported. Report-only - no threshold fails this check.

[github-actions]

Benchmark (macOS)

Scenario	transpileOnly	PR branch median (ms)	base branch median (ms)	Δ vs base branch
Cold build	false	822.9	821.2	+0.2%
Incremental rebuild (leaf touch)	false	50.3	50.3	-0.1%
Incremental rebuild (hub touch)	false	331.8	331.1	+0.2%
Cold build	true	222.8	225.3	-1.1%
Incremental rebuild (leaf touch)	true	50.1	49.3	+1.7%
Incremental rebuild (hub touch)	true	49.8	49.8	+0.1%
Cold build (all orphan files)	false	944.9	920.7	+2.6%
Incremental rebuild (hub touch, all orphan files)	false	387.4	406.7	-4.7%

PR branch = /Users/runner/work/ts-loader/ts-loader, base branch = /Users/runner/work/ts-loader/ts-loader-main. 2 warmup + 10 measured iterations per scenario, median reported. Report-only - no threshold fails this check.

[github-actions]

Benchmark (Windows)

Scenario	transpileOnly	PR branch median (ms)	base branch median (ms)	Δ vs base branch
Cold build	false	1822.5	1822.5	+0.0%
Incremental rebuild (leaf touch)	false	64.3	64.1	+0.2%
Incremental rebuild (hub touch)	false	751.8	761.8	-1.3%
Cold build	true	490.2	494.2	-0.8%
Incremental rebuild (leaf touch)	true	37.6	36.6	+2.9%
Incremental rebuild (hub touch)	true	44.0	38.6	+13.9%
Cold build (all orphan files)	false	1995.4	1980.5	+0.8%
Incremental rebuild (hub touch, all orphan files)	false	870.3	882.9	-1.4%

PR branch = C:\source\ts-loader, base branch = C:\source\ts-loader-main. 2 warmup + 10 measured iterations per scenario, median reported. Report-only - no threshold fails this check.

[Copilot]

🟡 Changes recommended

The test-name guard permits empty values that clear the entire staging root, alongside unresolved symlink traversal, erroneous snapshot deletion, and missing config-factory bail semantics.

4 open issues

• Reject empty or option-like test suite names · New
• Prevent recursive traversal of directory symlinks · New
• Propagate non-ENOENT errors when checking output files · New
• Pass bail to the config factory argv · New

What changed in this PR

This PR reduces ts-loader’s development-tooling footprint, updates its test harnesses for Mocha 12, and expands type checking in CI.

Changes:

• Trim root dependencies, unused scripts, and obsolete Docker/dev-container setups.
• Upgrade Mocha and replace glob, semver, and webpack-cli usage with local helpers.
• Add harness type checks, share compiler options, and use Node-aware resolution for source imports.

File	Description
test/​tsconfig.base.json	Shares JavaScript harness-check options.
test/​execution-tests/​tsconfig.json	Adds execution and shared harness type checking.
test/​execution-tests/​run-webpack.js	Adds a one-shot webpack Node API runner.
test/​execution-tests/​run-tests.js	Replaces semver and invokes the new webpack runner.
test/​comparison-tests/​tsconfig.json	Shares options and expands checked files.
test/​comparison-tests/​run-tests.js	Updates Mocha color and test-selection arguments.
test/​comparison-tests/​README.md	Documents manual builds using npx webpack-cli.
test/​comparison-tests/​newline.loader.js	Types the webpack loader context.
test/​comparison-tests/​globals.d.ts	Declares the Mocha globals used by the harness.
test/​comparison-tests/​create-and-execute-test.js	Replaces glob and adds argument parsing and compiler checks.
test/​benchmark-tests/​tsconfig.json	Removes the redundant interop option.
src/​tsconfig.json	Uses Node-aware module resolution.
README.md	Corrects the declaration-files TOC label.
package.json	Trims dependencies/scripts and adds harness checks.
package-lock.json	Prunes packages and refreshes tool resolutions.
knip.json	Removes exclusions for deleted dependencies.
Dockerfile	Deletes the obsolete root test image.
.npmignore	Removes exclusions for deleted container files.
.github/​workflows/​lint.yml	Runs all three harness type checks in CI.
.dockerignore	Deletes the unused Docker exclusions.
.devcontainer/​Dockerfile	Removes the outdated development image.
.devcontainer/​devcontainer.json	Removes the obsolete dev-container configuration.
.claude/​settings.json	Updates allowed commands for Mocha’s new CLI.

_{Review effort: Max}