cli: list every working loader name in the --loader help text

[robobun]

Problem

• bun --help, bun run --help and bun repl --help describe -l, --loader as: Valid loaders: js, jsx, ts, tsx, json, toml, text, file, wasm, napi.
• That list predates the css, html, yaml, json5, xml, md and sqlite loaders. The flag resolves its value through loader_resolver() (src/runtime/cli/Arguments.rs:38), which is bun_ast::Loader::from_string over LOADER_NAMES (src/ast/loader.rs:73), so all seven are accepted and applied today; the help text is the only in-binary place they are missing.
• The string lives in TRANSPILER_PARAMS_ at src/runtime/cli/Arguments.rs:137.

Fix

• Update the help string to js, jsx, ts, tsx, json, toml, yaml, json5, xml, text, md, css, html, wasm, napi, sqlite, file, the same list (same order) that docs, completions: update the loader name lists for bunfig [loader] and --loader #38266 puts in the --loader docs and shell completions. No behavior change.
• Every added name is accepted by the flag and does what its name says when mapped to an extension: mapping a made-up extension to each one and importing a fixture gives a parsed object for yaml/json5/xml, rendered HTML for md, an HTMLBundle for html, a Database for sqlite, and the same module shape as a real .css file for css. Without the mapping each fixture imports as a path (the file loader), so the flag is what makes them work.
• The list is deliberately the names that work as named, not everything from_string accepts: jsonc and sqlite_embedded collapse to json and sqlite in to_api() (src/options_types/bundle_enums.rs, being fixed in bundler: honor jsonc and sqlite_embedded in loader maps, accept every loader name in Bun.build #38247, after which both belong in this list), base64 and dataurl are unimplemented stubs (bundler: implement dataurl and base64 loaders #36327, runtime + --no-bundle: implement base64/dataurl loaders #36334), sh maps to file, and the aliases (mjs, cjs, mts, cts, node, txt, markdown) stay unlisted as before. The full accepted set is what the invalid loader error prints (cli: list the names --loader accepts in the invalid loader error #38283); the help line is the recommended subset.
• Tests, in test/cli/bun.test.ts:
  • bun --help and bun run --help: the --loader line's Valid loaders: list equals the 17 names above. Fails on the current release (the eight missing names), passes with this change.
  • One bun run that maps a distinct extension to every advertised name and imports a fixture for each of the 15 that are observable at runtime, asserting the loader-specific result; wasm and napi are only mapped (an unknown name fails argument parsing). This ties the advertised list to what the flag accepts.
• bun bd test test/cli/bun.test.ts: 24 pass.

Related

• The same list is hand-copied in docs/snippets/cli/run.mdx, docs/runtime/bunfig.mdx, completions/bun.zsh and completions/bun.bash; docs, completions: update the loader name lists for bunfig [loader] and --loader #38266 updates those copies to this list. This PR carries the binary half because a src/ change needs the help-output test. The Loader type in bun-types is updated separately in bun-types: add json5 and md to the Loader union, document both loaders #38267.
• Once docs, completions: update the loader name lists for bunfig [loader] and --loader #38266 and this land, a source lint (test/internal/source-lints/) asserting the copies equal LOADER_NAMES minus an explicit exclusion list would stop the next drift; it can only pass once both are in one tree, so it is a follow-up rather than part of either PR.
• loader: document the empty --loader extension, accept it in bunfig too #36904 edits this same help line (adds an extensionless example) and will need a one-line rebase.

Background

• A loader is the parser Bun applies to a file based on its extension (json parses to an object, text returns the contents, file returns the path, and so on). --loader .ext:name overrides the extension to loader mapping for the runtime and the bundler.
• LOADER_NAMES is the table of spellings --loader (and bunfig [loader]) accept; to_api() then converts the chosen loader into the API enum the bundler and runtime consume, which is where a few accepted names currently degrade into another loader.

[coderabbitai]

Warning

Review limit reached

@robobun, you've reached your PR review limit, so we couldn't start this review.

Next review available in: 6 seconds

Enable usage-based reviews in Billing to review now. Otherwise, wait until the next included review is available.
You're only billed for reviews past your plan's rate limits ($0.25/file).

How can I continue?

After more reviews become available, a review can be triggered using the @coderabbitai review command as a PR comment. Alternatively, push new commits to this PR.

To avoid repeated limits, reduce automatic review volume by pausing incremental auto-reviews earlier, using label-based review opt-in, excluding WIP or generated PR titles, or requesting reviews manually when the PR is ready. If your team needs uninterrupted high-volume reviews, an organization admin can enable usage-based reviews.

How do review limits work?

CodeRabbit enforces per-developer PR review limits for each organization. Most developers receive the normal plan review availability.

For paid Pro and Pro+ PR reviews, CodeRabbit uses adaptive limits for sustained high-volume activity. When a developer's recent PR review activity reaches the 95th percentile or higher among CodeRabbit users, additional reviews become available more gradually as earlier reviews age out of the rolling window.

Please refer docs for additional details.

Review details

⚙️ Run configuration

Configuration used: Path: .coderabbit.yaml

Review profile: ASSERTIVE

Plan: Pro

Run ID: 3c1f4d14-a8a9-4052-a8a2-51944e249f39

📥 Commits

Reviewing files that changed from the base of the PR and between 6855f52 and 41c0527.

📒 Files selected for processing (2)

• src/runtime/cli/Arguments.rs
• test/cli/bun.test.ts

---

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

[robobun]

Status: ready for review.

Reproduced on the current release (bun --version 1.4.0 canary): bun run --help prints Valid loaders: js, jsx, ts, tsx, json, toml, text, file, wasm, napi for -l, --loader, while bun --loader .l_yaml:yaml --loader .l_json5:json5 --loader .l_xml:xml --loader .l_md:md --loader .l_html:html --loader .l_sqlite:sqlite --loader .l_css:css main.ts accepts all seven unlisted names and applies them (parsed objects, rendered HTML, an HTMLBundle, a Database, a CSS module). The new help-text test in test/cli/bun.test.ts fails on that release with exactly those names missing and passes with this branch.

The src/ diff is the one-line string change. Rebased onto main once to pick up the fixes for the unrelated red checks on the first run (#38280 for the src/sourcemap clippy failure, #38291 for the macOS abort in the astro/vite tests). The docs, completions and types copies of the list are #38266 and #38267; the invalid loader error list is #38283.

[claude]

LGTM — help-text-only change with thorough tests.

What was reviewed:

• Confirmed all 17 advertised names exist in LOADER_NAMES (src/ast/loader.rs:73) and resolve via from_string; the excluded names (jsonc, sqlite_embedded, base64, dataurl, sh, aliases) are justified in the description.
• Checked that this is the only Valid loaders: string in src/ — no sibling copies missed.
• Tests follow harness conventions: tempDir, bunEnv, concurrent pipe draining, await using, stderr/stdout asserted before exit code, test.concurrent for independent subprocesses.

Extended reasoning...

Overview

The PR updates a single string literal in src/runtime/cli/Arguments.rs:137 — the --loader help text in TRANSPILER_PARAMS_ — to list seven additional loader names (yaml, json5, xml, md, css, html, sqlite) that the flag already accepts. It adds ~120 lines of tests to test/cli/bun.test.ts under a new --help describe block: one test asserts the help text matches the expected list for both bun --help and bun run --help, and a second test maps a made-up extension to each advertised loader and imports a fixture to prove the loader is actually applied.

Security risks

None. This is a documentation string in CLI help output; it does not touch parsing, validation, or any code path that handles untrusted input. The test fixtures are self-contained in a temp dir and do not touch the network.

Level of scrutiny

Low. The runtime change is a static help string with no behavioral effect — loader_resolver() already accepts every name being added (verified against LOADER_NAMES at src/ast/loader.rs:73-102). The PR description carefully justifies which accepted-but-degraded names (jsonc, sqlite_embedded, base64, dataurl, sh) and aliases are intentionally omitted, with references to the relevant tracking PRs/issues. A grep confirms this is the only Valid loaders: string in src/, so there are no sibling copies to update in this PR (docs/completions are handled in #38266 as noted).

Other factors

The tests are well above the bar for a help-text change: they tie the advertised list to actual runtime behavior by importing a fixture through each loader and asserting a loader-specific result (parsed object for data formats, HTMLBundle for html, Database instance for sqlite, rendered HTML for md, etc.). Harness conventions are followed — tempDir with using, bunEnv/bunExe, await using on spawned processes, Promise.all over stdout/stderr/exited, stderr asserted before exit code, test.concurrent for independent subprocess tests. No CODEOWNERS cover the touched files. No prior reviewer comments to address.

[robobun]

^{Updated 4:05 AM PT - Aug 14th, 2026}

❌ @robobun, your commit 41c0527 has 1 failures in Build #95552 (All Failures):

• test/bake/deinitialization.test.ts - pid 4488 segmentation fault at address 0x0 on 🪟 2019 x64

---

🧪   To try this PR locally:

bunx bun-pr 38270

That installs a local version of the PR into your bun-38270 executable, so you can run:

bun-38270 --bun

[claude]

LGTM — one-line help-string update bringing the --loader list in sync with what LOADER_NAMES already accepts.

What was reviewed:

• Confirmed all seven added names (yaml, json5, xml, md, css, html, sqlite) are present in LOADER_NAMES at src/ast/loader.rs:73 and reach loader_resolver via from_string.
• Checked the intentionally-omitted names (jsonc, sqlite_embedded, base64, dataurl, sh, aliases) against the table — the PR description's rationale holds.
• Tests follow harness conventions (concurrent pipe drain, tempDir, stderr-before-exitCode) and the help-text assertion fails on the old string.

Extended reasoning...

Overview

The functional change is a single string literal in TRANSPILER_PARAMS_ at src/runtime/cli/Arguments.rs:137 — the --loader help text — expanding the advertised loader list from 10 to 17 names. The rest of the diff is two new test blocks in test/cli/bun.test.ts: one asserting the help text on bun --help and bun run --help, and one that maps a synthetic extension to each advertised name and imports a fixture through it, proving each name is both accepted by argument parsing and applied at runtime.

Security risks

None. This edits a static help string and adds read-only tests that spawn the bun binary in a temp dir. No parsing, validation, or resolution logic is touched.

Level of scrutiny

Low. The runtime change is documentation-only — loader_resolver() already routes through bun_ast::Loader::from_string over LOADER_NAMES, and I verified every newly-listed name appears there. The PR description explicitly justifies each inclusion and each omission (jsonc/sqlite_embedded collapse in to_api(), base64/dataurl are stubs, sh maps to file, mjs/cjs/etc. are aliases), which matches the source. The test file additions are the larger surface but follow the existing patterns in the same file (the neighboring --help preserves <placeholder> block uses identical spawn/drain/assert structure).

Other factors

The tests are well-constructed per REVIEW.md: test.concurrent for independent subprocess spawns, Promise.all on stdout/stderr/exited, using tempDir, bunEnv/bunExe(), stderr asserted before exitCode, and the help-text assertion is exact (toEqual on the split list) rather than toContain. The second test ties the advertised list to actual behavior, so future drift in either direction (a name stops working, or a new loader is added without updating help) will surface here. No prior reviews or outstanding comments; the noted clippy failure is unrelated (from main, fixed in #38272).

[claude]

LGTM — one-line help-string update bringing the --loader list in line with what LOADER_NAMES already accepts.

What was reviewed

• Confirmed every added name (yaml, json5, xml, md, css, html, sqlite) is present in LOADER_NAMES at src/ast/loader.rs:73 and reachable via loader_resolver().
• Checked the new tests: pipes drained concurrently, tempDir/bunEnv/await using per harness convention, stderr/stdout asserted before exit code, test.concurrent for independent subprocess spawns.
• The earlier comment-cop note on Arguments.rs was addressed in 41c0527 and the thread is resolved; the src diff is now the single string edit.

Extended reasoning...

Overview

The PR updates a single hard-coded help string in src/runtime/cli/Arguments.rs (the -l, --loader entry in TRANSPILER_PARAMS_) to list seven loader names that the flag already accepts but the help text omitted. It adds a describe("--loader help text") block to test/cli/bun.test.ts with two tests: one asserting bun --help and bun run --help print the exact 17-name list, and one that maps a synthetic extension to each advertised name, imports a fixture, and asserts the loader-specific result — tying the advertised list to actual behavior.

Security risks

None. This is a display-only string in CLI help output; no parsing, resolution, or runtime logic changed. The test additions spawn the local bunExe() in a temp dir with no network access.

Level of scrutiny

Low. The source change is a mechanical documentation fix — the flag's resolver (loader_resolver → bun_ast::Loader::from_string over LOADER_NAMES) is untouched, and I verified against src/ast/loader.rs that every newly listed name maps to a real Loader variant. The PR description carefully justifies which accepted names are excluded (aliases, unimplemented stubs, names that collapse via to_api()) and cross-references the sibling PRs updating docs/completions/types.

Other factors

The tests follow repo conventions cleanly (tempDir from harness, test.concurrent, Promise.all on stdout/stderr/exited, assertions before exit-code). The one prior review comment (comment-cop about a verbose source comment) was addressed in the current head commit and the thread is marked resolved. No outstanding reviewer feedback and no prior claude[bot] review on this PR.