Output.err: print the generic form for an errno outside the SystemErrno table

[robobun]

Problem

• Output::err prints a bun_sys::Error whose errno is not in the SystemErrno table as UNKNOWN: <msg> (<syscall>). "UNKNOWN" is not an errno code. The plain error: <msg> form, with the number, is more useful.
• Cause: impl ErrName for bun_sys::Error always returns Some(SysErrInfo { tag_name: Error::name(self), .. }) (src/sys/Error.rs:522), and Error::name() falls back to the literal "UNKNOWN" (src/sys/Error.rs:319). So err_with_body (src/bun_core/output.rs:2400) always takes the syscall-error branch.

Fix

• SysErrInfo::tag_name is now Option<&'static str>. The bun_sys impl sets it from get_error_code_tag_name(), so it is None when the errno does not resolve.
• err_with_body prints error: <msg> (<syscall>, errno N) for that case. The known-errno forms do not change: ENOENT: No such file or directory: <msg> (open).
• Correct because the table is the only source of a code name. When it has none, the output says so with the generic prefix and keeps the number, which is the one fact the user can look up.
• Verified: test/js/bun/sys/error-name-from-libuv.test.ts. Before the fix the new test gets UNKNOWN: sysErrorOutputErr (open). After: error: sysErrorOutputErr (open, errno 524). Also test/internal/sigaction-layout.test.ts and test/internal/source-lints.

Background

• Output::err(name, fmt, args) is the CLI error printer. For a bun_sys::Error it renders <code>: <strerror>: <msg> (<syscall>). The SysErrInfo struct is how bun_sys (a higher crate) hands the pieces to bun_core without bun_core naming the error type.
• SystemErrno is bun's per-OS errno enum. The kernel is not bound to it: FUSE filesystems and some drivers return values it does not declare, for example Linux ENOTSUPP (524).
• No syscall in CI returns such a value, so the test uses a new sysErrorOutputErr(errno, fromLibuv) hook in bun:internal-for-testing. It builds the bun_sys::Error and calls Output::err on it. The hook is the same one Output.err: print the errno description for libuv errors on Windows #40852 adds, so the two branches merge without a second copy.

Notes

• Related: Output.err: print the errno description for libuv errors on Windows #40852 fixes the strerror label for from_libuv errors on Windows in the same as_sys_err_info. The two changes compose: that PR changes the errno field, this one changes tag_name. Whichever lands second has a small conflict in as_sys_err_info and in the test file.
• Error::name() keeps its "UNKNOWN" fallback. Other callers use it in free-form messages (failed to open: {name}), which this request does not cover.
• An errno that a syscall wrapper already collapsed to E::EUNKNOWN (via E::from_raw) is in the table and still prints as EUNKNOWN: <msg> (<syscall>). The raw number is lost before the Error is built, so there is nothing better to print there.
• Ran cargo fmt --check, cargo clippy -p bun_core -p bun_sys -p bun_sys_jsc, and prettier on the touched files.

---

no test proof · iteration 0 · platform-specific test(s) that do not run on this machine, deferring to CI, which covers all platforms: test/js/bun/sys/error-name-from-libuv.test.ts

[coderabbitai]

Warning

Review limit reached

• Run on-demand review

_{On-demand reviews are free for the next 23 days. After that, they cost $0.25 per reviewed file.}

Or wait 1 minute for your next included review.

View limit details

Limit details: You’ve used all 5 included reviews currently available.

Learn how review limits work.

Review configuration:

⚙️ Run configuration

Configuration used: Path: .coderabbit.yaml

Review profile: ASSERTIVE

Plan: Pro

Run ID: ef1e8d1b-77bb-4747-8a40-64d2efd90bdd

📥 Commits

Reviewing files that changed from the base of the PR and between d578a8c and df514dc.

📒 Files selected for processing (6)

• src/bun_core/output.rs
• src/js/internal-for-testing.ts
• src/runtime/dispatch_js2native.rs
• src/sys/Error.rs
• src/sys_jsc/error_jsc.rs
• test/js/bun/sys/error-name-from-libuv.test.ts

---

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

[robobun]

^{Updated 2:14 PM PT - Aug 28th, 2026}

❌ @robobun, your commit df514dc has 1 failures in Build #107985 (All Failures):

• test/js/web/url/url.test.ts - code 1: TypeError: Invalid URL on 🍎 any x64

---

🧪   To try this PR locally:

bunx bun-pr 40854

That installs a local version of the PR into your bun-40854 executable, so you can run:

bun-40854 --bun

[robobun]

Status: ready for review.

Reproduced with the new sysErrorOutputErr hook, errno 524 (Linux ENOTSUPP, not in the table):

• before: UNKNOWN: sysErrorOutputErr (open)
• after: error: sysErrorOutputErr (open, errno 524)

A known errno is unchanged: ENOENT: No such file or directory: sysErrorOutputErr (open).

CI on df514dc: the new tests pass on every lane. The one red job is test/js/web/url/url.test.ts on darwin x64 (Invalid URL in the Unicode 16 IDNA test), which also fails on main and is reported for triage. The other failures passed on retry.

Related: #40852 touches the same as_sys_err_info for the Windows from_libuv label. The two changes compose, and the testing hook is the same in both.

[claude]

Nothing blocking. The comments below are optional suggestions. There is no need to push a fix for them before merging.

[claude]

LGTM — the assertion-ordering nit from the earlier review is addressed; tests now assert the combined { stderr, stdout, exitCode } object.

What was reviewed:

• SysErrInfo.tag_name type change: only one producer (impl ErrName for bun_sys::Error) and one consumer (err_with_body), both updated together.
• sys_error_output_err host fn: arg validation before use, from_libuv is read on both cfg branches (early-return vs. struct field), no unused-binding warning.
• New tests spawn via bunExe()/bunEnv with -e, drain both pipes concurrently, and are test.concurrent.

Extended reasoning...

Overview

The PR changes how bun_core::output::err renders a bun_sys::Error whose errno is not in the SystemErrno table: instead of printing UNKNOWN: <msg> (<syscall>), it now prints the generic error: <msg> (<syscall>, errno N) form so the raw number is preserved. Mechanically, SysErrInfo.tag_name becomes Option<&'static str> (was &'static [u8]), the sole producer in src/sys/Error.rs populates it from get_error_code_tag_name() (which returns None for out-of-table errnos), and err_with_body gains an early branch for the None case. The known-errno paths just swap bstr::BStr::new(bytes) for direct &str display — same rendered output. A test-only host function sys_error_output_err is added and wired through dispatch_js2native.rs and internal-for-testing.ts, and two test.concurrent cases in test/js/bun/sys/error-name-from-libuv.test.ts cover both the known-errno (2 → ENOENT) and out-of-table (524) paths.

Security risks

None. This is CLI error-message formatting on a static SysErrInfo struct with &'static str fields. The new host function is confined to bun:internal-for-testing, validates its two arguments (is_number() / is_boolean() / u16::try_from) before use, and only writes to stderr. No user-controlled data reaches allocation, filesystem, or network paths.

Level of scrutiny

Low-to-moderate. The behavioral surface is a single formatted line in an error path that previously printed a less-useful literal. I confirmed via grep that SysErrInfo has exactly one construction site and one read site, both updated in this diff, so the field-type change cannot silently break another producer. The get_error_code_tag_name() helper already existed and is what Error::name() itself wraps, so the semantics of "None means not in the table" are established. Cross-platform: the from_libuv binding is consumed on both #[cfg] branches (non-Windows early-return, Windows struct field), so no dead-binding lint on either target.

Other factors

My earlier inline nit (assert exit code last) was addressed in a follow-up commit — the helper now returns { stderr, stdout, exitCode } and each test asserts the whole object with .toEqual, matching the CLAUDE.md/REVIEW.md convention. The tests follow harness patterns (bunExe(), bunEnv, -e for single-file spawn, concurrent pipe draining via Promise.all, test.concurrent). The change is ~60 net lines, self-contained, and the PR description accurately describes the before/after output. No outstanding third-party CHANGES_REQUESTED reviews in the timeline.