Remove dead code: write-only fields and never-set options in 11 crates

[robobun]

Problem

• 16 struct fields in 11 crates are never read. Two options are never set: BufferWriter.append_null_byte is never true, TransposeState.import_record_tag is never Some.
• No lint reports them. rustc's dead_code pass counts x.f = v as a use of f. hawk counts the initializer as a reference.

Fix

• Delete each field with its initializers and stores, and the two branches that the options guard. 18 files, 105 lines removed.
• Candidates come from a compiler probe: #[deprecated] on 12,134 struct fields, cargo check --force-warn deprecated for Linux, Windows, and macOS, then a syn pass marks each use as read or write. 134 fields have no read. Most stay (Notes).
• Verified: bun bd, bun run rust:check-all (12 of 12 targets), cargo check --workspace --all-targets, and the tests in the Notes.
• Self-reviewed: 4 concerns raised, 2 addressed. Rejected as outside a deletion: the now vestigial written_without_trailing_zero() and an older stale comment.

Background

• MultiArrayList<T> keeps each field of T in its own column. Code reads a column by name as a string (items::<"name", _>()), so the compiler sees no read. Those fields stay.
• BufferWriter is the JS printer's output buffer. done() could append a NUL byte. No caller asks for it.

Downsides

• bun_core::output::Source becomes Send + Sync, because its raw pointer fields are gone. It lives in a thread_local! only.
• The bun_resolver::Result in _resolve now drops when _resolve returns. It has no Drop effect: paths are borrowed, fds are Copy.

Notes

Removed, one line each

• bun_core::output::Source: buffered_stream, buffered_error_stream, stream, error_stream (*mut io::Writer). init() cached them, nothing read them. The accessors of the same names return *_backing.new_interface(), which is a pointer cast with no side effect.
• bun_bundler: InputFileInfo.import_count (MetafileBuilder.rs).
• bun_install: SecurityScanSubprocess.stderr_data (an empty Vec that is never filled, the scanner's stderr is inherited), PackageManager.root_progress_node. The progress.start(b"", 0) call stays, because it starts the progress root.
• bun_js_parser: PropertyOpts.async_range, ScanPassResult.approximate_newline_count, TransposeState.import_record_tag. Support import with { type: "json" } and others #16624 replaced the only setter of the tag with import_loader, which stays.
• bun_jsc: VirtualMachine.has_terminated (its readers were debug panics in enqueue_task_concurrent, which Worker / worker_threads: WebCore-shaped lifetimes, joined threads, one ordered VM teardown #37075 removed), ResolveFunctionResult.result (seven stores, no read). Nothing borrows from the stored value: path and query_string point into the resolver's arena and the specifier.
• bun_js_printer: BufferWriter.append_null_byte, with the seven = false stores in bun_jsc and bun_runtime.
• bun_libarchive: BufferReadStream.reading.
• bun_runtime: WindowsState.is_server (ipc.rs, Windows only). Its last reader went away in more child-process #18688, before the Rust port.
• bun_csrf: GenerateOptions.encoding. csrf_jsc.rs encodes the token itself, and VerifyOptions.encoding stays.
• bun_resolver: DataURL.url (always String::EMPTY).
• bun_s3_signing: S3CredentialsWithOptions.virtual_hosted_style. No code names it. Every reader uses credentials.virtual_hosted_style on the inner S3Credentials.

Passed the probe, kept on purpose

• MultiArrayList columns: fields of JSMeta, File, InputFile, BundledAst, Entry, Node, WatchItem, LineOffsetTable, ServerComponentBoundary, and others.
• Owners of memory that other fields borrow: ParseResult.source_contents_backing, LinkerContext.unique_key_buf, OutputFile.owned_src_path_text, HTTPResponseMetadata.owned_buf, PackageJSON.source_contents and json_tape, MatchedRoute.pathname_backing, SignResult.content_md5, KEventWaker.machport_buf.
• Values whose drop has an effect: Repl.last_error (a GC protect), SecurityScanSubprocess.process, Watcher.thread.
• Options that are unfinished features or open bugs, not dead code: md::Options.hard_soft_breaks and underline (Bun.markdown: implement the hardSoftBreaks and collapseWhitespace options #39495, Bun.markdown: implement the latexMath and underline options #39493), jsc::virtual_machine::Options.dns_result_order (Forward --inspect and --dns-result-order to the VM in compiled executables #40703), P.has_top_level_return (js_parser: make the ESM/CJS classification and the module/exports bindings agree #40840), ReactRefresh.last_hook_seen and force_reset, DebugOptions.dump_environment_variables (the --dump-environment-variables flag is parsed and ignored).
• Kept by a comment in the source: AllocatorConfiguration.long_running, DumpStackTraceOptions.skip_*, NameOfSymbol.has_property_key_comment.
• Removal needs more than a deletion: PostgresSQLQuery Flags.is_done (set by the JS done() call), WorkerPipe.done (leaves two empty reader callbacks), WTFTimer.repeat (leaves an unused FFI parameter), MaxHeapAllocator.len (leaves an empty reset()), ReadToEndResult.err (callers drop read errors, which looks like a bug), Subcommand::Pack with PACK_PARAMS (bun pm pack runs as Subcommand::Pm, so the pack help text is unreachable).

Self-review

Three independent read-only passes tried to prove each deletion wrong (readers through raw pointers, offset_of!, C++ layout mirrors, macros, other cfgs, unit tests, the Zig originals, drop effects). None found a reader. Concerns raised: the Send + Sync change of Source (now in Downsides), the style of the kept progress.start call (now let _ =, as in install_with_manager.rs), BufferWriter::written_without_trailing_zero() (nine callers, it only strips NUL bytes that the printer never appends now, a follow-up), and a comment above ResolveFunctionResult.path that names a function which no longer exists (it predates this change).

Other scans of this run, all clean or already in an open pull request

• A relink of the debug binary with --gc-sections --print-gc-sections, and a zero-mention identifier index over src/, packages/, scripts/ and build/debug/codegen/.
• macro_rules! without an invocation, Cargo features that nothing enables, #if 0 and never-true #if blocks in the C++ bindings, patches that no dependency script applies, unused exports under scripts/, tsc --noUnusedLocals over src/js and scripts/.
• A syn scan for enum variants that no expression constructs. The hits are FFI code tables, derive-constructed variants, or already claimed.

Overlap with the open dead-code pull requests

• Every removed line was checked per file against the diffs of the 36 open dead-code pull requests. None of them deletes these lines. Several touch the same files in other places (output.rs, VirtualMachine.rs, parser.rs, p.rs, js_printer/lib.rs, jsc_hooks.rs, PackageManager.rs, data_url.rs).

Tests run with the debug build

test/js/bun/util/csrf.test.ts (31 pass), test/bundler/metafile.test.ts (65 pass), test/js/bun/archive.test.ts (108 pass), test/bundler/transpiler/transpiler.test.js (222 pass), test/js/bun/resolve/import-empty.test.js, test/js/bun/resolve/esModule.test.ts, test/cli/install/bun-install-security-provider.test.ts (43 pass), test/internal/source-lints/ (195 pass). Also a data: URL import, and bun install under a pty so that the progress root starts.

No test is added. The change deletes fields that nothing reads, so there is no behavior to assert, and test/internal/source-lints/CLAUDE.md asks for no tests that pin dead symbols.

[coderabbitai]

Warning

Review limit reached

• Run on-demand review

_{This review includes 18 billable files and costs up to $4.50.}

• Ask an admin to make reviews automatic

Open in CodeRabbit

Reviews can continue after your included limit without a manual trigger. An admin must approve usage-based billing.

Or wait 15 minutes for your next included review.

Check out review usage here.

View limit details

Limit details: You’ve used all 10 included reviews currently available.

Learn how review limits work.

Review configuration:

⚙️ Run configuration

Configuration used: Repository: oven-sh/bun/.coderabbit.yaml

Review profile: ASSERTIVE

Plan: Essentials

Run ID: 7073c528-3c8f-4da1-9e7f-04cc6bc57190

📥 Commits

Reviewing files that changed from the base of the PR and between 2f9bf19 and 59e88d8.

📒 Files selected for processing (18)

• src/bun_core/output.rs
• src/bundler/linker_context/MetafileBuilder.rs
• src/csrf/lib.rs
• src/install/PackageManager.rs
• src/install/PackageManager/security_scanner.rs
• src/js_parser/p.rs
• src/js_parser/parse/parse_entry.rs
• src/js_parser/parse/parse_property.rs
• src/js_parser/parser.rs
• src/js_printer/lib.rs
• src/jsc/RuntimeTranspilerStore.rs
• src/jsc/VirtualMachine.rs
• src/libarchive/lib.rs
• src/resolver/data_url.rs
• src/runtime/api/csrf_jsc.rs
• src/runtime/ipc.rs
• src/runtime/jsc_hooks.rs
• src/s3_signing/credentials.rs

---

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

[robobun]

Status: ready for review. The diff is green. CI is red on one test that this change does not touch.

• Verification: bun bd, bun run rust:check-all (12 of 12 targets), cargo check --workspace --all-targets, and 8 test files with the debug build. The PR body lists them.
• Each removed field has no reader on Linux, Windows, or macOS in a compiler probe (#[deprecated] on the field, then a read or write classification of every reported use).
• CI, build 119513: the only test that stays red is test/bake/deinitialization.test.ts on Windows 2019 x64 (Segmentation fault at address 0x60600002E). The same crash at the same address is red on an unrelated branch (build 119503). The test also fails and passes on retry on Windows 2019 x64 in builds 119510, 119501, 119499 and 119498, and on Alpine aarch64 in recent main builds (119356, 119333). This change does not touch the dev server. Every other failure in the build passed on retry.

[robobun]

^{Updated 8:30 PM PT - Sep 21st, 2026}

❌ @robobun, your commit 59e88d8 has 1 failures in Build #119513 (All Failures):

• test/bake/deinitialization.test.ts - pid 4200 segmentation fault at address 0x60600002e on 🪟 2019 x64

---

🧪   To try this PR locally:

bunx bun-pr 43745

That installs a local version of the PR into your bun-43745 executable, so you can run:

bun-43745 --bun

[claude]

LGTM — mechanical removal of write-only fields; I found no remaining reader for any of them.

What was reviewed:

• Grepped src/ for every removed field name (append_null_byte, has_terminated, root_progress_node, import_record_tag, stderr_data, is_server, virtual_hosted_style on the outer struct, etc.) — no stray references remain, including in cfg(windows) code in ipc.rs.
• ResolveFunctionResult.result drop in VirtualMachine.rs: bun_resolver::Result holds only borrowed &'static [u8] paths, Copy fds and a raw PackageJSON pointer, no Drop impl, so ret.path does not depend on it staying alive.
• None of the touched structs are #[repr(C)] or referenced via size_of/offset_of assertions, so no FFI layout expectation changes.
• csrf_jsc.rs still uses the local encoding to encode the token after generate, so behavior is unchanged there.

Extended reasoning...

The change deletes 16 write-only struct fields and one never-true printer option across 18 Rust files in 11 crates (+10/-105), touching Output, the bundler metafile, csrf options, the package manager, the parser/printer, VM module resolution, libarchive, data URLs, Windows IPC state, and S3 credentials. It touches no security-sensitive logic: the csrf change removes an unread copy of the encoding while the actual encoding path is unchanged, and the S3 field removed was on the outer wrapper that nothing named. I could not run cargo in this environment, so compile verification rests on CI, but grep confirms no reader survives for any removed field and the one lifetime-relevant deletion (the resolver Result no longer stored) is safe because the Result owns nothing. The diff is mechanical, follows an existing let _ = progress.start pattern, and adds no behavior.