[robobun]

Stacked on #38528.

Behaviour change: none

Problem

• StyleSheet::parse takes Option<&mut Vec<ImportRecord>>. StyleSheet::to_css and StyleAttribute::to_css take Option<ImportInfo> (src/css/css_parser.rs). A caller that passes None compiles.
• With None, the parser rejects each url() with Unexpected token: url(...) and the printer fails with No import records found. Two callers did that: build_css_output (bun build --no-bundle: keep css url() references instead of failing to parse #38528) and normalize_css of bun pm diff (install: fix regressions and older bugs found after the package manager consolidation #45307).

Fix

• The three functions take &mut Vec<ImportRecord> and ImportInfo, as parse_bundler and StyleAttribute::parse already do. The 8 callers drop their Some(..).
• Correct because no caller passes None after bun build --no-bundle: keep css url() references instead of failing to parse #38528. Parser::new and to_css_with_writer keep the Option: An+B, Bun.color and the linker's url-free blocks have no list.
• Verified: with None put back in build_css_output, cargo check -p bun_bundler gives two E0308 errors. test/bundler/css/, test/js/bun/css/, test/bundler/cli.test.ts and bun-pm-diff.test.ts pass.
• Self-reviewed: 5 concerns raised, 5 addressed (list in Notes).

Background

• Import record: the CSS parser adds one bun_ast::ImportRecord per url(). The printer reads the url() back from it through ImportInfo (src/css/printer.rs).
• Considered a parser that keeps a url() as text with no list: the tests of bun build --no-bundle: keep css url() references instead of failing to parse #38528 pass, but a bundler caller that forgets its table then prints unresolved urls. Considered one parse-and-print entry point in bun_css: a new API, which needs a maintainer's decision.

Downsides

• None found. The release build differs from its base in one function of 163,450: StyleSheet::parse_with, which parse and parse_bundler call, is 16 bytes smaller. The binary keeps its size.

Notes

Why a separate PR

The base of this PR is the branch of #38528. The repair of bun build --no-bundle is two arguments in one function and is in that PR. This change reaches no user, so it can wait or be dropped without an effect on that repair.

parse_with is the shared body of parse and parse_bundler. It now takes the NonNull pointer to the list that both of them already have, and no longer an Option of it.

Callers

• StyleSheet::parse: src/bundler/transpiler.rs (build_css_output), src/runtime/cli/pm_diff_normalize.rs (normalize_css), src/css_jsc/css_internals.rs (testing_impl).
• StyleSheet::to_css: the same three, and src/bundler/linker_context/prepareCssAstsForChunk.rs.
• StyleAttribute::to_css: src/css_jsc/css_internals.rs (attr_test).
• None is in cfg-gated code, so the host build compiles each of them.

What keeps the Option

• Parser::new(.., None, ..): the An+B parser in src/css/css_parser.rs and Bun.color in src/css_jsc/color_js.rs. These grammars have no url().
• to_css_with_writer(.., None, ..): the linker prints url-free blocks with it in src/bundler/linker_context/generateCompileResultForCssChunk.rs.
• Printer::new(.., None, ..): Bun.color.

What this does not guard

• This guards one table: the import record list. A stand-alone caller still pairs the symbols and the source index by hand.
• build_css_output is a short copy of the bundler's CSS step. css: export view-transition pseudo-element classes from CSS modules #42729 repaired its symbol table and source index. The types here do not cover those.
• The browser targets of its minify step (bun build --no-bundle: minify css with the build target's browser targets #38544) and the location of its errors (bun build --no-bundle: report a CSS error with its file, line and column #46184) are separate.

The text fallback, built and run

• The design: Parser::add_import_record returns a sentinel index when there is no list, and Url::to_css prints the text of the Url for that index (4 files in src/css). build_css_output keeps its two Nones.
• Result on main with that patch and without bun build --no-bundle: keep css url() references instead of failing to parse #38528: 19 of 19 new tests of bun build --no-bundle: keep css url() references instead of failing to parse #38528 pass.
• It fails on @container style(composes: b from "./f.css") in a CSS module with error: No import records found while printing. composes .. from holds only an index.
• A bundler caller that forgets its table prints each url() unresolved and exits 0. Today that caller fails.

Open PRs that edit the same lines

• bundler: load with { type: "css" } as a module that exports a CSSStyleSheet in browser builds #41959 adds a caller sheet.to_css(.., Some(ImportInfo::init_outside_of_bundler(..)), ..) in src/bundler/css_module_script.rs. It must drop the Some(..) after this merges.
• css: print a stylesheet with the targets StyleSheet::minify recorded #39251 and css: log parser warnings with the Source and a byte position #45787 edit the same signatures in src/css/css_parser.rs.
• Each of the three conflicts with main today.

Self-review

1. Keep this change out of bun build --no-bundle: keep css url() references instead of failing to parse #38528, which repairs the bug. Done: this PR is stacked on it.
2. Say what the types guard and what they do not. See the two lists above.
3. Name the open PRs that edit the same lines. See above.
4. The reason against the text fallback was its cost in a bundled build, and that cost is one compare. I built the fallback and ran it. The reasons are now the two results above.
5. The doc comment of Parser::new changes by one line.

Measurements (release builds of #38528 and of this branch, linux x64)

• Size: stripped bun is 140,478,752 bytes on both. .text is 110,627,497 bytes on both.
• Of 163,450 sized text symbols (llvm-nm -S), 1 differs: StyleSheet::parse_with::<BundlerAtRuleParser>, 9,143 to 9,127 bytes and 1,719 to 1,715 instructions. It no longer tests the Option.
• StyleSheet::parse (282 bytes), parse_bundler (282), to_css_with_writer (1,535), StyleAttribute::to_css (1,231) and Transpiler::build_css_output (2,566) are equal on both.
• Compile check: with None put back at the two arguments in build_css_output, cargo check -p bun_bundler prints error[E0308]: mismatched types: expected &mut Vec<ImportRecord<'_>>, found Option<_> and error[E0308]: mismatched types: expected ImportInfo<'_>, found Option<_>.
• The 88 probe stylesheets that I ran under --no-bundle print the same bytes on both debug builds.

Suites run on the debug build of this branch

test/bundler/cli.test.ts, test/bundler/css/, test/js/bun/css/, test/bundler/esbuild/css.test.ts, test/cli/install/bun-pm-diff.test.ts, test/bundler/bundler_loader.test.ts, test/bundler/bundler_html.test.ts: 3,956 pass, 48 skip, 4 todo, 0 fail.

---

no test proof · iteration 2 · the description declares no behaviour change, so there is no failing test to prove; the existing suite in CI is the check

[robobun]

^{Updated 8:06 PM PT - Nov 8th, 2026}

❌ @robobun, your commit 2bd274a has some failures in Build #138973 (All Failures)

---

🧪   To try this PR locally:

bunx bun-pr 47990

That installs a local version of the PR into your bun-47990 executable, so you can run:

bun-47990 --bun

[robobun]

This PR changes three signatures and no behaviour, so there is no failing input to reproduce.

What it prevents: with None put back at the two arguments in Transpiler::build_css_output, cargo check -p bun_bundler fails:

src/bundler/transpiler.rs:3063:13: error[E0308]: mismatched types: expected `&mut Vec<ImportRecord<'_>>`, found `Option<_>`
src/bundler/transpiler.rs:3092:13: error[E0308]: mismatched types: expected `ImportInfo<'_>`, found `Option<_>`

Before this PR that code compiles, and bun build --no-bundle x.css then fails on each url() (#38528).

This PR (#47990) is stacked on #38528.

CI on 2bd274a (build 138973): no test failed. Each test lane that ran passed all of its shards: Linux (alpine, debian, debian with ASAN, ubuntu, on x64 and aarch64), Windows (x64 and aarch64) and macOS aarch64. The four macOS x64 test shards did not run. Each one waited in the queue without an agent until it expired, four times in a row.

[claude]

Code review found no issues

No high-confidence issues detected in this change.