You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
{{ message }}
Repository navigation
test(transpiler): record which early errors are erased in code that nothing runs - #48011
Bun's parser leaves some early errors to JavaScriptCore. Where a pass removes the code, nothing reports the error: if (0) { x = /(/; } runs.
bun test --coverage turns dead code elimination off and refuses the same file.
No test records this class.
Fix
A table in test/bundler/transpiler/runtime-transpiler.test.ts holds 95 programs. A child bun loads each program in two files: under if (0), and under a condition that Bun can't decide.
Today 11 early errors are reported in both files. 64 are erased under if (0), and 16 where the code stays. The other 4 programs are valid controls.
The table is a floor. The test fails when Bun reports an error less than its line says. A fix passes, and its author can move the lines up.
Verified: runtime-transpiler.test.ts on release and debug builds of b1305dd. Self-reviewed: 9 concerns raised, 9 addressed, and the 5 review comments too.
Background
An early error is a syntax error that the specification requires before any code runs. It applies to code that never runs too.
"Erased where the code stays" means that both files load although the engine refuses the first as written.
The table records what Bun does, not pending fixes. bun run --debug-aot refuses 67 of the 80 erased programs. 11 of the other 13 have an invalid RegExp literal.
Considered a scan of the lint suites' tables or the AOT suite's lists. A change to those lists then breaks this test.
Notes
The table on main (95 programs)
List
Reported
Erased under if (0)
Erased where it stays
No early error
accessors.js
2
1
assignment-targets.js
2
4
await.js
6
class-members.js
1
2
2
declarations.js
2
4
2
delete.js
2
delete.ts
1
labels-and-jumps.js
1
6
modules.js
1
new.js
4
1
regexp-literals.js
1
4
7
reserved-names.js
2
18
reserved-names.ts
2
sloppy-code.cjs
3
strict-code.cjs
2
5
2
using.js
4
1
Total
11
64
16
4
/\P{assigned}/u; in regexp-literals.js is the program of #15219.
How an outcome is decided
Each program is in two files that nothing runs: if (globalThis.never) { <program> } and if (0) { <program> }. A "use strict"; at the start of a program stays above the if.
A child bun --no-install loads each file, a .cjs file with require and any other file with import(). The package has "type": "module".
If both files load, the engine reads the first file as written: a .js file as a module, a .cjs file as the body of a CommonJS function. If it refuses the text, the error is erased where the code stays. The engine can't read TypeScript, so a .ts program that loads in both files is an error of the table.
The table holds no error message, so a WebKit upgrade that rewords an error changes nothing.
Other checks, outside the test
Node.js 26.3.0: node --check refuses all 88 .js and .cjs programs that have an early error. It accepts the 4 controls. The other 3 programs are TypeScript.
bun run --debug-aot was asked about each of the 80 erased programs in the file where it is erased. It stops on 67 with a TypeScript diagnostic, such as TS2392 for two constructors. It runs 13: the 11 with an invalid RegExp literal, "use strict"; x = 010; and function f() { "use strict"; with ({}) {} }. So bun run and bun run --aot already disagree on 67 of these files.
Positions that are in the tree
An invalid RegExp literal in removed code: test/aot/parallel/language/invalid-regexp-in-dropped-code.js has to run under --aot as it does under bun. src/aot/lower/expression/array.rs says that bun run never parses a literal that its transpiler leaves out.
new.target outside a function: src/js_parser/visit/visit_expr.rs says that the parser's error "is intentionally not emitted". The AOT front end refuses it, also under if (0).
Question for a maintainer: for if (0) { new.target; } in a module, is bun run (it runs) or bun run --debug-aot (TS17013) right?
The outcomes have an order: reported, erased under if (0), erased where the code stays. The test fails when a program is below its line, or when a valid program is refused.
A strict comparison would fail on every fix. A fix and this PR would each pass alone, and main would be red after the second merge.
Each line is exact on main today. Commit 0dfa8b3 compared the whole table with toEqual and passed.
The test passes on the base build, by design. The PR has no fix to prove: it records what the base build does.
The second version scanned the AOT suite's early-error lists with Bun.Transpiler. It kept the result in a list file. A program added to those lists would fail this test with no conflict for git to show. With the lists of AOT: front end, test runs and tests for the rewritten compiler #47942 it asked for six more lines. Programs with an export were never examined, because a block can't hold one.
Turning dead code elimination off at run time is not a fix. It also changes which files are CommonJS: if (0) { module.exports = 1; } new.target; is a module under bun test and CommonJS under bun test --coverage.
What the test doesn't find
An early error in a program that is not in the table.
An error in syntax that only the top level holds. let a; export { a as "\uD83D" }; runs, and Node.js refuses it.
An error that JavaScriptCore accepts too.
An erased error in TypeScript where the code stays. The engine can't read that source.
Cost
The test takes 40 ms on a release build and 1 s on a debug build with ASAN.
Test-only change.Debug/ASAN (expected pass):
$ bun bd test'test/bundler/transpiler/runtime-transpiler.test.ts'
$ BUN_DEBUG_QUIET_LOGS=1 bun scripts/build.ts --profile=debug --quiet test test/bundler/transpiler/runtime-transpiler.test.tsbun test v1.4.4 (367d939d9)test/bundler/transpiler/runtime-transpiler.test.ts:(pass) use strict causes CommonJS [372.76ms](pass) non-ascii regexp literals [4.05ms](pass) ascii regex with escapes [1.82ms](pass) // @bun > async transpiler [26.11ms](pass) // @bun > require() [10.95ms](pass) // @bun > synchronous [212.46ms](pass) json imports > require(*.json) [13.51ms](pass) json imports > import(*.json) [6.93ms](pass) json imports > should support comments in tsconfig.json [7.54ms](pass) json imports > should handle non-boecjts in tsconfig.json [5.36ms](pass) json imports > should handle duplicate keys [5.02ms](pass) with statement > works [203.38ms](pass) math.pow [3.97ms](pass) unterminated string literals in large files > reports an unterminated string literal at the end of a large JavaScript file [368.31ms](pass) unterminated string literals in large files > reports an unterminated string literal at the end of a large JSON file [274.12ms](skip) printing more than 2 GiB of modules in one process keeps the space after a keyword(pass) an early error in code that nothing runs > is reported no less than the table says [966.84ms] 16 pass 1 skip 0 fail 37 expect() callsRan 17 tests across 1 file. [4.81s]Exit: 0
…ed code
Bun's parser leaves some early errors to the engine: it prints the code
and JavaScriptCore refuses the printed text. Where a pass removes the
code or prints it differently, nothing reports the error.
The test transpiles each program under `if (1) { }` and under
`if (0) { }` and asks the engine about the printed text.
erased-early-errors.txt lists the programs that are refused in one place
and accepted in the other. A second test loads each listed program as a
file in a child process.
The programs are the early-error lists of the AOT suite, read in place,
and a table in the test for the rules that those lists lack.
Status: the PR is ready for a maintainer. The test passes on a release build and a debug build of b1305dd.
How to see what the table records, with a build of main:
echo'if (globalThis.never) { x = /(/; }'> kept.js && bun kept.js
# SyntaxError: Invalid regular expression: missing )echo'if (0) { x = /(/; } console.log("ran");'> removed.js && bun removed.js
# ran
How to run the test: bun bd test test/bundler/transpiler/runtime-transpiler.test.ts -t "an early error"
CI: build 137962 is finished. test/bundler/transpiler/runtime-transpiler.test.ts passed at the first attempt on each of the 11 lanes that run it. The build has 28 red tests. All 28 are tests of the AOT suite, and all 28 are red on main (build 137822). One ASAN job reached its time limit in test/aot/three-layers-agree.test.ts. This PR changes none of these files.
No actionable comments were generated in the recent review.
Walkthrough
The change adds a regression test for early-error behavior in JavaScript, CommonJS, and TypeScript snippets. It compares expected classifications with engine rejection checks and Bun’s results for retained and removed files.
Changes
Early-error comparison
Layer / File(s)
Summary
Case table and engine check test/bundler/transpiler/runtime-transpiler.test.ts
The test adds source snippets grouped by expected error behavior. Helpers preserve a leading strict-mode directive and check whether node:vm rejects .js and .cjs source.
The test generates retained and removed files for each case, loads .cjs files with require and other files with import, then compares results with the expected classifications. It also checks for empty stderr and a zero exit code.
Suggested reviewers: alii
Merge Risk: Minimal · up to e3a5a
The test provides the intended early-error regression coverage, with no identified issue requiring a fix before merge.
Pre-merge checks: 4
Check name
Status
Explanation
Title check
The title describes the test for early errors erased in code that does not run.
Linked Issues check
Check skipped because no linked issues were found for this pull request.
Out of Scope Changes check
Check skipped because no linked issues were found for this pull request.
Description check
The description explains the test-only change and how the author verified it. The headings differ from the template, but both required topics are covered.
Autofix · Keep fixing CodeRabbit findings and required CI, and resolving merge conflicts
Tip
🎯 Customize your risk assessments
Guide how CodeRabbit assesses merge risk and security architecture risk. Highlight sensitive changes and explain your security boundaries using Merge Risk Instructions and Security Risk Instructions in your review settings. Security architecture risk assessments require the Advanced plan or above.
✅ Automatic approvals, with your rules
Let CodeRabbit approve pull requests once required threads are resolved, pre-merge checks pass, and your risk limits are met. Enable Automatic approvals in review settings, with overrides for teams or file paths.
Thanks for using CodeRabbit! It's free for OSS, and your support helps us grow. If you like it, consider giving us a shout-out.
Included review availability: This review used your included allowance. Your plan provides up to 10 included reviews per hour; 3 remain after this review.
…st.ts
The table has its own programs, each under the outcome that it has
today: reported, erased under `if (0)`, erased where the code stays, or
no early error. One child process loads each program in two files.
The test reads no list of another suite, and it has no list file and no
switch to write one.
robobun
changed the title
test(transpiler): list the early errors that nothing reports in removed code
test(transpiler): record which early errors are erased in code that nothing runs
Nov 6, 2026
The reason will be displayed to describe this comment to others. Learn more.
Nothing blocking. The comments below are optional suggestions. There is no need to push a fix for them before merging.
Beyond the inline nit, I also checked that vm.SourceTextModule parses eagerly in its constructor, so the isRefusedAsWritten oracle for .js rows is live; that the classifier throws on any unexpected error name and yields an outcome absent from the table (so a surprise fails toEqual rather than passing silently); and that the "use strict" hoisting in fileOf matches the three directive rows. The rewrite into runtime-transpiler.test.ts with an inline table resolves the earlier points about a separate file and the AOT-list coupling.
Extended reasoning...
Test-only change: 197 added lines in test/bundler/transpiler/runtime-transpiler.test.ts adding one test that spawns a single Bun process over ~200 generated files and compares a hard-coded four-way outcome table of ~100 early-error programs against observed behavior. It touches no production code and no security-sensitive surface. The table is an empirical snapshot of current Bun+JSC behavior that cannot be verified here without running the test, and a maintainer may want to weigh its maintenance cost on WebKit or parser changes, so a human look is still worthwhile.
…r table can't judge
The table takes .js, .cjs and .ts lists. Any other extension is an
error. A .ts program that loads in both files is an error too: the
engine can't read it, so nothing says whether it has an early error.
The reason will be displayed to describe this comment to others. Learn more.
Nothing blocking. The comments below are optional suggestions. There is no need to push a fix for them before merging.
The extension fall-through raised last round is addressed: unknown list extensions now throw in the row builder, and isRefusedAsWritten throws for .ts instead of filing the row under "has no early error". Beyond the inline nit, I also checked hermeticity (--no-install, harness tempDir, using registered before any assertion), that stdout/stderr/exited are drained together, and that vm.SourceTextModule exists in Bun's node:vm (src/js/node/vm.ts:479) — none of these are a problem. Since the table deliberately records current behaviour rather than desired behaviour, and the description itself poses a question for a maintainer about new.target under if (0), a maintainer's look is still worthwhile.
Extended reasoning...
The change adds one characterization test (+205 lines) to test/bundler/transpiler/runtime-transpiler.test.ts with no source changes and touches no security-sensitive surface. The prior-round classifier gap for .mjs/.cts rows is fixed in this push, and the remaining inline finding is a nit about the test passing under USE_SYSTEM_BUN=1. Deferring rather than approving because the table encodes a product position (which early errors Bun is allowed to erase) that the author explicitly asks a maintainer to decide.
The test fails when Bun reports an error less than its line says, or
refuses a valid program. A change that reports more of these errors
passes, so a fix does not fail the test before its lines are moved.
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
![claude[bot]](https://avatars.githubusercontent.com/in/1236702?s=60&v=4){kind=small}
robobun
changed the title
Behaviour change: none
Problem
if (0) { x = /(/; }runs.bun test --coverageturns dead code elimination off and refuses the same file.Fix
test/bundler/transpiler/runtime-transpiler.test.tsholds 95 programs. A childbunloads each program in two files: underif (0), and under a condition that Bun can't decide.if (0), and 16 where the code stays. The other 4 programs are valid controls.runtime-transpiler.test.tson release and debug builds of b1305dd. Self-reviewed: 9 concerns raised, 9 addressed, and the 5 review comments too.Background
bun run --debug-aotrefuses 67 of the 80 erased programs. 11 of the other 13 have an invalid RegExp literal.Notes
The table on main (95 programs)
if (0)accessors.jsassignment-targets.jsawait.jsclass-members.jsdeclarations.jsdelete.jsdelete.tslabels-and-jumps.jsmodules.jsnew.jsregexp-literals.jsreserved-names.jsreserved-names.tssloppy-code.cjsstrict-code.cjsusing.js/\P{assigned}/u;inregexp-literals.jsis the program of #15219.How an outcome is decided
if (globalThis.never) { <program> }andif (0) { <program> }. A"use strict";at the start of a program stays above theif.bun --no-installloads each file, a.cjsfile withrequireand any other file withimport(). The package has"type": "module"..jsfile as a module, a.cjsfile as the body of a CommonJS function. If it refuses the text, the error is erased where the code stays. The engine can't read TypeScript, so a.tsprogram that loads in both files is an error of the table.Other checks, outside the test
node --checkrefuses all 88.jsand.cjsprograms that have an early error. It accepts the 4 controls. The other 3 programs are TypeScript.bun run --debug-aotwas asked about each of the 80 erased programs in the file where it is erased. It stops on 67 with a TypeScript diagnostic, such as TS2392 for two constructors. It runs 13: the 11 with an invalid RegExp literal,"use strict"; x = 010;andfunction f() { "use strict"; with ({}) {} }. Sobun runandbun run --aotalready disagree on 67 of these files.Positions that are in the tree
test/aot/parallel/language/invalid-regexp-in-dropped-code.jshas to run under--aotas it does underbun.src/aot/lower/expression/array.rssays thatbun runnever parses a literal that its transpiler leaves out.new.targetoutside a function:src/js_parser/visit/visit_expr.rssays that the parser's error "is intentionally not emitted". The AOT front end refuses it, also underif (0).if (0) { new.target; }in a module, isbun run(it runs) orbun run --debug-aot(TS17013) right?Open PRs that change an outcome
var/letpattern and ofusingin afor (;;)head #47992 is about a pattern with no initializer (let [a];), Keep the parentheses of an optional chain undernewand as a template tag, and rejectnew a?.b()#41638 aboutnew a?.b(), and AOT: front end, test runs and tests for the rewritten compiler #47942 about a function-level"use strict". Each of them is about errors of this table. If one reports more of them, the test still passes, and its lines can move up.A floor, not an exact record
if (0), erased where the code stays. The test fails when a program is below its line, or when a valid program is refused.toEqualand passed.Earlier shapes of this PR
lint.test.tsandoxc-errors.test.ts. To share them, 13,400 lines move, andgit merge-treeshows thatbun check: TypeScript 7.1 #46383 and Reject ? and ! after an object literal property name like oxc #47762 then conflict.Bun.Transpiler. It kept the result in a list file. A program added to those lists would fail this test with no conflict for git to show. With the lists of AOT: front end, test runs and tests for the rewritten compiler #47942 it asked for six more lines. Programs with anexportwere never examined, because a block can't hold one.if (0) { module.exports = 1; } new.target;is a module underbun testand CommonJS underbun test --coverage.What the test doesn't find
let a; export { a as "\uD83D" };runs, and Node.js refuses it.Cost
[auto-merge] gate passed · iteration 1 · 1 files touched
passes on PR (with fix)
diff hotspot
gate history · 4 passed · 0 rejected · iteration 1
evidence per changed file