[robobun]

Problem

• A test can fail with a syntax error or a missing module. The JUnit report then has <failure type="Error" /> with no type, message or location.
• bun test captures a failure through the hook on_print_error_zig_exception, which print_error_instance_body calls (src/jsc/VirtualMachine.rs:6769). BuildMessage and ResolveMessage print from their Msg in print_error_from_maybe_private_data (:6036) and return before that call.

Fix

• print_error_from_maybe_private_data calls the hook for both classes. It passes the class name, the message text and one frame from the location of the message.
• The call is under allow_side_effects, like the existing call. It also runs when the message was logged before.
• Verified: the new test in test/js/junit-reporter/junit.test.js fails on main. Also ran inspect-error.test.js, resolve-error.test.ts, build-error.test.ts, reportError.test.ts.
• Self-reviewed: 1 concern raised, 1 addressed.

Background

• BuildMessage and ResolveMessage are what bun throws for a transpile error and a failed resolution. Each holds a bun_ast::Msg: text, file, line, column.
• The hook is a function pointer on the VM. bun test sets it while it prints the error of a failed test, and TestFailure::record_cb fills <failure>. No other code sets it.
• Considered: read name and message from the JS value in the reporter. That can run user code, and bun already has the text.

Downsides

• Each printed BuildMessage or ResolveMessage executes 7 more instructions when no hook is installed (89 to 96 in print_error_from_maybe_private_data).
• Release binary: +1,125 bytes in the two functions (nm -S).
• <failure> has one frame, without the notes of the message. The inspector and GitHub annotations still get nothing for these classes.

Notes

• This is the first PR of a stack. bun test --reporter=junit: record a failure that is not a finished test #44435 records a failure that is not a finished test in the JUnit report. A file that fails to load with a syntax error gets <failure type="BuildMessage" message="Unexpected ;"> there because of this PR.
• Before: <failure type="Error" />. After: <failure type="BuildMessage" message="Unexpected ;">BuildMessage: Unexpected ; and at broken.js:1:11.
• print_error_instance_body does three things for an error under allow_side_effects: it reports to the inspector, it calls this hook, and it prints a GitHub annotation. This PR adds only the hook call for the two classes. bun test: emit one anchored GitHub Actions annotation per failed test #41876 is open for the annotations.
• The instruction count is the number of guest instructions that qemu-user executes inside the function (-one-insn-per-tb -d exec,nochain -dfilter <range>) for bun run of a file with const x = ;, on release builds of the merge base and of this change. It is 90 to 97 for a missing import. The whole process executes about 2.5 million, with a spread of 3% between runs. perf, valgrind and bloaty are not installed on the machine I used.
• The bytes: print_error_from_maybe_private_data 1,520 to 1,668, plus 977 for the cold helper that builds the exception.
• ZigException::with_stack holds the field initialiser that Holder::zig_exception had. The hook call does not need a Holder.
• The self-review concern: the hook is one of three observers. The Downsides bullet and the note above state that limit.

[robobun]

Status: ready for review. First PR of a stack: #44434 (this one), #44435, #44436 (draft, optional).

How I reproduced it: broken.js is const x = ;. A test does await import("./broken.js"). Run bun test --reporter=junit --reporter-outfile=junit.xml.

• main: <failure type="Error" />
• this PR: <failure type="BuildMessage" message="Unexpected ;">

[coderabbitai]

Navigate logical layers of code changes, visualize relationships, and explore their blast radius.

No actionable comments were generated in the recent review. 🎉

ℹ️ Recent review info

⚙️ Run configuration

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

Review profile: ASSERTIVE

Plan: Essentials

Run ID: 7d677973-8130-4e29-8f19-b72989572b0a

📥 Commits

Reviewing files that changed from the base of the PR and between bc7a813 and b3ede46.

📒 Files selected for processing (3)

• src/jsc/VirtualMachine.rs
• src/jsc/ZigException.rs
• test/js/junit-reporter/junit.test.js

Included review availability: This review used your included allowance. Your plan provides up to 10 included reviews per hour; 0 remain after this review.

---

Walkthrough

When side effects are allowed, VirtualMachine reports build and resolve messages to the configured exception hook as ZigException instances. The change adds exception initialization with an existing stack and a JUnit test for dynamic-import build and resolution failures.

Changes

Message Exception Reporting

Layer / File(s)	Summary
Initialize exceptions with an existing stack src/jsc/ZigException.rs	ZigException::with_stack initializes exception defaults with a supplied stack. Holder::zig_exception uses this helper.
Report build and resolve messages src/jsc/VirtualMachine.rs, test/js/junit-reporter/junit.test.js	VirtualMachine reports build and resolve messages to the configured exception hook when side effects are allowed. The test checks failure types, messages, source location, and process exit code.

Suggested reviewers: jarred-sumner

Priority: ⬇️ Low

Merge Risk: ⚪ Minimal · up to b3ede

The change appears mergeable after normal checks. The new JUnit test has not been executed in this review.

🚥 Pre-merge checks | ✅ 4

✅ Passed checks (4 passed)

Check name	Status	Explanation
Linked Issues check	✅ Passed	Check skipped because no linked issues were found for this pull request.
Out of Scope Changes check	✅ Passed	Check skipped because no linked issues were found for this pull request.
Title check	✅ Passed	The title clearly describes the main change: populating JUnit details for build and resolve errors.
Description check	✅ Passed	The description explains the problem, implementation, verification steps, scope, and trade-offs. It does not use the exact template headings, but it provides the required information and is substantia…

• Autopilot · Keep fixing CodeRabbit findings and required CI, and resolving merge conflicts

Autopilot is currently an internal CodeRabbit preview.

---

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

[claude]

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

Beyond the inline findings, I also checked the lifetime of the synthesized ZigException: name/message/source_url are borrowed from the Msg and a stack array, but the only hook consumer (TestFailure::record in src/runtime/cli/test_command.rs) copies name, message and each frame into owned Vecs synchronously, so nothing outlives the call. The with_stack refactor in src/jsc/ZigException.rs is field-for-field identical to the initializer it replaced.

Extended reasoning...

The change adds a hook call for BuildMessage/ResolveMessage in src/jsc/VirtualMachine.rs, extracts a ZigException constructor in src/jsc/ZigException.rs, and adds one junit reporter test; it touches no security-sensitive surface. Two inline findings (missing JS frames for synchronous throws, and the test discarding stdout/stderr) were already posted, so a human look is signaled by those; this note records only what else was examined and ruled out.

[robobun]

Closing: one PR in place of the stack. This change is the first commit of #44435, which now targets main.