[robobun]

Behaviour change: none

Problem

• The console formatter keeps a set of the values that it is rendering (the visited set). The code that takes the pooled map on first use and records a value has two copies: print_as_prelude (src/jsc/ConsoleObject.rs) and the loop that prints the cause of an error (print_error_instance_body, src/jsc/VirtualMachine.rs).
• error printer: one render per error, [Circular] under the key that closes the cycle #37270 moves the record of an error into the error printer and needs that code again.

Fix

• Add Formatter::visited_insert and Formatter::visited_remove. visited_insert takes the pooled map on first use. It returns true when the value was not in the set, like HashSet::insert.
• print_as_prelude, the cause loop, and the three sites that drop the record before print_as changes the tag of a value call them.
• Verified: the existing suites pass unchanged (Notes). 367 of 380 printed outputs are byte-identical to main. The other 13 run into the end of the stack.

Background

• The visited set is a hash map from a thread-local pool. A value that the formatter finds in it prints [Circular].
• print_as recurses once per nesting level, so its frame must stay small. The locals of the record live in an outlined leaf function for that reason.

Downsides

• print_as_prelude now calls visited_insert: one more call per object, array, Map, Set, class, function, error and event printed. Pool takes and hash operations per print equal main in 15 of 15 measured cases. Release .text: 1280 bytes smaller. Instructions: not measured, perf and valgrind are not installed.

Notes

Status. #44268 covers this PR and lists it under Fixes. This PR stays open as the small fallback until that one lands. I do not push to it any more.

All numbers: linux x64, main a4f1429 and this branch.

Pool takes / hash inserts / hash lookups per print (release build, gdb hit counts of LocalKey::with for the pool, HashMap<JSValue, ()>::get_or_put_slot and ::get_index, difference of 80 and 40 prints):

print	main	this PR
Bun.inspect(new Error("x"))	1 / 2 / 2	1 / 2 / 2
Bun.inspect of an Error with an object property	1 / 3 / 3	1 / 3 / 3
Bun.inspect of an Error with a cause	1 / 3 / 3	1 / 3 / 3
Bun.inspect({ a: 1, b: [1, { c: 2 }] })	1 / 3 / 3	1 / 3 / 3
reportError(new Error("x"))	0 / 0 / 0	0 / 0 / 0
reportError of an Error with an object property	1 / 1 / 1	1 / 1 / 1
reportError of an Error with a cause	1 / 1 / 1	1 / 1 / 1

The other 8 cases (primitive properties, assigned cause, AggregateError of 2, console.log) are equal too.

Stack reserved per function (sub ...,%rsp):

function	main, release	this PR, release	main, debug+ASAN	this PR, debug+ASAN
Formatter::print_as	56 B	56 B	320 B	320 B
Formatter::print_as_prelude	88 B	16 B	544 B	384 B
Formatter::visited_insert (new leaf)		56 B		288 B
VirtualMachine::print_error_instance_body	376 B	360 B	10816 B	10560 B

print_as is the frame that recurses. print_as_prelude and visited_insert return before the formatter descends.

Binary. size of the release binary: .text 80660492 -> 80659212 bytes. Code of the printer functions (nm -S): 18958 -> 18491 bytes.

Output corpus. 38 values (plain errors, errors with properties, cause chains of 4 and 12, AggregateErrors, errors that reach themselves, objects and arrays that hold errors, primitives) through 10 entries (console.log, console.error, console.log(v, v), Bun.inspect at depth 2, 0 and Infinity, util.inspect, throw, Promise.reject, reportError). One process per output, debug+ASAN build. 367 of 380 outputs are byte-identical.

• 12 are e.cause = e; e.errors = [e] and a.x = [b]; b.y = b on the six entries where main recurses until the stack ends. The exit status is the same (134 or 139). The recursion prints 7 more renders before that, because the frame of print_error_instance_body is 256 B smaller.
• 1 is Bun.inspect at depth Infinity of an AggregateError that lists itself: 1014 levels before the stack check stops it, main 1013.

Suites run (debug+ASAN): test/js/bun/util/inspect.test.js, test/js/bun/util/inspect-error.test.js, test/js/node/util/bun-inspect.test.ts, test/regression/issue/circular-error-stack.test.ts, test/js/web/console/console-log.test.ts, test/js/bun/util/reportError.test.ts, test/js/bun/test/stack.test.ts.

---

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

[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: 97713e77-c458-4596-abd0-d3a8e44163a2

📥 Commits

Reviewing files that changed from the base of the PR and between a4f1429 and 7bfdf30.

📒 Files selected for processing (2)

• src/jsc/ConsoleObject.rs
• src/jsc/VirtualMachine.rs

Included review availability: This review used your included allowance. Your plan provides up to 10 included reviews per hour; 5 remain after this review.

---

Walkthrough

Formatter now provides methods to insert and remove visited values. Console object printing and error-chain cycle tracking use these methods instead of manipulating the visited map directly.

Changes

Visited-map handling

Layer / File(s)	Summary
Visited-map handling in console printing src/jsc/ConsoleObject.rs	Formatter adds visited_insert and visited_remove. Console printing uses them for circular-reference checks and before recursive printing. The node_data_mut safety documentation is shortened; its implementation is unchanged.
Error-chain cycle tracking src/jsc/VirtualMachine.rs	Error-chain cycle detection uses Formatter methods to insert and remove visited errors. Imports now use direct Tag and TagPayload imports instead of the console_object module alias.

Suggested reviewers: jarred-sumner, dylan-conway

Priority: ⬇️ Low

Merge Risk: ⚪ Minimal · up to 7bfdf

No merge-blocking behavior change is identified; this refactor appears ready to merge after normal checks.

🚥 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 is concise and accurately identifies the main change: centralizing visited-set handling in one formatter helper.
Description check	✅ Passed	The description explains the problem, fix, behavior impact, verification results, performance data, and test suites. It does not use the exact template headings, but it provides the required informati…

---

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

[claude]

Code review found no issues

No high-confidence issues detected in this change.

[robobun]

@robobun wake up!!

[Jarred-Sumner]

@robobun wake up!!

[robobun]

I am here. The status:

• This PR is ready for review. CI is green at 7bfdf30 (Buildkite 121294). It changes no behaviour: it moves the code that records a value in the visited set of the console formatter into Formatter::visited_insert and Formatter::visited_remove.
• error printer: one render per error, [Circular] under the key that closes the cycle #37270 is stacked on this branch. It fixes the crash for an Error that reaches itself: const e = new Error('cyc'); e.cause = e; e.errors = [e]; throw e; ends in SIGSEGV on main. CI is green there too (Buildkite 121295).
• Merge order: this PR first. Then I rebase error printer: one render per error, [Circular] under the key that closes the cycle #37270 on main and set its base to main.
• One push to error printer: one render per error, [Circular] under the key that closes the cycle #37270 is pending. It removes a check in the cause loop that the last review found to be dead code.

If you want a different shape, for example one PR and not two, tell me and I change it.

[robobun]

Closing: this change is on main. #46021 merged the helpers of this PR (Formatter::visited_insert and Formatter::visited_remove in src/jsc/ConsoleObject.rs) together with the fix of #37270 that uses them.