[robobun]

Problem

• On a debug build, cell values survive a full GC between the width and render passes takes 2.3 to 3.6 s on an idle machine, and fails with this test timed out after 5000ms under load. Six open PRs report that timeout.
• The test runs Bun.gc(true) in each of its 64 getters. One full GC takes about 50 ms on a debug build.
• CI does not show it: CI has no debug lane, and its runner passes --timeout=90000 (scripts/runner.node.ts:1167).

Fix

• Run the GC in 4 of the 64 getters (0, 16, 32, 48). The rows and the assertion do not change.
• The test still finds a freed cell value. With the values kept in a Vec<JSValue> (a local change), 47 of 64 ids are missing on debug and release builds. The old test: 62.
• Verified: test/js/bun/console/console-table.test.ts on a debug build, 32 pass in 10 of 10 runs, and the test takes 0.18 to 0.30 s. Also a release build.

Background

• console.table formats each cell one time, in its width pass (console.table: invoke each cell getter once #32924). An earlier revision kept the cell values for the render pass, and the test guards that design against a GC between the passes.
• The test sees a freed value through its memory: the object of a later getter takes it, and the earlier row shows a later id.
• Considered N = 8 with a GC in each getter: 8 GCs, and 6 missing ids. Considered isDebug: a release build finds the bug with 4 GCs too.

Notes

All measurements: Linux x64, 12 cores. The debug build is bun bd of main 449877f (debug, ASAN). The release build for the timings is 367d939.

Timings of the test

	before	after
debug build, whole file, 10 runs each	2.33 to 3.61 s	0.18 to 0.30 s
debug build, 12 copies of the file at once	7.05 to 9.75 s, 12 of 12 time out	0.25 to 0.77 s, 12 of 12 pass
debug build, 36 busy loops on the 12 cores, 6 runs each	3.51 to 4.41 s	0.45 to 0.88 s
release build, 5 runs each	373 to 450 ms	20 to 31 ms

The whole file on the debug build: 4.45 to 5.70 s before, 2.32 to 2.42 s after.

Cost of one full GC. Bun.gc(true) in the getter takes 24 to 155 ms on the debug build, with a median of 41 to 61 ms (two runs of 64 calls). fullGC() from bun:jsc costs the same. Bun.gc(false) only requests a concurrent collection, so it cannot replace the call.

The local change that frees cell values. TablePrinter::collect_row (src/jsc/ConsoleObject.rs) formats a cell as soon as it reads it. For the experiment it stores the value, and print_table formats the stored values after the last row:

// collect_row, in place of the format_cell call
self.pending.push((self.row_idx, slot, value)); // Vec<(usize, usize, JSValue)>

// print_table, after the width pass
for (row_idx, slot, value) in core::mem::take(&mut self.pending) {
    let cell = self.format_cell::<ENABLE_ANSI_COLORS>(&mut cell_text, value)?;
    columns[slot + 1].width = columns[slot + 1].width.max(cell.width);
    rows[row_idx].cells[slot] = Some(cell);
}

The GC does not scan the Vec, so a full GC in a later getter frees the stored objects. REVIEW.md names this bug: "Root or copy every JSValue held beyond the current call". Two builds have the change: the debug build, and a release build of 449877f (--profile=release --lto=off).

What the new test prints with the local change. Both builds print the same table. Rows 0 to 46 show the id of a row from 47 to 63:

0:48 1:49 2:50 ... 14:62 15:47 16:48 ... 31:63 32:48 ... 46:62 47:47 48:48 ... 63:63

The assertion fails with the ids 0 to 46 in missing. In each build the whole file has 1 fail and 31 pass, so no other test in the file finds this bug.

Ids missing from the table with the local change. One scratch test for each row of this table. Each count is the same in every run (3 or more runs for each build).

GC in getter	GCs	debug build	release build
each of 64 (the old test)	64	62	62
0, 16, 32, 48 (the new test)	4	47	47
0, 8, ..., 56	8	55	55
0, 32	2	31	31
8, 24, 40, 56	4	55	39
15, 31, 47, 63	4	62	32
31, 63	2	62	1
32 only	1	31	0
63 only	1	62	0
each of 8 rows (N = 8)	8	6	6
each of 2 rows (N = 2)	2	0	0

Why getter 0 runs a GC. On the release build, the values that the getters return before the first GC do not show the bug. With GCs in getters 8, 24, 40 and 56, ids 0 to 7 are not missing. One GC in getter 32, or in getter 63, finds nothing. With a GC in getter 0, the two builds give the same count: all the rows before the last GC, less one.

CI.

• .buildkite/ci.ts has release lanes and one asan lane, which is a release build with ASAN (--asan=on). No lane tests a debug build.
• The runner passes --timeout=90000 to bun test, and --timeout=270000 on the ASAN lane (scripts/runner.node.ts:1167, :267). The 5 s default applies only to a direct bun test, for example bun bd test.
• The ASAN lane ran this file in 0.63 s in build 131281. test/expected-durations.json has 1160 ms for the asan column and 140 ms for the default lane.
• No annotation (error, flaky, or new flaky) names this file in 989 builds: 389 builds from 132751 to 133153 (25 and 26 October), and a sample of 600 builds from 127157 to 132250 (17 to 24 October).

History.

• console.table: invoke each cell getter once #32924 added the test in its commit 31564bd, when each cell value was kept in a MarkedArgumentBuffer for the render pass. The final revision formats each cell in the width pass and keeps no value.
• The test does not fail on a build without console.table: invoke each cell getter once #32924, as the body of console.table: invoke each cell getter once #32924 says. On bun 1.3.14, the last release before it, 8 tests of console.table reads each cell once fail and this test passes, with the old test and with the new test.
• A MarkedArgumentBuffer holds 8 values inline. The new test keeps 64 rows, so a design with such a buffer still goes past its inline storage before the second GC.
• Six open PRs report this timeout on a debug build as a failure that their change does not cause: console, bun:test: print a native function that is not a constructor as a function #44214, Bun.inspect.table: read the options after an undefined or null properties argument #45027, console: print [Circular] for %j of a cyclic value #45325, Remove dead code from the code generators, the build scripts, and ten crate roots #45709, console.table: log only the first argument when it is not an object #46313, console.table: skip properties entries whose key is a Symbol #46986.

[coderabbitai]

Warning

Review limit reached

You've used all free OSS reviews for now. Wait for the free limit to reset to keep reviewing this public repository.

Next included review available in 1 minute.

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

• Source mode: Auto
• Repository: oven-sh/bun
• Branch: robobun/f666cf36/console-table-gc-test-debug-timeout
• File: .coderabbit.yaml
• Commit: 085b985
• Layers: Repository YAML
• Review profile: ASSERTIVE
• Plan: Advanced
• Run ID: bdd0835b-ac77-4533-a97f-9d6ad8f93fdf

Commits

Reviewing files that changed from the base of the PR and between 449877f and 085b985.

Files selected for processing: 1

• test/js/bun/console/console-table.test.ts

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

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.

❤️ Share

• X
• Mastodon
• Reddit
• LinkedIn

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

[robobun]

Status: ready for review. Waiting for CI.

Reproduction.

• On main 449877f, with a debug build: bun bd test test/js/bun/console/console-table.test.ts, 10 runs. cell values survive a full GC between the width and render passes takes 2.33 to 3.61 s of the 5 s limit.
• With 12 copies of the file at once on 12 cores, the test fails in 12 of 12 runs: this test timed out after 5000ms, after 7.05 to 9.75 s.

With this branch.

• The same command gives 32 pass in 10 of 10 runs, and the test takes 0.18 to 0.30 s.
• With 12 copies of the file at once, the test passes in 12 of 12 runs, in 0.25 to 0.77 s.

PR: #47004

[robobun]

^{Updated 11:56 AM PT - Oct 26th, 2026}

✅ @robobun, your commit 085b985445aeb234579c568f90a6f4eaafde2b91 passed in Build #133180! 🎉

---

🧪   To try this PR locally:

bunx bun-pr 47004

That installs a local version of the PR into your bun-47004 executable, so you can run:

bun-47004 --bun

[claude]

Code review found no issues

No high-confidence issues detected in this change.