fix: use try_init() to avoid SetGlobalDefaultError panic by fengmk2 · Pull Request #9078 · rolldown/rolldown

fengmk2 · 2026-04-12T13:42:42Z

Summary

Replace .init() with .try_init().ok() in rolldown_devtools::DebugTracer::init() and all 3 call sites in rolldown_tracing::try_init_tracing()
Prevents SetGlobalDefaultError panic when rolldown is embedded in a host binary that has already installed a global tracing subscriber

Context

When rolldown is bundled into a host binary (e.g., vite-plus bundles rolldown_binding via pub extern crate), the host may install a global tracing subscriber during NAPI module init. Later, when DebugTracer::init() runs (triggered by build.rolldownOptions.devtools: {}), it calls .init() which panics:

thread '<unnamed>' panicked at tracing-subscriber-0.3.23/src/util.rs:94:14:
failed to set global default subscriber: SetGlobalDefaultError("a global default trace dispatcher has already been set")

.try_init() returns Result instead of panicking, so a previously-installed subscriber is silently accepted.

Ref: voidzero-dev/vite-plus#1356

Test plan

cargo check -p rolldown_devtools -p rolldown_tracing passes
CI passes

When rolldown is embedded in a host binary (e.g., vite-plus) that has already installed a global tracing subscriber, DebugTracer::init() and try_init_tracing() panic because .init() calls set_global_default() which fails if a subscriber is already set. Replace .init() with .try_init().ok() at all 4 call sites so a previously-installed subscriber is silently accepted instead of causing a panic. Ref: voidzero-dev/vite-plus#1356

netlify · 2026-04-12T13:42:59Z

✅ Deploy Preview for rolldown-rs canceled.

Name	Link
🔨 Latest commit	`ac096e6`
🔍 Latest deploy log	https://app.netlify.com/projects/rolldown-rs/deploys/69dba1574495080008b8fdb1

codspeed-hq · 2026-04-12T13:49:25Z

Merging this PR will not alter performance

✅ 4 untouched benchmarks
⏩ 10 skipped benchmarks¹

_{Comparing fengmk2:fix/tracing-try-init (ac096e6) with main (b3843b3)}

10 benchmarks were skipped, so the baseline results were used instead. If they were deleted from the codebase, click here and archive them to remove them from the performance reports. ↩

Copilot

Pull request overview

This PR updates tracing initialization to avoid panicking when a global tracing subscriber has already been installed by an embedding host binary.

Changes:

Replaces tracing_subscriber::util::SubscriberInitExt::init() with try_init() in rolldown_tracing::try_init_tracing().
Replaces init() with try_init() in rolldown_devtools::DebugTracer::init().
Silently ignores SetGlobalDefaultError instead of panicking during devtools/tracing setup.

Reviewed changes

Copilot reviewed 2 out of 2 changed files in this pull request and generated 4 comments.

File	Description
`crates/rolldown_tracing/src/lib.rs`	Switches tracing subscriber setup to `try_init()` across output modes to avoid `SetGlobalDefaultError` panics.
`crates/rolldown_devtools/src/init_tracing.rs`	Switches devtools tracing subscriber setup to `try_init()` to avoid global subscriber re-install panics.

Copilot · 2026-04-12T15:14:34Z

      tracing_subscriber::registry()
        .with(Targets::from_str(&env_var).unwrap())
        .with(chrome_layer.with_filter(filter_for_removing_devtools_event))
-        .init();
+        .try_init()
+        .ok();


try_init().ok(); still discards a #[must_use] return value (both Result and the Option from .ok()), which will trigger unused_must_use warnings. Also, this branch returns Some(guard) even if try_init() fails (e.g., global subscriber already set), which can leave an unused guard/file handle around. Capture the Result from try_init() (e.g., let _ = ... or if ... .is_ok() { ... }) and only return the guard when initialization actually succeeded (or short-circuit before building the chrome layer when a global subscriber is already installed).

Copilot · 2026-04-12T15:14:34Z

      tracing_subscriber::registry()
        .with(filter_for_removing_devtools_event)
        .with(Targets::from_str(&env_var).unwrap())
        .with(
          fmt::layer().pretty().with_span_events(FmtSpan::NONE).with_level(true).with_target(false),
        )
-        .init();
+        .try_init()
+        .ok();
      tracing::debug!("Tracing initialized");


try_init().ok(); discards a #[must_use] value and is likely to produce an unused_must_use warning. Additionally, tracing::debug!("Tracing initialized") will run even if try_init() fails (e.g., subscriber already set), which makes the log message inaccurate; consider logging only on Ok(()) (or logging a different message on Err).

Copilot · 2026-04-12T15:14:34Z

      tracing_subscriber::registry()
        .with(filter_for_removing_devtools_event)
        .with(Targets::from_str(&env_var).unwrap())
        .with(fmt::layer().pretty().with_span_events(FmtSpan::CLOSE | FmtSpan::ENTER))
-        .init();
+        .try_init()
+        .ok();
      tracing::debug!("Tracing initialized");
      None


Same issue as the other branch: try_init().ok(); still drops a #[must_use] value (warning), and the unconditional Tracing initialized debug log is misleading when try_init() returns Err because a global subscriber is already installed.

Copilot · 2026-04-12T15:14:34Z

    tracing_subscriber::registry()
      .with(DevtoolsLayer.with_filter(devtools_event_filter))
      .with(fmt::layer().event_format(DevtoolsFormatter))
-      .init();
+      .try_init()
+      .ok();


try_init().ok(); still discards a #[must_use] return value (the Option from .ok()), so this will likely trigger an unused_must_use warning during cargo check. Prefer capturing the result explicitly (e.g., let _ = ...try_init();) or handling/logging the Err case.

sapphi-red · 2026-04-13T02:44:57Z

refs #8698 (comment)

sapphi-red · 2026-04-13T02:45:51Z

I guess this would disable the debug file output is used by the devtools feature and will break the feature.

fengmk2 · 2026-04-13T02:48:51Z

@sapphi-red Indeed, we need to think of a better solution. Do you have any suggestions?

sapphi-red · 2026-04-13T02:55:31Z

I'm not familiar with the code around here and not sure how it's working in details. @hyf0 Do you have any ideas?

fengmk2 · 2026-04-15T06:37:19Z

After communicating with @hyf0, I will handle it at the Vite+ side for now to avoid simultaneous registration. The long-term solution needs @hyf0 to take another look.

hyf0 · 2026-04-15T07:05:01Z

For the future plan:

rolldown/meta/design/devtools.md

Line 263 in 0af9533

#### `tracing` Scoped Subscriber Mechanisms

Should be able to solve it without causing conflict with other crates.

…s is enabled (#1364) ## Summary Fixes a panic that occurs when `build.rolldownOptions.devtools` is set in `vite.config.ts` (commonly via `@vitejs/devtools`): ``` Rolldown panicked. This is a bug in Rolldown, not your code. thread '<unnamed>' panicked at tracing-subscriber-0.3.23/src/util.rs:94:14: failed to set global default subscriber: SetGlobalDefaultError("a global default trace dispatcher has already been set") ``` ## Root cause The vite-plus NAPI binary bundles `rolldown_binding` (including `rolldown_devtools`) into a single `.node` file. Two paths inside this same binary each try to install a global tracing subscriber: 1. **NAPI module load** → `vite_shared::init_tracing()` — previously installed a no-op subscriber (empty `Targets` filter) even when `VITE_LOG` was unset, occupying the global default slot. 2. **Build with devtools enabled** → `DebugTracer::init()` — tries to install its own subscriber and panics because the slot is already taken. ## Fix - Skip subscriber installation entirely when `VITE_LOG` is not set, leaving the global default slot free for `DebugTracer::init()` (the common case — no one pays for tracing they didn't ask for). - Use `.try_init().ok()` instead of `.init()` so a double-init still fails gracefully rather than panicking. ## Test plan - [x] Reproduced the panic at `/tmp/repro-1356` with `@vitejs/devtools@0.1.13` + `build.rolldownOptions.devtools: {}`. - [x] Rebuilt with `RELEASE_BUILD=1 pnpm bootstrap-cli`; `npx vp build` now completes successfully. - [x] `VITE_LOG=debug vp build` still prints traces (subscriber is installed when requested). ## Related Upstream rolldown also makes the same `.init()` → `.try_init()` change as defense-in-depth: rolldown/rolldown#9078. That upstream fix is not required for this patch — the vite-plus fix here is sufficient on its own. Closes #1356

fengmk2 mentioned this pull request Apr 12, 2026

fix(tracing): avoid SetGlobalDefaultError panic when rolldown devtools is enabled voidzero-dev/vite-plus#1364

Merged

3 tasks

hyf0 requested a review from Copilot April 12, 2026 15:11

Copilot started reviewing on behalf of hyf0 April 12, 2026 15:11 View session

Copilot AI reviewed Apr 12, 2026

View reviewed changes

fengmk2 marked this pull request as draft April 13, 2026 02:48

fengmk2 closed this Apr 15, 2026

Conversation

fengmk2 commented Apr 12, 2026

Summary

Context

Test plan

Uh oh!

netlify Bot commented Apr 12, 2026

✅ Deploy Preview for rolldown-rs canceled.

Uh oh!

codspeed-hq Bot commented Apr 12, 2026

Merging this PR will not alter performance

Footnotes

Uh oh!

Copilot AI left a comment

Choose a reason for hiding this comment

Pull request overview

Reviewed changes

Uh oh!

Copilot AI Apr 12, 2026

Choose a reason for hiding this comment

Uh oh!

Copilot AI Apr 12, 2026

Choose a reason for hiding this comment

Uh oh!

Copilot AI Apr 12, 2026

Choose a reason for hiding this comment

Uh oh!

Copilot AI Apr 12, 2026

Choose a reason for hiding this comment

Uh oh!

sapphi-red commented Apr 13, 2026

Uh oh!

sapphi-red commented Apr 13, 2026

Uh oh!

fengmk2 commented Apr 13, 2026

Uh oh!

sapphi-red commented Apr 13, 2026

Uh oh!

fengmk2 commented Apr 15, 2026

Uh oh!

hyf0 commented Apr 15, 2026

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

4 participants