Exit with code 2 when a plugin config file fails to load by WooWan · Pull Request #1947 · webpro-nl/knip

WooWan · 2026-08-15T12:56:33Z

When a plugin config file (e.g. vite.config.ts) fails to load, Knip logs the error and continues, but the exit code previously ignored it.

Before:

Scenario	Exit
load error, no other findings	0
load error, with findings	1
load error, `--no-exit-code`	0

After: every case above exits 2.

Analysis still runs to completion and findings are still reported. --no-exit-code continues to suppress lint findings, but not config load errors, consistent with how Knip already handles failures in its own configuration.

The existing CLI test was only green because its fixture also had unrelated findings. A new load-error-only fixture verifies that a config load error alone exits 2.

Docs are updated accordingly.

pkg-pr-new · 2026-08-15T12:57:10Z

Open in StackBlitz

npm i https://pkg.pr.new/knip@1947

npm i https://pkg.pr.new/@knip/language-server@1947

npm i https://pkg.pr.new/@knip/mcp@1947

commit: f37936a

WooWan · 2026-08-27T05:36:21Z

Although this is a bug fix, changing the exit code could cause existing services to suddenly start failing after a minor or patch upgrade. It may therefore be safer to ship this in the next major release, but I’ll leave that decision to the maintainers. 🙇

webpro

Thanks for the pull request, and for tracking down why the existing test was green.

The gap is real. With the knip.json from this PR in place, main reports the failure and still exits 0:

$ knip --no-progress
ERROR: Error loading vite.config.ts (Cannot find module './missing-bootstrap-output.js')
ERROR: Please fix or visit https://knip.dev/reference/known-issues
$ echo $?
0

Without it the exit code comes from the unused vite devDependency, the unlisted binary and the unresolved import, so the CLI test never proved anything about the load failure. Your reading is correct.

On the release question from your comment: exiting non-zero here has been the intended behavior since 0914bd383 added that test, so this is a bug fix rather than a breaking change and does not need to wait for a major.

The code should be 2 rather than 1 though, and --no-exit-code should not suppress it. A vite.config.ts that fails to load is not a finding about the user's code. It means the analysis input was incomplete, so every remaining result may be wrong. Knip already treats that class this way: a knip.config.ts that fails to load exits 2 even with --no-exit-code, because the error escapes to the catch block in cli.ts, which sets the code regardless of the flag. A plugin config file is the same failure, only non-fatal, and reference/cli.md already describes 2 as "Knip did not run successfully due to bad input or internal error". Anyone who cannot fix the file keeps the documented escape hatch of overriding the plugin's config, which skips the load attempt entirely.

That moves the assertions in load-error.test.ts: 1 becomes 2 in the CLI test and in the cache test, and the --no-exit-code case expects 2 rather than 0, with a name to match.

P2 The exit code table in reference/cli.md and the "Exceptions from config files" section in reference/known-issues.md both need the new behavior: a plugin config file that fails to load exits 2, and --no-exit-code does not suppress it.

The branch also conflicts with main in cli.ts, which now reads options.maxIssues instead of parsing the argument inline. Merging main into the branch updates the pull request without a force push.

webpro · 2026-09-02T09:31:30Z

              } catch (error) {
                if (!(error instanceof Error)) throw error;
                hasLoadConfigError = true;
+                if (isLoaderError(error)) hasConfigLoadErrors = true;


P1 This narrows to LoaderError, but the try block also covers plugin.resolveConfig and filterTransitiveDependencies, and both logError calls run for anything they throw. A plugin throwing a plain Error prints the same two lines, discards the cache entry, and still exits 0.

hasLoadConfigError on the line above already carries this fact for the cache. Set the outer flag from that one and drop the isLoaderError import. The cache already treats every error here as a failed load; the exit code should agree.

Thanks for the detailed feedback! Your approach makes more sense. Applied in 4765a10.

webpro · 2026-09-02T09:31:30Z

    debugLogArray(wsName, 'Plugin dependencies', () => compact(inputs.map(input => toDebugString(input, rootCwd))));

-    return inputs;
+    return { inputs, hasConfigLoadErrors };


P1 Returning a boolean here changes the return type and ripples through build, run and cli. IssueCollector already carries run-level facts of exactly this kind in tagHints and configurationHints, collector is in scope in build, and WorkspaceWorker already takes callbacks such as handleInput. One more callback keeps runPlugins(): Input[] and leaves build and run untouched.

It also puts the fact in results, within reach of main() and the reporters. As submitted, the github-actions reporter has nothing to annotate, so the job turns red with no summary entry explaining why. That annotation can be a separate change.

Applied in 4765a10

webpro · 2026-09-02T09:31:30Z

    if (
      !args['no-exit-code'] &&
      (totalErrorCount > Number(args['max-issues'] ?? 0) ||
+        hasConfigLoadErrors ||


P1 This sits inside the !args['no-exit-code'] guard and behind totalErrorCount, so the flag suppresses the load failure and a findings exit can claim the code ahead of it. Both halves matter: incomplete input then reports as a clean run for anyone who has opted out of exit codes, and otherwise reports as ordinary findings.

Set it before this branch instead, after the reporters have run, reading the flag from results once it moves into IssueCollector:

if (hasConfigLoadErrors) { process.exitCode = 2; return; }

Applied in 4765a10.

webpro · 2026-09-02T09:31:30Z

@@ -0,0 +1,8 @@
+{


P3 This turns the fixture into the clean-report case and drops the case it covered before, a load failure alongside genuine issues. Both are worth keeping: a copy under fixtures/plugin-config/load-error-only carrying this file leaves the original fixture and its assertions intact.

Applied it in fec9d99.

# Conflicts: # packages/knip/src/cli.ts

WooWan · 2026-09-05T05:56:25Z

 export const finalizeConfigurationHints = (
-  results: Results,
+  results: Pick<Results, 'counters' | 'issues' | 'includedWorkspaceDirs' | 'configurationHints'>,


Results now requires hasConfigLoadErrors, which isn't in ReporterOptions. Narrowed this parameter to the four fields it uses. It can be extended when GitHub Actions annotations are added.

WooWan · 2026-09-05T06:18:37Z

Added docs in 81e0474. English isn't my first language, so feel free to tweak the wording as needed.

webpro

Every finding from the previous round is addressed, and the behavior holds up under direct checks rather than just the diff. A vite.config.ts that loads and then throws inside resolveConfig now exits 2, where the isLoaderError narrowing let it exit 0, and the same fixture with the load failure removed still exits 0. Thanks for the quick turnaround.

Two smaller things and a nit, all inline.

webpro · 2026-09-07T06:32:23Z


 export const finalizeConfigurationHints = (
-  results: Results,
+  results: Pick<Results, 'counters' | 'issues' | 'includedWorkspaceDirs' | 'configurationHints'>,


P2 This narrowing works around a gap that is better closed directly. hasConfigLoadErrors reaches Results but not ReporterOptions, which cli.ts assembles by hand, so reporters and preprocessors still cannot see it. Reaching them was half the reason for moving the fact into IssueCollector.

finalizeConfigurationHints is also exported from knip/session and consumed by packages/mcp-server/src/tools.js, so this changes a published signature as a side effect of the type gap.

Adding hasConfigLoadErrors to ReporterOptions and passing it through initialData closes the gap and makes this narrowing unnecessary.

Updated ReporterOptions to include hasConfigLoadErrors and restored the Results parameter type. After merging main, toReporterOptions() passes the flag through to reporters and preprocessors, applied it in 4c75139

webpro · 2026-09-07T06:32:23Z

              } catch (error) {
                if (!(error instanceof Error)) throw error;
                hasLoadConfigError = true;
+                this.handleConfigLoadError();


P2 Nothing pins this line. Restoring if (isLoaderError(error)) in front of it leaves all 1096 tests passing, and that guard is the behavior the fix exists to remove.

A config file that loads and then throws inside resolveConfig covers it. A vite.config.ts of export default { test: { projects: 42 } } is enough:

ERROR: Error loading vite.config.ts (config.test.projects is not iterable)

That exits 2 on this branch and 0 with the guard back.

Added a regression test using projects: 42 that expects exit code 2, including with --no-exit-code. Confirmed that restoring the isLoaderError guard makes both cases fail. f241e7c, f37936a

webpro · 2026-09-07T06:32:23Z

+  "ignoreBinaries": ["vite"],
+  "ignoreIssues": {
+    "vite.config.ts": ["unresolved"],
+    "missing-bootstrap-output.js": ["files"]


P3 This entry does nothing here. It carries the cache recovery test in the sibling fixture, which writes missing-bootstrap-output.js partway through; nothing creates that file in this fixture. Removing the line leaves the output identical.

Thanks, applied it 302da6a

Include hasConfigLoadErrors in ReporterOptions and the data assembled by the CLI. Restore the Results parameter type for finalizeConfigurationHints and pass all required fields through its caller.

…fig-load-error

webpro · 2026-09-09T05:46:21Z

Thank you! 🌱

webpro · 2026-09-09T06:40:39Z

🚀 This pull request is included in v6.35.1. See Release 6.35.1 for release notes.

Using Knip in a commercial project? Please consider becoming a sponsor.

… config knip loads plugin configs through its own CJS transpile, which cannot call ESM-only @vitejs/plugin-react, so loading electron.vite.config.js has been failing all along. knip >= 6.35.1 turns that failure into exit code 2 (webpro-nl/knip#1947), which breaks the Check workflow on the 6.36.0 bump. Disable both the vite and electron-vite plugins (electron-vite auto-detects the same config file) and add the config as an entry so @vitejs/plugin-react and ./build/global-externals.js are still counted as used. Co-authored-by: Leo Capuano <60476114+leo-capvano@users.noreply.github.com> Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>

changwan-wu added 2 commits August 15, 2026 21:27

Exit non-zero on plugin config load errors

f75e650

update test and fixture

25a5bf8

webpro requested changes Sep 2, 2026

View reviewed changes

changwan-wu added 6 commits September 2, 2026 18:52

Merge remote-tracking branch 'upstream/main' into fix/config-load-error

5308406

# Conflicts: # packages/knip/src/cli.ts

Record plugin config load errors in IssueCollector and exit with code 2

4765a10

Add load-error-only fixture for load errors without findings

fec9d99

Narrow finalizeConfigurationHints to the fields it reads

c65c3e9

Clarify exit codes for config load errors in docs

81e0474

remark

7d23941

WooWan commented Sep 5, 2026

View reviewed changes

WooWan requested a review from webpro September 5, 2026 06:04

WooWan changed the title ~~Exit non-zero on plugin config load errors~~ Exit with code 2 when a plugin config file fails to load Sep 6, 2026

webpro requested changes Sep 7, 2026

View reviewed changes

changwan-wu added 5 commits September 9, 2026 13:00

Pass config load errors to reporters and preprocessors

4c75139

Include hasConfigLoadErrors in ReporterOptions and the data assembled by the CLI. Restore the Results parameter type for finalizeConfigurationHints and pass all required fields through its caller.

add resolve-config-error test

f241e7c

unnecessary test config

302da6a

Merge branch 'main' of https://github.com/webpro-nl/knip into fix/con…

bd8e0e0

…fig-load-error

Align config resolution error tests with existing test style

f37936a

webpro merged commit 37b2642 into webpro-nl:main Sep 9, 2026
30 checks passed

claude Bot mentioned this pull request Sep 16, 2026

Update dependency knip to v6.36.0 freelensapp/freelens-agentbridge-extension#48

Open

1 task

Conversation

WooWan commented Aug 15, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

pkg-pr-new Bot commented Aug 15, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

WooWan commented Aug 27, 2026

Uh oh!

webpro left a comment

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

WooWan Sep 5, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

WooWan Sep 5, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

WooWan Sep 5, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

WooWan commented Sep 5, 2026

Uh oh!

webpro left a comment

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

WooWan Sep 9, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Uh oh!

webpro commented Sep 9, 2026

Uh oh!

webpro commented Sep 9, 2026

Uh oh!

Reviewers

Uh oh!

Assignees

Labels

Projects

Milestone

Development

Uh oh!

3 participants

WooWan commented Aug 15, 2026 •

edited

Loading

pkg-pr-new Bot commented Aug 15, 2026 •

edited

Loading

WooWan Sep 5, 2026 •

edited

Loading

WooWan Sep 5, 2026 •

edited

Loading

WooWan Sep 5, 2026 •

edited

Loading

WooWan Sep 9, 2026 •

edited

Loading