fix(cli): traverse parent directories to find vite.config.ts in vp pack

[kazupon]

resolve #942

[netlify]

✅ Deploy Preview for viteplus-preview canceled.

Name	Link
🔨 Latest commit	925fd97
🔍 Latest deploy log	https://app.netlify.com/projects/viteplus-preview/deploys/69bfb8c68f584c00082cf550

[kazupon]

@codex review

[chatgpt-codex-connector]

Codex Review: Didn't find any major issues. More of your lovely PRs please.

ℹ️ About Codex in GitHub

Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you

• Open a pull request for review
• Mark a draft as ready
• Comment "@codex review".

If Codex has suggestions, it will comment; otherwise it will react with 👍.

Codex can also answer questions or update the PR. Try commenting "@codex address that feedback".

[fengmk2]

What about other commands? Does the build need to consider similar logic?

[kazupon]

Good point!

I investigated how each command resolves vite config:

Command	Config resolution	Call site	traverseUp status
vp pack	resolveViteConfig (JS)	pack-bin.ts:131	OK: Added in this PR
vp staged	resolveViteConfig (JS)	staged/bin.ts:143	NG: Should add
vp build / vp dev	delegate::execute → Vite's resolveConfig	cli.rs:1876 → Vite	NG: Should add
vp lint / vp fmt	resolveUniversalViteConfig (Rust→JS)	cli.rs → bin.ts:92	NG: Needs investigation
vp run	Rust workspace-aware task runner	cli.rs:1262	OK: Already handled

Since vite+ is the unified toolchain and users interact through vp commands, config resolution (including traverseUp) should be vite+'s responsibility for all commands, not just vp pack.

For vp build/vp dev, we can find the config via findViteConfigUp and pass it as --config to Vite.
For vp staged, we can add { traverseUp: true } to the existing resolveViteConfig call.

[kazupon]

One thing I noticed during this investigation, the vite config resolution layer is currently split across JS and Rust:

• JS side: vp pack, vp staged → resolveViteConfig
• Rust→JS side: vp lint, vp fmt, vp run → VitePlusConfigLoader → resolveUniversalViteConfig
• Vite itself: vp build, vp dev → Vite's internal resolveConfig

The Rust side handles config resolution for vp lint/vp fmt/vp run because oxlint, oxfmt, and vite-task are Rust binaries that need the config.
This is reasonable, but the split means that cross-cutting features like traverseUp need to be applied in multiple places, which can lead to inconsistent behavior between commands.

Since vite+ unifies tool configuration through vite.config.ts (a JS/TS file), it might be worth considering consolidating config resolution to the JS side in the long term.
The Rust side's vite_static_config (static analysis) could remain as a performance optimization with JS as the fallback, which is already the current design.

The solution to this root cause goes well beyond the scope of this issue.
we should discuss it in a separate issue. /cc @sapphi-red

[chatgpt-codex-connector]

💡 Codex Review

vite-plus/packages/cli/src/pack-bin.ts

Lines 139 to 142 in ec46391

	const packConfigs = Array.isArray(viteConfig.pack)
	? viteConfig.pack
	: [viteConfig.pack ?? {}];
	for (const packConfig of packConfigs) {

Avoid building every root pack entry from a package subdirectory

If the newly discovered workspace-root vite.config.* contains pack: [...], this code will execute every entry even when vp pack is launched from a single package directory. There is no cwd-based narrowing before iterating viteConfig.pack, so a monorepo script like vp run foo#build can unexpectedly build sibling packages unless the user manually adds -F; that defeats the point of making package-local invocations reuse the parent config.

ℹ️ About Codex in GitHub

Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you

• Open a pull request for review
• Mark a draft as ready
• Comment "@codex review".

If Codex has suggestions, it will comment; otherwise it will react with 👍.

Codex can also answer questions or update the PR. Try commenting "@codex address that feedback".

[kazupon]

@codex

If the newly discovered workspace-root vite.config.* contains pack: [...], this code will execute every entry even when vp pack is launched from a single package directory. There is no cwd-based narrowing before iterating viteConfig.pack, so a monorepo script like vp run foo#build can unexpectedly build sibling packages unless the user manually adds -F; that defeats the point of making package-local invocations reuse the parent config.

As I understand it, tasks are cached by vite-task, the task runner within vp, so they will not be executed again.
Verify test project is here:
https://github.com/kazupon/vp-942-array-test

[chatgpt-codex-connector]

To use Codex here, create an environment for this repo.

[kazupon]

I reconsidered the behavior of vp staged, vp lint, and vp fmt when run from a subpackage without a vite.config.ts.

vp staged is called from the pre-commit hook (.vite-hooks/pre-commit), which always runs from the repository root.
vp staged is unlikely to be called from a subpackage directory rarely.
I think that vp staged may not need traverseUp.

For vp lint / vp fmt, when run directly from a subpackage without a vite.config.ts, they fall back to default settings (no custom rules).
Fortunately, these commands should not result in an error and should run by default. If that happens, the user will realize that no configuration has been set and will configure vite.config.

I'll keep traverseUp scoped to vp pack for this PR.
The other commands can be addressed in a follow-up if needed.

[fengmk2]

snap tests result change after vite-task update. I will fix it before merge.

[fengmk2]

snap tests result change after vite-task update. I will fix it before merge.

follow up here #1095