Forward CLI extra args only to explicitly requested tasks

[branchseer]

What changes for users

Arguments typed after a task name on vp run are now forwarded only to that task. Tasks pulled in via dependsOn no longer receive them.

Before

Given vite-task.json:

{
  "tasks": {
    "test": { "command": "vitest", "dependsOn": ["build"] },
    "build": { "command": "tsc" }
  }
}

Running:

vp run test some-filter

…would invoke:

• test → vitest some-filter ✅
• build → tsc some-filter ❌ — tsc has no idea what some-filter means and fails (or worse, silently interprets it as a source file).

After

The same command now invokes:

• test → vitest some-filter ✅
• build → tsc ✅ — runs with its own config, unaffected by the caller's CLI.

This matches the intuitive mental model most users already had: "positional args I type after test are for test." It also fixes a subtle cache-invalidation side effect — dependency tasks used to incorporate the caller's extra args into their cache key, so vp run test foo and vp run test bar each produced a distinct build cache entry even though build ran identically in both cases. Dependency tasks now cache under a stable key regardless of what the caller typed.

Scope

Only explicit dependsOn dependencies are affected. Tasks pulled in via topological (-r, -t) or package-graph expansion still receive the extras, because each of those packages runs the same task name the user typed — vp run -r test some-filter still forwards some-filter to every selected package's test, which is the intended behavior.

No config changes or flags are required; the new behavior applies unconditionally.

---

Implementation notes

• TaskExecutionGraph (in crates/vite_task_graph/src/query/mod.rs) changed from a type alias for DiGraphMap<TaskNodeIndex, ()> into a small pub struct with two public fields: the same graph, plus a requested: FxHashSet<TaskNodeIndex> tracking the stage-2 (user-requested) nodes.
• map_subgraph_to_tasks populates requested; add_dependencies (stage 3, dependsOn expansion) deliberately does not.
• plan_query_request (in crates/vite_task_plan/src/plan.rs) no longer sets extra_args globally on the planning context. Instead, it duplicates the context per task and sets either the real extra_args or a shared empty Arc<[Str]> based on requested.contains(&task_index).
• New plan snapshot fixture extra-args-not-forwarded-to-depends-on records the observable behavior: test's spawn args and cache key include some-filter, while build's do not.

Fixes #324

https://claude.ai/code/session_01Hay4ywHGNEzjuh1WW5SnBr

[Copilot]

Pull request overview

This PR fixes vp run <task> <extra-args...> behavior so that extra CLI args are forwarded only to tasks explicitly requested by the user, and not to tasks that are included solely via dependsOn expansion (fixes #324).

Changes:

• Extend the task query result to track which task nodes were explicitly requested vs. added via dependsOn.
• Apply extra_args per task during planning based on that “requested” set, using an empty args slice for dependency-only tasks.
• Add a snapshot fixture verifying extra args reach only the requested task and do not reach its dependsOn dependency.

Reviewed changes

Copilot reviewed 8 out of 8 changed files in this pull request and generated 2 comments.

File	Description
crates/vite_task_graph/src/query/mod.rs	Changes TaskExecutionGraph to track both the graph and explicitly requested nodes.
crates/vite_task_plan/src/plan.rs	Applies extra args per task based on the requested set while building the execution graph.
crates/vite_task_plan/tests/plan_snapshots/**	Adds a fixture + snapshots proving extra args aren’t forwarded to dependsOn tasks.
CHANGELOG.md	Documents the user-visible behavior change for vp run extra args forwarding.

---

💡 Add Copilot custom instructions for smarter, more guided reviews. Learn how to get started.

[fengmk2]

bugfix includes breaking changes 😄