feat: auto output restoration & env detection by wan9chi · Pull Request #410 · voidzero-dev/vite-task

wan9chi · 2026-05-28T13:40:13Z

vp run now supports two cache-correctness features that don't require listing every output/env in the task config:

Auto output restoration. With fspy on, the cache archives the files the task wrote and restores them on a hit — no output: [...] config needed. Tools can opt scratch paths out per-path via ignoreOutput().
Auto env detection. Tools can fetch env values via getEnv / getEnvs and have the runner register the read as a cache-key dependency in the same call — no env: [...] config needed.

Both rely on a new IPC channel between the runner and the tools it spawns. The JS-side surface — the @voidzero-dev/vite-task-client package — shipped in #409 and is now on npm. This PR is the runner side: the IPC protocol, the napi addon the JS wrapper loads at runtime, the executor reshape that hosts a per-task server, the cache integration that consumes the recorded reports, and end-to-end tests.

disableCache() and ignoreInput() round out the tool-side API.

Reviewing

5 commits, one per logical layer — each builds and tests on its own, so commit-by-commit review is recommended:

#	Commit	What it adds
1	`feat(ipc): protocol + Rust transport`	`vite_task_ipc_shared` (wire format), `vite_task_server` (per-task async server), `vite_task_client` (sync blocking client). End-to-end integration tests.
2	`feat(client): napi binding`	`crates/vite_task_client_napi` — napi addon exposing the Rust client to JS as the `RunnerClient` returned from `load()`. (The npm wrapper itself shipped in #409.)
3	`refactor(execute): reshape for IPC server lifecycle`	Reshape `vite_task::session::execute` and supporting modules to host a per-task IPC server. The server is still a placeholder here — runner behavior is byte-for-byte identical to today's. Also folds in the blocking output-config refactor.
4	`feat(cache): integrate runner-aware IPC reports`	Wire up the real `serve(Recorder::new(env_map))`, embed the napi addon, inject `VP_RUN_NODE_CLIENT_PATH` into the child, and merge the resulting `Reports` into the post-run fingerprint and cache update.
5	`test(cache): real-vite integration fixtures + playground`	End-to-end fixtures exercising the full stack against a real Vite build.

Lays the groundwork for publishing `@voidzero-dev/vite-task-client` to npm. JS-side only — `src/index.js` (JSDoc as source of truth), `src/index.d.ts` regenerated from it, and the metadata required to publish. No Rust changes; no runner integration. The package is a thin wrapper with no `dependencies`: under a `vp run` task it loads the runner-provided napi addon at runtime, otherwise every method is a graceful no-op. ## Next steps 1. Publish `@voidzero-dev/vite-task-client@0.1.0` to npm. 2. Update the [Vite PR](vitejs/vite#22453) to depend on the published `@voidzero-dev/vite-task-client`. 3. Merge #410 — the remaining Rust implementation (IPC protocol + transport, napi binding, executor refactor, cache integration, e2e tests). Co-authored-by: Claude Opus 4.7 (1M context) <noreply@anthropic.com>

Add three new crates that together define the runner-tool IPC: - `vite_task_ipc_shared` — message types + wire format shared by both ends. Spawned tools tell the runner what they read, wrote, or cared about; the runner uses that to decide what to fingerprint in the cache. - `vite_task_server` — async server hosted in the runner. One server instance per task execution. - `vite_task_client` — synchronous blocking client used by spawned tools. The sync API is deliberate: most tools are JS, called one-method-at-a-time, and don't want a runtime imposed on them. The two sides are wired together end-to-end in `vite_task_server/tests/integration.rs`, so this PR is independently reviewable as "does the wire format and transport work correctly?" without depending on runner internals. Design notes: `docs/runner-task-ipc/`. Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>

Adds the JS surface that lets spawned Node tools talk to the runner over the IPC introduced in the previous PR: - `crates/vite_task_client_napi` — napi binding that exposes the Rust `Client` to JS as a `RunnerClient` returned from `load()`. - `packages/vite-task-client` — JS wrapper with `ignoreInput`, `ignoreOutput`, `disableCache`, `getEnv`, and `getEnvs`. Types live in `index.js` as JSDoc; `index.d.ts` is generated via `pnpm build-vite-task-client-types` and a CI staleness check fails the build if the committed `.d.ts` drifts from the source. The package is consumable as a no-op outside the runner: when `VP_RUN_NODE_CLIENT_PATH` isn't set, `load()` returns `null` and every exported function becomes a no-op (or returns `undefined`/`{}`). That means the wrapper is safe to add as a regular dependency from a tool that wants to opt in to runner-aware behavior without requiring its users to run under `vp`. Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>

Reshape `vite_task::session::execute` (and the supporting modules: `spawn`, `fingerprint`, `cache`, `tracked_accesses`, `event`, and the summary reporter) to accommodate a per-task IPC server. The server itself is a placeholder on this branch: instead of actually calling `vite_task_server::serve(...)`, we construct an empty `Recorder` whose driver future resolves immediately with no traffic, and bind a `StopAccepting::noop()`. The downstream plumbing — async-join of the child with the server, `Reports` flowing into post-run fingerprinting and into the cache update — is fully in place but sees only the empty `Reports`, so behaviour is byte-for-byte identical to today's runner. The follow-up wires up the real `serve(...)`, embeds the napi addon, and injects `VP_RUN_NODE_CLIENT_PATH` into the child. That PR's `serve(Recorder::new(env_map))` is the only call that changes from "future::ready(Ok(recorder))" to the real bind. Adds `StopAccepting::noop()` to `vite_task_server` for the same placeholder use case; it's a tiny helper that's also useful for tests that need a value of the type without running a server. Also folds in the unrelated-but-blocking output-config refactor (`output: Option<UserInputsConfig>` instead of `Option<Vec<UserOutputEntry>>`) — both sides now route through `ResolvedGlobConfig::from_user_config`. The 80-odd plan-snapshot updates are mechanical consequences of that type change. Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>

Replaces the previous PR's placeholder with the real `vite_task_server::serve(...)` call: per-task IPC server, napi addon embedded into the runner binary and materialized to disk on first use, `VP_RUN_NODE_CLIENT_PATH` injected into the child so the JS wrapper can `require()` it. Cache integration: `Reports` collected from the IPC drive `PostRunFingerprint` and the cache-update path — - `ignore_input` reads → excluded from input fingerprint - `ignore_output` writes → excluded from output archive - `tracked_envs` (single name) + `tracked_env_globs` → folded into the post-run fingerprint so a value change misses the cache - `disable_cache` → skips the cache-update path entirely (`ToolRequested`) - IPC server bind/runtime failure → `IpcServerError` cache disable End-to-end coverage via the `ipc_client_test` fixture set: one fixture per API method, each exercising the real Rust ↔ JS path and asserting the right cache behaviour. Adds `vtt` (test-only) helpers `grep_file` and `stat_file` that the fixtures need. Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>

Adds the real-vite e2e coverage: `vite_build_cache` exercises runner-aware cache behaviour through `vite build` against the companion vite PR (NODE_ENV-change invalidation, `envPrefix`-driven tracked-env set, `dist/` write restoration on cache hit), and `vite_dev_disable_cache` verifies `disableCache()` short-circuits the dev server's run from the cache. Wires the rest of the consumer-side bits needed for vite (consumed via the `github:` spec for vite-task-client) to install cleanly: - Adds `vite: catalog:` to `packages/tools/package.json` so `pnpm install` materialises `packages/tools/node_modules/.bin/vite` for the harness's `link_tools_packages(stage, ["vite"])` symlink setup. - Pins root `packageManager` to `pnpm@10.33.4` matching the vite PR's pnpm, removing a per-install version switch that produced an outer-vs-inner lockfile-settings mismatch when vite consumed vite-task as a git dep. - Drops the root `prepare: husky` so pnpm's git-dep flow doesn't invoke `pnpm pack`, which had been producing a lockfile `resolution` entry without the `path:` field — pnpm then extracted the *entire* vite-task monorepo into `node_modules/@voidzero-dev/vite-task-client/` instead of the subpath. - Pins the vite catalog entry to a commit-specific `pkg.pr.new` URL so future bumps require updating this line. - Sets `blockExoticSubdeps: false` so pnpm 11 doesn't reject the transitive `github:` spec. Refreshes the playground to use the published vite + the runner- aware tools. Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>

chatgpt-codex-connector

💡 Codex Review

Here are some automated review suggestions for this pull request.

Reviewed commit: fef1237ff3

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".

chatgpt-codex-connector · 2026-06-01T03:12:25Z

    pub inferred_inputs: HashMap<RelativePathBuf, PathFingerprint>,
+
+    /// Env vars observed via runner-aware IPC `getEnv` with `tracked: true`.
+    /// Key is the env name; value is the env value at execution time (or
+    /// `None` if unset). Validated at cache lookup by comparing against the
+    /// current parent env.
+    pub tracked_envs: BTreeMap<Str, Option<Str>>,
+


Bump the cache DB version for the new value schema

Adding these fields changes the wincode layout of CacheEntryValue because it embeds PostRunFingerprint, but ExecutionCache::load_from_path still treats existing user_version = 13 databases as current. On an upgrade with a populated v13 cache, a task whose cache key matches an old entry will try to deserialize the old shorter value into this new struct, causing cache lookup to error and the task to fail instead of running or treating it as a miss. Please bump/reset the cache version for this schema change.

Useful? React with 👍 / 👎.

chatgpt-codex-connector · 2026-06-01T03:12:25Z

+            let value = record.value.as_ref().and_then(|v| v.to_str().map(Str::from));
+            Some((Str::from(name_str), value))


Preserve non-UTF-8 tracked env values

On Unix, when a native/Rust tool tracks an env var whose value is not valid UTF-8, this conversion stores None, the same representation as an unset variable. Validation uses the same to_str() path, so changing between unset and a non-UTF-8 value, or between two different non-UTF-8 values, will still hit the cache even though the tool observed a different value. Since the protocol and Rust client carry OsStr/NativeStr, the fingerprint should use a byte-faithful representation or explicitly disable caching for such values.

Useful? React with 👍 / 👎.

This was referenced May 28, 2026

feat(client): vite-task-client JS package #409

Merged

feat(ipc): protocol + Rust transport #404

Closed

feat(client): napi binding + JS wrapper #405

Closed

refactor(execute): reshape for IPC server lifecycle #406

Closed

This was referenced May 28, 2026

feat(cache): integrate runner-aware IPC reports #407

Closed

test(cache): real-vite integration fixtures + playground #408

Closed

wan9chi force-pushed the feat/vite-task-client-js branch from 2fd44c0 to 7a668fd Compare May 28, 2026 13:55

wan9chi force-pushed the feat/runner-tools branch from 1fed8fa to 93aa7c1 Compare May 28, 2026 13:55

wan9chi force-pushed the feat/vite-task-client-js branch from 7a668fd to fcfd107 Compare May 28, 2026 14:13

wan9chi force-pushed the feat/runner-tools branch 2 times, most recently from 976fad0 to 9583e9d Compare May 28, 2026 14:20

wan9chi force-pushed the feat/vite-task-client-js branch from fcfd107 to 3061c61 Compare May 28, 2026 14:20

wan9chi force-pushed the feat/runner-tools branch from 9583e9d to 2def2db Compare May 28, 2026 14:24

wan9chi force-pushed the feat/vite-task-client-js branch from 3061c61 to 2bc64fb Compare May 28, 2026 14:24

Base automatically changed from feat/vite-task-client-js to main May 28, 2026 14:38

wan9chi and others added 5 commits May 28, 2026 22:39

wan9chi force-pushed the feat/runner-tools branch from 2def2db to fef1237 Compare May 28, 2026 14:39

wan9chi changed the title ~~feat(ipc): protocol + Rust transport~~ feat: runner-aware tools Jun 1, 2026

wan9chi marked this pull request as ready for review June 1, 2026 03:09

wan9chi changed the title ~~feat: runner-aware tools~~ feat: auto output restoration for vite build Jun 1, 2026

wan9chi changed the title ~~feat: auto output restoration for vite build~~ feat: auto output restoration & env detection Jun 1, 2026

chatgpt-codex-connector Bot reviewed Jun 1, 2026

View reviewed changes

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

feat: auto output restoration & env detection#410

feat: auto output restoration & env detection#410
wan9chi wants to merge 5 commits into
mainfrom
feat/runner-tools

wan9chi commented May 28, 2026 •

edited

Loading

Uh oh!

wan9chi commented May 28, 2026 •

edited

Loading

Uh oh!

socket-security Bot commented May 28, 2026 •

edited

Loading

Uh oh!

chatgpt-codex-connector Bot left a comment

Uh oh!

chatgpt-codex-connector Bot Jun 1, 2026

Uh oh!

chatgpt-codex-connector Bot Jun 1, 2026

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

1 participant

		let value = record.value.as_ref().and_then(\|v\| v.to_str().map(Str::from));
		Some((Str::from(name_str), value))

Conversation

wan9chi commented May 28, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Reviewing

Related

Uh oh!

wan9chi commented May 28, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

socket-security Bot commented May 28, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

chatgpt-codex-connector Bot left a comment

Choose a reason for hiding this comment

💡 Codex Review

Uh oh!

chatgpt-codex-connector Bot Jun 1, 2026

Choose a reason for hiding this comment

Uh oh!

chatgpt-codex-connector Bot Jun 1, 2026

Choose a reason for hiding this comment

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

1 participant

wan9chi commented May 28, 2026 •

edited

Loading

wan9chi commented May 28, 2026 •

edited

Loading

socket-security Bot commented May 28, 2026 •

edited

Loading