feat(cache): add output globs for cache restoration

Adds an `output` field to cached tasks: archives files matching the
configured globs after a successful run and restores them on cache hit.

Supports glob patterns, negative patterns, and `{pattern, base}` form
with explicit base directory. When `output` is omitted or empty, no
output archiving happens (matches prior behavior).

Schema version bumped to 12 (CacheEntryKey carries output_config,
CacheEntryValue carries output_archive). Old caches are reset on upgrade.

Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>

Author: branchseer

CHANGELOG.md

@@ -1,5 +1,6 @@
 # Changelog
 
+- **Added** `output` field for cached tasks: archives output files matching the configured globs after a successful run and restores them on cache hit. Patterns are relative to the package directory; supports negative patterns (e.g. `"!dist/cache/**"`) and `{pattern, base}` form for explicit base. ([#321](https://github.com/voidzero-dev/vite-task/pull/321))
 - **Fixed** Windows cached tasks can now run package shims rewritten through PowerShell; default env passthrough now preserves `PATHEXT` ([#366](https://github.com/voidzero-dev/vite-task/pull/366))
 - **Added** Platform support for targets without `input` auto-inference (e.g. Android). Tasks still run; those relying on auto-inference run uncached, with the summary noting that `input` must be configured manually to enable caching ([#352](https://github.com/voidzero-dev/vite-task/pull/352))
 - **Fixed** `vp run` no longer aborts with `failed to prepare the command for injection: Invalid argument` when the user environment already has `LD_PRELOAD` (Linux) or `DYLD_INSERT_LIBRARIES` (macOS) set. The tracer shim is now appended to any existing value and placed last, so user preloads keep their symbol-interposition precedence ([#340](https://github.com/voidzero-dev/vite-task/issues/340))

Cargo.lock

@@ -163,6 +163,7 @@ winsafe = { version = "0.0.27", features = ["kernel"] }
 xxhash-rust = { version = "0.8.15", features = ["const_xxh3"] }
 ntest = "0.9.5"
 terminal_size = "0.4"
+zstd = "0.13"
 
 [workspace.metadata.cargo-shear]
 ignored = [

Cargo.toml

@@ -29,6 +29,7 @@ rustc-hash = { workspace = true }
 serde = { workspace = true, features = ["derive", "rc"] }
 serde_json = { workspace = true }
 thiserror = { workspace = true }
+tar = { workspace = true }
 tokio = { workspace = true, features = [
   "rt-multi-thread",
   "io-std",
@@ -40,13 +41,15 @@ tokio = { workspace = true, features = [
 tokio-util = { workspace = true }
 tracing = { workspace = true }
 twox-hash = { workspace = true }
+uuid = { workspace = true, features = ["v4"] }
 vite_path = { workspace = true }
 vite_select = { workspace = true }
 vite_str = { workspace = true }
 vite_task_graph = { workspace = true }
 vite_task_plan = { workspace = true }
 vite_workspace = { workspace = true }
 wax = { workspace = true }
+zstd = { workspace = true }
 
 [dev-dependencies]
 tempfile = { workspace = true }

crates/vite_task/Cargo.toml

@@ -0,0 +1,61 @@
+//! Output archive creation and extraction using tar + zstd compression.
+
+use std::fs::File;
+
+use vite_path::{AbsolutePath, RelativePathBuf};
+
+/// Create a tar.zst archive from workspace-relative output file paths.
+///
+/// Files that no longer exist are silently skipped (the task may delete
+/// temporary files during execution).
+///
+/// # Errors
+///
+/// Returns an error if creating the archive file or adding entries fails.
+pub fn create_output_archive(
+    workspace_root: &AbsolutePath,
+    output_files: &[RelativePathBuf],
+    archive_path: &AbsolutePath,
+) -> anyhow::Result<()> {
+    let file = File::create(archive_path.as_path())?;
+    let encoder = zstd::Encoder::new(file, 0)?.auto_finish();
+    let mut builder = tar::Builder::new(encoder);
+
+    for rel_path in output_files {
+        let abs_path = workspace_root.join(rel_path);
+        // Skip files that no longer exist (task may delete temp files)
+        if !abs_path.as_path().exists() {
+            continue;
+        }
+        let metadata = std::fs::metadata(abs_path.as_path())?;
+        if metadata.is_file() {
+            let mut file = File::open(abs_path.as_path())?;
+            let mut header = tar::Header::new_gnu();
+            header.set_metadata(&metadata);
+            header.set_cksum();
+            builder.append_data(&mut header, rel_path.as_str(), &mut file)?;
+        }
+    }
+
+    builder.finish()?;
+    Ok(())
+}
+
+/// Extract a tar.zst archive, restoring files relative to workspace root.
+///
+/// Parent directories are created automatically. Existing files are overwritten.
+///
+/// # Errors
+///
+/// Returns an error if opening the archive or extracting entries fails.
+pub fn extract_output_archive(
+    workspace_root: &AbsolutePath,
+    archive_path: &AbsolutePath,
+) -> anyhow::Result<()> {
+    let file = File::open(archive_path.as_path())?;
+    let decoder = zstd::Decoder::new(file)?;
+    let mut archive = tar::Archive::new(decoder);
+
+    archive.unpack(workspace_root.as_path())?;
+    Ok(())
+}

crates/vite_task/src/session/cache/archive.rs

@@ -174,6 +174,7 @@ pub fn format_cache_status_inline(cache_status: &CacheStatus) -> Option<Str> {
                     }
                 }
                 FingerprintMismatch::InputConfig => "input configuration changed",
+                FingerprintMismatch::OutputConfig => "output configuration changed",
                 FingerprintMismatch::InputChanged { kind, path } => {
                     let desc = format_input_change_str(*kind, path.as_str());
                     return Some(vite_str::format!("○ cache miss: {desc}, executing"));

crates/vite_task/src/session/cache/display.rs

@@ -1,5 +1,6 @@
 //! Execution cache for storing and retrieving cached command results.
 
+pub mod archive;
 pub mod display;
 
 use std::{collections::BTreeMap, fmt::Display, fs::File, io::Write, sync::Arc, time::Duration};
@@ -14,7 +15,8 @@ use rusqlite::{Connection, OptionalExtension as _, config::DbConfig};
 use serde::{Deserialize, Serialize};
 use tokio::sync::Mutex;
 use vite_path::{AbsolutePath, RelativePathBuf};
-use vite_task_graph::config::ResolvedInputConfig;
+use vite_str::Str;
+use vite_task_graph::config::ResolvedGlobConfig;
 use vite_task_plan::cache_metadata::{CacheMetadata, ExecutionCacheKey, SpawnFingerprint};
 use wincode::{
     SchemaRead, SchemaReadOwned, SchemaWrite,
@@ -43,14 +45,18 @@ pub struct CacheEntryKey {
     pub spawn_fingerprint: SpawnFingerprint,
     /// Resolved input configuration that affects cache behavior.
     /// Glob patterns are workspace-root-relative.
-    pub input_config: ResolvedInputConfig,
+    pub input_config: ResolvedGlobConfig,
+    /// Resolved output configuration that affects cache restoration.
+    /// Glob patterns are workspace-root-relative.
+    pub output_config: ResolvedGlobConfig,
 }
 
 impl CacheEntryKey {
     fn from_metadata(cache_metadata: &CacheMetadata) -> Self {
         Self {
             spawn_fingerprint: cache_metadata.spawn_fingerprint.clone(),
             input_config: cache_metadata.input_config.clone(),
+            output_config: cache_metadata.output_config.clone(),
         }
     }
 }
@@ -103,6 +109,9 @@ pub struct CacheEntryValue {
     /// Path is relative to workspace root, value is `xxHash3_64` of file content.
     /// Stored in the value (not the key) so changes can be detected and reported.
     pub globbed_inputs: BTreeMap<RelativePathBuf, u64>,
+    /// Filename of the output archive (e.g. `{uuid}.tar.zst`) stored alongside
+    /// `cache.db` in the cache directory. `None` if no output files were produced.
+    pub output_archive: Option<Str>,
 }
 
 #[derive(Debug)]
@@ -142,6 +151,8 @@ pub enum FingerprintMismatch {
     },
     /// Found a previous cache entry key for the same task, but `input_config` differs.
     InputConfig,
+    /// Found a previous cache entry key for the same task, but `output_config` differs.
+    OutputConfig,
 
     InputChanged {
         kind: InputChangeKind,
@@ -158,6 +169,9 @@ impl Display for FingerprintMismatch {
             Self::InputConfig => {
                 write!(f, "input configuration changed")
             }
+            Self::OutputConfig => {
+                write!(f, "output configuration changed")
+            }
             Self::InputChanged { kind, path } => {
                 write!(f, "{}", display::format_input_change_str(*kind, path.as_str()))
             }
@@ -201,16 +215,16 @@ impl ExecutionCache {
                         "CREATE TABLE task_fingerprints (key BLOB PRIMARY KEY, value BLOB);",
                         (),
                     )?;
-                    conn.execute("PRAGMA user_version = 11", ())?;
+                    conn.execute("PRAGMA user_version = 12", ())?;
                 }
-                1..=10 => {
+                1..=11 => {
                     // old internal db version. reset
                     conn.set_db_config(DbConfig::SQLITE_DBCONFIG_RESET_DATABASE, true)?;
                     conn.execute("VACUUM", ())?;
                     conn.set_db_config(DbConfig::SQLITE_DBCONFIG_RESET_DATABASE, false)?;
                 }
-                11 => break, // current version
-                12.. => {
+                12 => break, // current version
+                13.. => {
                     return Err(anyhow::anyhow!(
                         "Unrecognized database version: {user_version}. \
                          The cache may have been created by a newer version of Vite Task. \
@@ -270,11 +284,20 @@ impl ExecutionCache {
             self.get_cache_key_by_execution_key(execution_cache_key).await?
         {
             // Destructure to ensure we handle all fields when new ones are added
-            let CacheEntryKey { spawn_fingerprint: old_spawn_fingerprint, input_config: _ } =
-                old_cache_key;
+            let CacheEntryKey {
+                spawn_fingerprint: old_spawn_fingerprint,
+                input_config: old_input_config,
+                output_config: old_output_config,
+            } = old_cache_key;
             let mismatch = if old_spawn_fingerprint == *spawn_fingerprint {
-                // spawn fingerprint is the same but input_config or glob_base changed
-                FingerprintMismatch::InputConfig
+                // spawn fingerprint is the same but input_config or output_config changed
+                if old_input_config != cache_metadata.input_config {
+                    FingerprintMismatch::InputConfig
+                } else if old_output_config != cache_metadata.output_config {
+                    FingerprintMismatch::OutputConfig
+                } else {
+                    FingerprintMismatch::InputConfig
+                }
             } else {
                 FingerprintMismatch::SpawnFingerprint {
                     old: old_spawn_fingerprint,
@@ -288,16 +311,33 @@ impl ExecutionCache {
     }
 
     /// Update cache after successful execution.
+    ///
+    /// If a previous entry exists for the same cache key with a different
+    /// `output_archive`, the stale archive file in `cache_dir` is removed
+    /// (best-effort) so it doesn't accumulate on disk.
     #[tracing::instrument(level = "debug", skip_all)]
     pub async fn update(
         &self,
         cache_metadata: &CacheMetadata,
         cache_value: CacheEntryValue,
+        cache_dir: &AbsolutePath,
     ) -> anyhow::Result<()> {
         let execution_cache_key = &cache_metadata.execution_cache_key;
 
         let cache_key = CacheEntryKey::from_metadata(cache_metadata);
 
+        // If a previous entry exists with a stale output archive, delete the
+        // old file so the cache directory doesn't accumulate orphaned archives.
+        if let Some(old_value) = self.get_by_cache_key(&cache_key).await?
+            && let Some(old_archive) = old_value.output_archive
+            && cache_value.output_archive.as_ref() != Some(&old_archive)
+        {
+            let old_archive_path = cache_dir.join(old_archive.as_str());
+            // Best-effort cleanup: a missing file (e.g. after a crash or manual
+            // cache clear) is fine, so we ignore the error.
+            let _ = std::fs::remove_file(old_archive_path.as_path());
+        }
+
         self.upsert_cache_entry(&cache_key, &cache_value).await?;
         self.upsert_task_fingerprint(execution_cache_key, &cache_key).await?;
         Ok(())

crates/vite_task/src/session/cache/mod.rs

@@ -109,6 +109,58 @@ pub fn compute_globbed_inputs(
     Ok(result)
 }
 
+/// Collect file paths matching positive globs, filtered by negative globs.
+///
+/// Like [`compute_globbed_inputs`] but only collects paths (no hashing).
+/// Used for determining which output files to archive.
+pub fn collect_glob_paths(
+    workspace_root: &AbsolutePath,
+    positive_globs: &std::collections::BTreeSet<Str>,
+    negative_globs: &std::collections::BTreeSet<Str>,
+) -> anyhow::Result<Vec<RelativePathBuf>> {
+    if positive_globs.is_empty() {
+        return Ok(Vec::new());
+    }
+
+    let negatives: Vec<Glob<'static>> = negative_globs
+        .iter()
+        .map(|p| Ok(Glob::new(p.as_str())?.into_owned()))
+        .collect::<anyhow::Result<_>>()?;
+    let negation = wax::any(negatives)?;
+
+    let mut result = Vec::new();
+
+    for pattern in positive_globs {
+        let glob = Glob::new(pattern.as_str())?.into_owned();
+        let walk = glob.walk(workspace_root.as_path());
+        for entry in walk.not(negation.clone())? {
+            let entry = match entry {
+                Ok(entry) => entry,
+                Err(err) => {
+                    let io_err: io::Error = err.into();
+                    if io_err.kind() == io::ErrorKind::NotFound {
+                        continue;
+                    }
+                    return Err(io_err.into());
+                }
+            };
+            if !entry.file_type().is_file() {
+                continue;
+            }
+            let path = entry.path();
+            let Some(stripped) = path.strip_prefix(workspace_root.as_path()).ok() else {
+                continue;
+            };
+            let relative = RelativePathBuf::new(stripped)?;
+            result.push(relative);
+        }
+    }
+
+    result.sort();
+    result.dedup();
+    Ok(result)
+}
+
 #[expect(clippy::disallowed_types, reason = "receives std::path::Path from wax glob walker")]
 fn hash_file_content(path: &std::path::Path) -> io::Result<u64> {
     super::hash::hash_content(io::BufReader::new(File::open(path)?))