haz_cache/

info.rs

//! Read-only cache introspection per `AUX-017`..`AUX-021`.
//!
//! [`CacheReader::info`] walks the cache root and returns a
//! [`CacheInfoReport`] classifying every entry, summing the total
//! disk footprint, and breaking down well-formed entries by their
//! manifest's `(chapter_revision, hash_function)` prefix. The walk
//! is strictly read-only: no file or directory under the cache
//! root is created, removed, renamed, or modified.
//!
//! The walk shares its tree-shape recognition with
//! `CacheWriter::clean` but with different intent:
//!
//! - `clean` removes objectively-stale entries and orphan
//!   `.tmp-` / `.restore-` directories (under `--soft`), and
//!   age- or size-bounded entries (under `--max-age` /
//!   `--max-size`).
//! - `info` classifies the same shapes into counts without
//!   touching anything.
//!
//! Unparseable manifests and schema mismatches surface as the
//! `corrupt_entries` count rather than as errors; the only failure
//! mode is a filesystem read error during the walk.

use std::collections::BTreeMap;
use std::path::Path;

use haz_vfs::{EntryKind, Filesystem, FsError};
use snafu::{ResultExt, Snafu};

use crate::layout;
use crate::manifest::{HashFunctionLabel, Manifest};
use crate::reader::CacheReader;

/// Failure modes for [`CacheReader::info`].
///
/// Unlike [`crate::clean::CleanError`], `info` does not mutate the
/// cache, so the only error class is a filesystem read failure
/// encountered during the walk. Unparseable manifests and schema
/// mismatches are NOT errors; they fold into the
/// [`CacheInfoReport::corrupt_entries`] count.
#[derive(Debug, Snafu)]
pub enum CacheInfoError {
    /// Underlying filesystem error during the read-only walk.
    /// Wraps the [`FsError`] for the originating path.
    #[snafu(display("filesystem error during cache info walk: {source}"))]
    Io {
        /// The originating filesystem error.
        source: FsError,
    },
}

/// Schema-prefix key used by [`CacheInfoReport::by_schema`].
///
/// Equal to the `(chapter_revision, hash_function)` pair recorded
/// in each well-formed manifest per `CACHE-003` / `CACHE-011`.
pub type SchemaPrefix = (u8, HashFunctionLabel);

/// Outcome of [`CacheReader::info`]: every category `AUX-019` demands.
///
/// Counts are accumulated across the entire cache root. The total
/// byte footprint is the sum of regular-file sizes for every file
/// reached during the walk (apparent size per
/// [`haz_vfs::FsMetadata::size`]). The `by_schema` breakdown
/// counts only well-formed entries; corrupt entries are not
/// classified by manifest fields they may fail to expose.
#[derive(Debug, Default, Clone, PartialEq, Eq)]
pub struct CacheInfoReport {
    /// Number of entries whose manifest is present, parses, and
    /// declares the cache's current schema (`AUX-019` step 2).
    pub well_formed_entries: u64,
    /// Number of entries whose manifest is absent, unparseable, or
    /// declares a non-current schema (`AUX-019` step 3).
    pub corrupt_entries: u64,
    /// Number of `<shard>/.tmp-<key>-<random>` directories left by
    /// an incomplete two-phase store (`AUX-019` step 6).
    pub orphan_tmp_dirs: u64,
    /// Number of `<cache_root>/.restore-<key>-<random>` directories
    /// left by an interrupted restoration (`AUX-019` step 7).
    pub orphan_restore_dirs: u64,
    /// Total byte footprint of every regular file under the cache
    /// root (`AUX-019` step 4, apparent-size choice).
    pub total_bytes: u64,
    /// Per-schema-prefix count of well-formed entries (`AUX-019`
    /// step 5).
    pub by_schema: BTreeMap<SchemaPrefix, u64>,
}

impl<Fs: Filesystem> CacheReader<Fs> {
    /// Walk the cache root and classify every entry per `AUX-019`.
    ///
    /// Idempotent on an absent cache root: when
    /// `<workspace_root>/.haz/cache` does not exist, returns the
    /// default-zero report rather than an error.
    ///
    /// # Errors
    ///
    /// Returns [`CacheInfoError::Io`] wrapping the underlying
    /// [`FsError`] when any filesystem read fails. Manifest parse
    /// failures and schema mismatches are NOT errors: those entries
    /// fold into the [`CacheInfoReport::corrupt_entries`] count.
    pub fn info(&self) -> Result<CacheInfoReport, CacheInfoError> {
        let mut report = CacheInfoReport::default();

        let cache_entries = match self.fs().read_dir(self.cache_root()) {
            Ok(es) => es,
            Err(FsError::NotFound { .. }) => return Ok(report),
            Err(e) => return Err(CacheInfoError::Io { source: e }),
        };

        for entry in cache_entries {
            let name = entry
                .path
                .file_name()
                .map(|n| n.to_string_lossy().into_owned())
                .unwrap_or_default();

            if name.starts_with(".restore-") {
                report.orphan_restore_dirs += 1;
                report.total_bytes = report
                    .total_bytes
                    .saturating_add(self.sum_recursive(&entry.path)?);
                continue;
            }

            match entry.metadata.kind {
                EntryKind::Dir => self.walk_shard(&entry.path, &mut report)?,
                EntryKind::File => {
                    report.total_bytes = report.total_bytes.saturating_add(entry.metadata.size);
                }
                _ => {}
            }
        }

        Ok(report)
    }

    fn walk_shard(
        &self,
        shard_dir: &Path,
        report: &mut CacheInfoReport,
    ) -> Result<(), CacheInfoError> {
        let shard_entries = self.fs().read_dir(shard_dir).context(IoSnafu)?;
        for shard_entry in shard_entries {
            let sname = shard_entry
                .path
                .file_name()
                .map(|n| n.to_string_lossy().into_owned())
                .unwrap_or_default();

            if sname.starts_with(".tmp-") {
                report.orphan_tmp_dirs += 1;
                report.total_bytes = report
                    .total_bytes
                    .saturating_add(self.sum_recursive(&shard_entry.path)?);
                continue;
            }

            match shard_entry.metadata.kind {
                EntryKind::Dir => self.classify_entry(&shard_entry.path, report)?,
                EntryKind::File => {
                    report.total_bytes =
                        report.total_bytes.saturating_add(shard_entry.metadata.size);
                }
                _ => {}
            }
        }
        Ok(())
    }

    fn classify_entry(
        &self,
        entry_dir: &Path,
        report: &mut CacheInfoReport,
    ) -> Result<(), CacheInfoError> {
        report.total_bytes = report
            .total_bytes
            .saturating_add(self.sum_recursive(entry_dir)?);

        let manifest_path = entry_dir.join(layout::MANIFEST_FILE_NAME);
        let bytes = match self.fs().read(&manifest_path) {
            Ok(b) => b,
            Err(FsError::NotFound { .. } | FsError::NotAFile { .. }) => {
                report.corrupt_entries += 1;
                return Ok(());
            }
            Err(e) => return Err(CacheInfoError::Io { source: e }),
        };

        let Ok(manifest) = Manifest::from_json(&bytes) else {
            report.corrupt_entries += 1;
            return Ok(());
        };

        let chapter_ok = manifest.current_chapter_revision_matches();
        let hash_ok = HashFunctionLabel::from(self.hash_algo()) == manifest.hash_function;
        if !chapter_ok || !hash_ok {
            report.corrupt_entries += 1;
        } else {
            report.well_formed_entries += 1;
            let schema_key: SchemaPrefix = (manifest.chapter_revision, manifest.hash_function);
            *report.by_schema.entry(schema_key).or_insert(0) += 1;
        }
        Ok(())
    }

    /// Recursively sum file sizes under `path`. Non-file entries
    /// (symlinks, devices, etc.) contribute zero; directories
    /// contribute the sum of their files. The walk does NOT follow
    /// symlinks: [`Filesystem::read_dir`] reports symlink kinds
    /// directly and we skip them.
    fn sum_recursive(&self, path: &Path) -> Result<u64, CacheInfoError> {
        let entries = self.fs().read_dir(path).context(IoSnafu)?;
        let mut sum = 0u64;
        for entry in entries {
            match entry.metadata.kind {
                EntryKind::File => sum = sum.saturating_add(entry.metadata.size),
                EntryKind::Dir => sum = sum.saturating_add(self.sum_recursive(&entry.path)?),
                _ => {}
            }
        }
        Ok(sum)
    }
}

#[cfg(test)]
mod tests {
    use std::collections::BTreeMap;
    use std::path::Path;

    use haz_domain::path::CanonicalPath;
    use haz_domain::settings::cache::HashAlgo;
    use haz_vfs::{Filesystem, WritableFilesystem};
    use haz_vfs_testing::MemFilesystem;

    use crate::info::{CacheInfoReport, SchemaPrefix};
    use crate::key::CacheKey;
    use crate::key::prefix::CHAPTER_REVISION;
    use crate::layout;
    use crate::manifest::{HashFunctionLabel, Manifest, OutputBlob};
    use crate::store::{StoreInputs, StoredOutput};
    use crate::writer::CacheWriter;

    const WORKSPACE_ROOT: &str = "/ws";

    fn cp(s: &str) -> CanonicalPath {
        CanonicalPath::parse_workspace_absolute(s)
            .expect("test helper expects a valid workspace-absolute path")
    }

    fn make_cache(fs: MemFilesystem, algo: HashAlgo) -> CacheWriter<MemFilesystem> {
        CacheWriter::new(fs, Path::new(WORKSPACE_ROOT), algo)
    }

    fn key_with_first_byte(first: u8) -> CacheKey {
        let mut bytes = [0u8; 32];
        bytes[0] = first;
        CacheKey::from_bytes(bytes)
    }

    fn store_a_valid_entry(
        cache: &CacheWriter<MemFilesystem>,
        key: &CacheKey,
        rel: &str,
        bytes: &[u8],
    ) {
        let target = Path::new(WORKSPACE_ROOT).join(rel);
        let anchored = format!("/{rel}");
        cache.fs().create_dir_all(target.parent().unwrap()).unwrap();
        cache.fs().write_file(&target, bytes).unwrap();
        let outs = [StoredOutput {
            workspace_absolute_path: &anchored,
            on_disk_path: &target,
            mode: 0o644,
        }];
        cache
            .store(
                key,
                &StoreInputs {
                    outputs: &outs,
                    stdout: b"",
                    stderr: b"",
                    created_at_unix: 0,
                },
            )
            .unwrap();
    }

    fn write_manifest_to_entry(
        cache: &CacheWriter<MemFilesystem>,
        key: &CacheKey,
        manifest: &Manifest,
    ) {
        cache
            .fs()
            .create_dir_all(&layout::entry_dir(cache.cache_root(), key))
            .unwrap();
        cache
            .fs()
            .write_file(
                &layout::manifest_path(cache.cache_root(), key),
                &manifest.to_json_bytes(),
            )
            .unwrap();
    }

    fn schema_blake3_current() -> SchemaPrefix {
        (CHAPTER_REVISION, HashFunctionLabel::Blake3)
    }

    // ---- AUX-019 missing-root branch ----

    #[test]
    fn aux_019_info_on_absent_cache_root_reports_zero() {
        let mut fs = MemFilesystem::new();
        fs.add_dir("/ws").unwrap();
        let cache = make_cache(fs, HashAlgo::Blake3);
        let report = cache.reader().info().unwrap();
        assert_eq!(report, CacheInfoReport::default());
    }

    // ---- AUX-019 well-formed entry ----

    #[test]
    fn aux_019_info_counts_one_well_formed_entry() {
        let mut fs = MemFilesystem::new();
        fs.add_dir("/ws").unwrap();
        let cache = make_cache(fs, HashAlgo::Blake3);
        let key = key_with_first_byte(0xAB);
        store_a_valid_entry(&cache, &key, "proj/out", b"hello");

        let report = cache.reader().info().unwrap();
        assert_eq!(report.well_formed_entries, 1);
        assert_eq!(report.corrupt_entries, 0);
        assert_eq!(report.orphan_tmp_dirs, 0);
        assert_eq!(report.orphan_restore_dirs, 0);
        let mut expected = BTreeMap::new();
        expected.insert(schema_blake3_current(), 1);
        assert_eq!(report.by_schema, expected);
        // total_bytes covers the manifest, stdout, stderr, and the
        // single output blob. The minimum is the 5-byte "hello"
        // blob; the manifest and stream files push the sum higher.
        assert!(
            report.total_bytes >= 5,
            "expected at least 5 bytes for the `hello` blob, got {}",
            report.total_bytes,
        );
    }

    // ---- AUX-019 corrupt-entry: missing manifest ----

    #[test]
    fn aux_019_info_counts_entry_without_a_manifest_as_corrupt() {
        let mut fs = MemFilesystem::new();
        fs.add_dir("/ws").unwrap();
        let cache = make_cache(fs, HashAlgo::Blake3);
        let key = key_with_first_byte(0xAB);
        cache
            .fs()
            .create_dir_all(&layout::entry_dir(cache.cache_root(), &key))
            .unwrap();

        let report = cache.reader().info().unwrap();
        assert_eq!(report.corrupt_entries, 1);
        assert_eq!(report.well_formed_entries, 0);
        assert!(report.by_schema.is_empty());
    }

    // ---- AUX-019 corrupt-entry: unparseable manifest ----

    #[test]
    fn aux_019_info_counts_entry_with_unparseable_manifest_as_corrupt() {
        let mut fs = MemFilesystem::new();
        fs.add_dir("/ws").unwrap();
        let cache = make_cache(fs, HashAlgo::Blake3);
        let key = key_with_first_byte(0xAB);
        cache
            .fs()
            .create_dir_all(&layout::entry_dir(cache.cache_root(), &key))
            .unwrap();
        cache
            .fs()
            .write_file(
                &layout::manifest_path(cache.cache_root(), &key),
                b"this is not json",
            )
            .unwrap();

        let report = cache.reader().info().unwrap();
        assert_eq!(report.corrupt_entries, 1);
        assert_eq!(report.well_formed_entries, 0);
        assert!(report.by_schema.is_empty());
    }

    // ---- AUX-019 corrupt-entry: schema mismatch ----

    #[test]
    fn aux_019_info_counts_schema_mismatched_entry_as_corrupt() {
        let mut fs = MemFilesystem::new();
        fs.add_dir("/ws").unwrap();
        let cache = make_cache(fs, HashAlgo::Blake3);
        let key = key_with_first_byte(0xAB);
        let manifest = Manifest {
            chapter_revision: CHAPTER_REVISION,
            hash_function: HashFunctionLabel::Sha256, // mismatch vs cache's Blake3
            key,
            outputs: vec![],
            stdout_len: 0,
            stderr_len: 0,
            stdout_hash: [0u8; 32],
            stderr_hash: [0u8; 32],
            exit_status: 0,
            created_at_unix: 0,
        };
        write_manifest_to_entry(&cache, &key, &manifest);

        let report = cache.reader().info().unwrap();
        assert_eq!(report.corrupt_entries, 1);
        assert_eq!(report.well_formed_entries, 0);
        assert!(report.by_schema.is_empty());
    }

    // ---- AUX-019 orphan tmp / restore dirs ----

    #[test]
    fn aux_019_info_counts_orphan_tmp_directory() {
        let mut fs = MemFilesystem::new();
        fs.add_dir("/ws").unwrap();
        let cache = make_cache(fs, HashAlgo::Blake3);
        let key = key_with_first_byte(0xAB);
        let tmp = layout::tmp_entry_dir(cache.cache_root(), &key, "abcdef");
        cache.fs().create_dir_all(&tmp).unwrap();
        cache
            .fs()
            .write_file(&tmp.join("partial.bin"), &[0u8; 17])
            .unwrap();

        let report = cache.reader().info().unwrap();
        assert_eq!(report.orphan_tmp_dirs, 1);
        assert_eq!(report.well_formed_entries, 0);
        assert_eq!(report.corrupt_entries, 0);
        assert!(report.total_bytes >= 17);
    }

    #[test]
    fn aux_019_info_counts_orphan_restore_directory() {
        let mut fs = MemFilesystem::new();
        fs.add_dir("/ws").unwrap();
        let cache = make_cache(fs, HashAlgo::Blake3);
        let key = key_with_first_byte(0xAB);
        let staging = layout::restore_staging_dir(cache.cache_root(), &key, "feedface");
        cache.fs().create_dir_all(&staging).unwrap();
        cache
            .fs()
            .write_file(&staging.join("leftover.bin"), &[0u8; 9])
            .unwrap();

        let report = cache.reader().info().unwrap();
        assert_eq!(report.orphan_restore_dirs, 1);
        assert_eq!(report.orphan_tmp_dirs, 0);
        assert!(report.total_bytes >= 9);
    }

    // ---- AUX-019 schema-prefix breakdown ----

    #[test]
    fn aux_019_info_breaks_down_by_schema_prefix() {
        let mut fs = MemFilesystem::new();
        fs.add_dir("/ws").unwrap();
        let cache = make_cache(fs, HashAlgo::Blake3);

        // Two well-formed entries under the current Blake3 schema.
        let key_a = key_with_first_byte(0xAA);
        store_a_valid_entry(&cache, &key_a, "proj/out_a", b"x");
        let key_b = key_with_first_byte(0xBB);
        store_a_valid_entry(&cache, &key_b, "proj/out_b", b"y");

        let report = cache.reader().info().unwrap();
        assert_eq!(report.well_formed_entries, 2);
        let mut expected = BTreeMap::new();
        expected.insert(schema_blake3_current(), 2);
        assert_eq!(report.by_schema, expected);
    }

    // ---- AUX-019 mixed state ----

    #[test]
    fn aux_019_info_classifies_mixed_state_correctly() {
        // Mirror clean_soft's mixed-state fixture: one valid entry,
        // one schema-mismatched entry, one orphan tmp dir, one
        // orphan restore dir.
        let mut fs = MemFilesystem::new();
        fs.add_dir("/ws").unwrap();
        let cache = make_cache(fs, HashAlgo::Blake3);

        let key_good = key_with_first_byte(0xAB);
        store_a_valid_entry(&cache, &key_good, "proj/out", b"x");

        let key_stale = key_with_first_byte(0xCD);
        let stale_manifest = Manifest {
            chapter_revision: CHAPTER_REVISION,
            hash_function: HashFunctionLabel::Sha256,
            key: key_stale,
            outputs: vec![],
            stdout_len: 0,
            stderr_len: 0,
            stdout_hash: [0u8; 32],
            stderr_hash: [0u8; 32],
            exit_status: 0,
            created_at_unix: 0,
        };
        write_manifest_to_entry(&cache, &key_stale, &stale_manifest);

        let key_tmp = key_with_first_byte(0xEF);
        let tmp = layout::tmp_entry_dir(cache.cache_root(), &key_tmp, "rnd1");
        cache.fs().create_dir_all(&tmp).unwrap();

        let key_restore = key_with_first_byte(0x12);
        let staging = layout::restore_staging_dir(cache.cache_root(), &key_restore, "rnd2");
        cache.fs().create_dir_all(&staging).unwrap();

        let report = cache.reader().info().unwrap();
        assert_eq!(report.well_formed_entries, 1);
        assert_eq!(report.corrupt_entries, 1);
        assert_eq!(report.orphan_tmp_dirs, 1);
        assert_eq!(report.orphan_restore_dirs, 1);
        let mut expected = BTreeMap::new();
        expected.insert(schema_blake3_current(), 1);
        assert_eq!(report.by_schema, expected);
    }

    // ---- AUX-019 total-bytes accuracy ----

    #[test]
    fn aux_019_info_total_bytes_sums_blob_sizes() {
        // A well-formed entry with two output blobs whose total
        // declared size is known. The blob bytes alone are 5 + 7 =
        // 12; the manifest and stream files push the apparent-size
        // total higher, but it MUST be at least 12.
        let mut fs = MemFilesystem::new();
        fs.add_dir("/ws").unwrap();
        let cache = make_cache(fs, HashAlgo::Blake3);
        let key = key_with_first_byte(0xAB);

        let small = Path::new(WORKSPACE_ROOT).join("proj/small.txt");
        let larger = Path::new(WORKSPACE_ROOT).join("proj/larger.txt");
        cache.fs().create_dir_all(small.parent().unwrap()).unwrap();
        cache.fs().write_file(&small, b"hello").unwrap();
        cache.fs().write_file(&larger, b"helloXX").unwrap();
        let outs = [
            StoredOutput {
                workspace_absolute_path: "/proj/small.txt",
                on_disk_path: &small,
                mode: 0o644,
            },
            StoredOutput {
                workspace_absolute_path: "/proj/larger.txt",
                on_disk_path: &larger,
                mode: 0o644,
            },
        ];
        cache
            .store(
                &key,
                &StoreInputs {
                    outputs: &outs,
                    stdout: b"",
                    stderr: b"",
                    created_at_unix: 0,
                },
            )
            .unwrap();

        let report = cache.reader().info().unwrap();
        assert!(
            report.total_bytes >= 12,
            "expected at least 12 bytes for the two blobs; got {}",
            report.total_bytes,
        );
        assert_eq!(report.well_formed_entries, 1);
    }

    // ---- AUX-018 read-only invariant ----

    #[test]
    fn aux_018_info_does_not_mutate_the_cache_root() {
        // The walk must not create, remove, rename, or modify any
        // file under the cache root. We snapshot every (path,
        // bytes) under the cache root before and after `info` and
        // require them to be byte-identical.
        let mut fs = MemFilesystem::new();
        fs.add_dir("/ws").unwrap();
        let cache = make_cache(fs, HashAlgo::Blake3);

        // Populate a mix of shapes.
        let key_good = key_with_first_byte(0xAB);
        store_a_valid_entry(&cache, &key_good, "proj/out", b"x");
        let key_stale = key_with_first_byte(0xCD);
        let stale_manifest = Manifest {
            chapter_revision: CHAPTER_REVISION,
            hash_function: HashFunctionLabel::Sha256,
            key: key_stale,
            outputs: vec![OutputBlob {
                workspace_absolute_path: cp("/proj/missing"),
                content_hash: [0u8; 32],
                size: 0,
                mode: 0o644,
            }],
            stdout_len: 0,
            stderr_len: 0,
            stdout_hash: [0u8; 32],
            stderr_hash: [0u8; 32],
            exit_status: 0,
            created_at_unix: 0,
        };
        write_manifest_to_entry(&cache, &key_stale, &stale_manifest);
        let tmp = layout::tmp_entry_dir(cache.cache_root(), &key_with_first_byte(0xEF), "r1");
        cache.fs().create_dir_all(&tmp).unwrap();
        cache.fs().write_file(&tmp.join("x"), b"y").unwrap();
        let staging =
            layout::restore_staging_dir(cache.cache_root(), &key_with_first_byte(0x12), "r2");
        cache.fs().create_dir_all(&staging).unwrap();

        let before = snapshot_cache(&cache);
        cache.reader().info().unwrap();
        let after = snapshot_cache(&cache);
        assert_eq!(
            before, after,
            "cache root state must not change under info()",
        );
    }

    fn snapshot_cache(cache: &CacheWriter<MemFilesystem>) -> BTreeMap<String, Vec<u8>> {
        let mut out = BTreeMap::new();
        snapshot_into(cache, cache.cache_root(), &mut out);
        out
    }

    fn snapshot_into(
        cache: &CacheWriter<MemFilesystem>,
        path: &Path,
        out: &mut BTreeMap<String, Vec<u8>>,
    ) {
        let Ok(entries) = cache.fs().read_dir(path) else {
            return;
        };
        for entry in entries {
            match entry.metadata.kind {
                haz_vfs::EntryKind::File => {
                    let key = entry.path.to_string_lossy().into_owned();
                    let bytes = cache.fs().read(&entry.path).unwrap_or_default();
                    out.insert(key, bytes);
                }
                haz_vfs::EntryKind::Dir => snapshot_into(cache, &entry.path, out),
                _ => {}
            }
        }
    }
}