build_shards() iterates all three catalog parts (base, dependency,
summary) and each lists the same packages, causing every stem to be
inserted into the FTS index 3 times. Track seen stems in a HashSet
and skip duplicates. Fixes#22.
fetch_manifest and the receiver were parsing legacy text manifests which
reference SHA1 hashes, while pkg5-imported repos store files under SHA-256.
Prefer the JSON manifest (.json) everywhere so payload lookups find the
correct files. Fixes archive creation, archive import, and local
FileBackend-to-FileBackend receive for pkg5-imported repositories.
Also completes the search REST API: v0 now returns 204 for empty results
per the pkg5 spec, and publisher-less routes (/search/0/, /search/1/) are
added for clients that omit the publisher prefix.
Implement ArchiveBackend (read) and ArchiveWriter (create) using zip with
zstd compression. Adds archive/import-archive subcommands to pkg6repo and
.p6p source support to pkg6recv.
Manifest text is now carried through the solver's ResolvedPkg and written
directly to disk during install, eliminating the redundant re-fetch from
the repository that could silently fail. save_manifest() is now mandatory
(fatal on error) since the .p5m file on disk is the authoritative record
for pkg verify and pkg fix.
Add ADRs for libips API layer (GUI sharing), OpenID Connect auth, and
SQLite catalog as query engine (including normalized installed_actions
table). Add phase plans for code hygiene, client completion, catalog
expansion, and OIDC authentication.
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
Merge both the UI handler (index, ui modules) and the legacy pkg5
search endpoint changes. Remove the obsolete JSON-based
build_search_index method since search now uses FTS5 via
sqlite_catalog::build_shards.
Multiple file actions in a manifest can reference the same payload hash.
When downloaded in parallel via rayon, multiple threads would write to
the same temp file path simultaneously, causing rename failures.
Now deduplicates payloads by digest before parallel download, then maps
results back to all file actions that reference each payload.
- Store compressed payloads under the primary (uncompressed) hash as
the main key, matching IPS protocol where clients request files by
the manifest's primary hash via /file/0/<hash>
- Create hardlink under compressed hash for internal lookups
- Detect SHA256 (64 chars) vs SHA1 (40 chars) from bare hex hash
length in Digest::from_str instead of always defaulting to SHA1
- Remove digest verification for local file copies in FileBackend
(files are compressed, hash is of uncompressed content)
- Simplify recv.rs payload fetch to use primary hash directly
instead of trying multiple digest formats against the source
Files were only stored under their compressed SHA256 hash, but pkg5
clients and the IPS protocol look up files by the manifest's primary
hash (typically SHA1/SHA256 uncompressed). Create a hardlink from the
primary hash path to the compressed hash file so both old and new
clients can find payloads via the server's /file/0/<hash> endpoint.
The previous digest fallback used Display format (source:algorithm:hash)
which caused REST servers to look up files by SHA256 compressed hash
instead of the expected SHA1 primary hash. Now tries raw hash first
(compatible with pkg5 REST servers), then formatted variants with
algorithm info (compatible with local FileBackend storage).
Multiple methods held a read or write lock on self.index while calling
fallback methods that also needed to acquire locks, causing deadlocks:
- get_obsoleted_package_metadata: read lock held during filesystem
fallback that calls update_index_entry (write lock)
- get_obsoleted_package_manifest: same pattern
- remove_obsoleted_package: write lock held while calling build_index
(also needs write lock)
- search_obsoleted_packages: read lock held during fallback
Fix: scope all lock acquisitions in blocks so locks are dropped before
calling any method that may also acquire a lock. For remove, extract a
needs_rebuild flag set inside the lock scope, then call build_index
after release.
All 95 libips tests now pass (previously 3 hung indefinitely).
- Use tempdir_in(repo_path) instead of system /tmp to keep temp files on
the same filesystem, preventing cross-device rename failures
- Replace fs::copy with fs::rename for atomic file moves in transaction
commit and signature payload storage
- Fix Digest construction to use correct SHA256 algorithm instead of
defaulting to SHA1 for bare hex hashes
- Try compressed hash (additional_identifiers) as fallback when primary
hash lookup fails during payload fetch
- Remove duplicate publisher resolution block in Transaction::commit()
- Add integration test for multi-file parallel package receive
Closes: https://codeberg.org/Toasterson/ips/issues/21
- Made `receive_one` public to allow external use.
- Improved signature payload fetching, storing, and fallback logic for `FileBackend` repositories.
- Added test case `test_receive_with_signature` to verify signature handling and repository storage.
- Extended manifest fields in transaction logic to support additional attributes (e.g., `users`, `groups`, `drivers`).
- Updated `RestBackend` to handle extended file path lookups for signature payloads.
- Added `resolve_packages` to centralize wildcard pattern processing and match the latest package versions consistently.
- Updated both `FileBackend` and `RestBackend` to support wildcard patterns via `glob_to_regex` logic.
- Made `glob_to_regex` public to facilitate regex conversion for glob patterns.
- Improved `catalog_manager` handling in `RestBackend` by explicitly loading catalog parts to ensure accurate package matching.
- Replaced redundant FMRI parsing logic with `resolve_packages` for cleaner and more maintainable code.
- Migrated `_SIGNATURE` handling in `CatalogPart`, `UpdateLog`, and `PackageUpdateEntry` to use metadata fields.
- Introduced `set_signature` and `signature` methods for cleaner access and manipulation of signature data.
- Updated deserialization logic to accommodate flattened metadata storage.
- Improved structure by centralizing signature and metadata interactions.
- Introduced specific error variants (e.g., `FileOpenError`, `FileWriteError`, `DirectoryCreateError`) in `RepositoryError` for better error diagnostics.
- Applied `.map_err()` handling for operations like file creation, copying, renaming, and directory manipulation.
- Replaced direct `HashMap` usage with `Mutex<HashMap>` for thread-safe access to catalog managers in `RestBackend`.
- Refactored `list_packages` to use `list_packages_from_catalog`, removing reliance on the search API.
- Added temporary directory management for non-local-cache scenarios in `RestBackend`.
- Implemented support for fetching, importing, and storing signature payloads during transaction creation.
- Added the `Signature` struct to represent signature-related action data.
- Updated `Manifest` to include and process `signatures` as part of its fields.
- Enabled signature file imports with proper path resolution in `pkg6repo`.
- Updated `FileBackend::open` calls to use immutable `repo` variables across multiple modules.
- Improved dependency action string construction by integrating optional properties, predicates, and facets for better clarity and flexibility.
- Eliminated LZ4-based compression and related decoding utilities for manifest handling.
- Removed unused private helper and legacy methods in `catalog.rs`.
- Standardized database path handling, replacing `catalog_db_path` and `obsoleted_db_path` with `active_db_path` and `obsolete_db_path`.
- Added `#[allow(dead_code)]` annotations for unused methods in `file_backend.rs` to reduce warnings.
- Transitioned the catalog backend from `redb` to `rusqlite` for better compatibility and concurrency.
- Updated `IpsProvider` to use SQLite for package querying, dependency resolution, and obsolescence handling.
- Removed `decode_manifest_bytes_local` and unused `manifest` cache logic.
- Simplified catalog-related functions and integrated `sqlite_catalog` module.
- Enhanced test functions and added schemas for managing SQLite databases.
- Introduced `pkg6recv` for downloading packages from IPS repositories with support for recursive dependency fetching.
- Implemented CLI commands with `clap` for source/destination setup, package selection, and publisher defaults.
- Added `ConsoleProgressReporter` for detailed receive progress visibility.
- Enhanced the `Manifest` parser to support JSON format alongside IPS text format.
- Updated `FileBackend` and `RestBackend` repositories to fetch manifests and payloads in new operations.
- Extended `Digest` utilities for payload integrity checks.
- Added tests to verify basic functionality and manifest format preservation.
- Introduced scripts and configurations for manual testing using `anyvm` with OpenIndiana and OmniOS.
- Implemented repository fetching (`fetch_repo.sh`) and server startup (`run_depotd.sh`) scripts.
- Enhanced `pkg6depotd` to support default publisher routes and trailing slashes.
- Updated integration tests to verify new publisher route behavior.
- Renamed `pkg6dev` to `pkg6` across build scripts, workflows, and documentation.
- Added support for additional binaries (`pkg6repo` and `pkg6depotd`) in release builds.
- Disabled `RUSTFLAGS` warnings-as-errors policy in workflows for improved flexibility.
- Simplified error handling in `manifest_fmri` with streamlined conditionals.
- Introduced `#[allow(clippy::result_large_err)]` to suppress clippy warnings for large error types.
- Replaced `values.get(0)` with `values.first()` for improved clarity and consistency across multiple files.
- Simplified conditional checks with `.is_empty()` and `format_args!` to enhance readability.
- Implemented `Default` trait for `VariableMode` directly within the enum.
- Updated error handling by boxing errors in `RepositoryError` and `CatalogError`.
- Replaced redundant `else` blocks with streamlined logic.
- Introduced a searchable index with structured `IndexEntry` support for packages, files, directories, and dependencies.
- Added `search` method in `DepotRepo` with wildcard and case-sensitive query handling.
- Created `/search/0` and `/search/1` routes for search API, supporting publishers and token-based queries.
- Updated `SearchIndex` handling to map tokens to detailed `IndexEntry` structures.
- Improved index building to include attributes for files, directories, and dependencies.
- Introduced `legacy_manifest_content` field in `Transaction` to allow preserving byte-identical legacy manifests.
- Updated `commit` method to save both JSON and legacy manifests, falling back to JSON if no legacy content is provided.
- Enhanced manifest copy logic to handle both JSON and legacy formats during transaction finalization.
- Improved catalog rebuild to skip `.json` files when a corresponding legacy manifest exists.
- Corrected FMRI formatting to `pkg:/name` for entries without a publisher.
- Excluded `pkg.fmri` from summary action extraction in file backend.
- Removed unused signature handling in catalog package processing.
- Reordered `PackageVersionEntry` struct fields for logical consistency.
- Removed unnecessary spaces in JSON key-value and array formatting for Python-style compatibility.
- Enhanced `format_iso8601_basic` to ensure precise microsecond-level time formatting.
- Updated `write_update_log` to return and store SHA-1 signatures for update log verification.
- Added `/file/1/{digest}` route to support file retrieval without specifying the algorithm.
- Implemented a Python-compatible JSON serializer to ensure consistent formatting for catalog artifacts.
- Replaced `HashMap` with `BTreeMap` for deterministic ordering in catalog serialization and updates.
- Updated integration tests to validate the new route functionality and ensure response correctness.
- Refactored `format_iso8601_basic` to improve timestamp formatting consistency.
- Adjusted indentation and line breaks for structs, functions, and method calls to comply with Rust formatting standards.
- Improved error message formatting and consistency across `PkgTreeError` instances.
- Restructured long function arguments and chained calls for clarity and maintainability.
- Simplified conditional statements and loops for better readability.
- No functional changes introduced.
- Introduced `BatchOptions` for configurable batched catalog rebuilds.
- Added `rebuild_catalog_batched` in `file_backend` for batched processing with deterministic package order and progress tracking.
- Updated `rebuild_catalog` to leverage the new batched implementation with default options.
- Added integration test `test_ini_only_repo_serving_catalog` to validate functionality with an INI-only repository configuration.
- Enhanced test coverage for legacy catalog handling via `pkg5.repository`.
- Introduced `write_catalog_attrs`, `write_catalog_part`, and `write_update_log` functions for streamlined and secure file writing.
- Refactored `file_backend` to use `catalog_writer` for managing catalog updates, improving readability and maintainability.
- Updated `save_catalog_attrs`, `save_catalog_part`, and `append_update` to leverage atomic writes and ensure signature computation.
- Replaced manual serialization logic with centralized writing utilities for consistency and error resilience.
- Updated dependencies for JSON handling and signature computation.
- Introduced fallback for legacy `pkg5.repository` configuration in INI format alongside the existing `pkg6.repository` JSON format.
- Enabled SHA-1 signature computation for compatibility with legacy catalog signatures.
- Added methods to save update logs in legacy format and enhance catalog compatibility.
- Updated dependencies to include `sha1` for hashing.
- Replaced string-based `FileReadError`, `FileWriteError`, and `DirectoryCreateError` with structured error types using `PathBuf` and `source`.
- Updated all affected methods in `rest_backend`, `file_backend`, and tests to use the new structured error variants.
- Improved error messages for clarity and debugging by embedding file paths and sources directly into error data.
- Removed now-redundant catalog path resolution logic in `repo.rs`.
- Updated `file_backend` with a unified method for fetching catalog file paths, addressing invalid path prefixes.
- Adjusted HTTP handler to use the updated method.
- Expanded repository structure by introducing methods for fetching legacy catalogs, catalog file paths, and repository info.
- Added new REST API v1 endpoints for catalog, manifest, file, and publisher handling.
- Implemented `publisher` handler module with `get_publisher_v0` and `get_publisher_v1` methods to retrieve publisher details in pkg5 format.
- Updated `integration_tests` to validate new endpoints and ensure compatibility with legacy and modern catalog/manifest handling.
- Removed unused dependency `walkdir` and refactored test cases for clarity and efficiency.
- Added `add_set`, `add_license`, `add_link`, and `add_depend` methods for streamlined manifest construction.
- Updated documentation with new examples showcasing the extended `ManifestBuilder` API.
- Enhanced test coverage to validate the new helper methods.
- Introduced `api.rs` to provide a stable, struct-first API surface for building, linting, resolving, and publishing IPS packages.
- Encapsulated existing functionality for better abstraction and usability in external integrations.
- Enhanced `libips` with a high-level repository and transaction interface.
- Added support for dependency generation, manifest transformations, and linting via configurable rulesets.
- Updated documentation with integration flow examples and usage guidelines.
- Introduced `depend.rs` to handle dependency generation for ELF files, scripts, Python modules, and SMF manifests.
- Implemented file classification and analysis logic with configurable bypass rules and runpath handling.
- Added utility functions to resolve file dependencies into manifest actions using a provided repository.
- Updated `Cargo.toml` with `goblin` dependency for ELF processing.
- Enhanced codebase with default runpath insertion, dynamic token expansion, and Python module import detection.
- Included `pkgdepend` documentation for dependency resolution overview.
