Replace in-memory hashing with a disk-backed external merge sort and `PersistentCompactIntVec` to drastically reduce peak RAM. Unify both phases using a custom `PtrHash` MPHF, eliminating `GOFunction` and `boomphf`. Introduce a concrete three-step `count_partition()` pipeline with adaptive chunk sizing based on available system memory. Update dependencies to `memmap2`, `ptr_hash`, and `obicompactvec`. Additionally, document strict genomics-only memory constraints and enforce an architectural feedback workflow requiring explicit user authorization before structural changes.
Formalize the two-phase MPHF indexing architecture and update Phase 6 to use `evidence.bin` for direct kmer extraction. Simplify the evidence and unitig storage layouts to flat packed formats enabling O(1) random access. Introduce aggregation traits (`ColumnWeights`, `CountPartials`, `BitPartials`) to support additive distance metric decomposition across partitions. Narrow the documented scope from metagenomic to individual genome datasets, and replace speculative open questions with concrete implementation specifications.
Extracts MPHF construction, evidence encoding, and unitig I/O into a new `MphfLayer` module. This removes direct dependencies on `Evidence`, `PersistentBitMatrix`, and `PersistentCompactIntMatrix` from `Layer`. The `query` method is simplified to perform direct MPHF lookups, while build logic and serialization are consolidated within the new module.
Introduces ColumnWeights, CountPartials, and BitPartials traits to compute and finalize partial distance matrices. Implements these traits for PersistentBitMatrix, PersistentCompactIntMatrix, and a new LayeredStore<S> wrapper that aggregates metrics across layers via parallel reduction. Adds ndarray for numerical aggregation and updates architecture documentation to reflect the trait-driven design and pending refactoring roadmap.
Add comprehensive documentation for the `obilayeredmap` crate, `PersistentCompactIntVec`, `PersistentBitVec`, and the hierarchical k-mer index architecture, including sidebar navigation updates across all documentation pages. Refactor the Bray-Curtis distance computation in `obicompactvec` to decouple numerator and denominator calculations, replacing direct pairwise calls with explicit loops over precomputed sums. Update tests to verify column sum accuracy and align with the simplified API.
Introduces parallel `count_ones` for `BitMatrix` and parallel column-sum aggregation alongside three pairwise distance constructors (Bray-Curtis, Euclidean, Hellinger) for `IntMatrix`. These methods support partial, layer-wise data by accepting precomputed global column sums for normalization, enabling additive decomposition across partitions. Includes unit tests verifying mathematical equivalence and partition additivity.
Introduce parallel distance matrix generation using `ndarray` and `rayon` for both `BitMatrix` and `IntMatrix`. Adds full and additive-partial variants for Jaccard, Hamming, Bray-Curtis, Euclidean, and Hellinger metrics. Includes comprehensive unit tests verifying matrix symmetry, zero diagonals, and numerical correctness against pairwise calculations.
Introduces `partial_jaccard_dist` to return raw intersection and union counts, improving Jaccard distance flexibility. Corrects `not()` to explicitly zero padding bits in the final word, ensuring accurate bit-counting for partially-filled words. Adds an optimized `build_from_counts` constructor.
Introduces `PersistentBitMatrix` and `PersistentCompactIntMatrix` to replace single-file vector storage with a column-major, directory-based layout. Each column is persisted as an individual file alongside a lightweight `meta.json` for dimension tracking. Migrates `obilayeredmap` to use these multi-column structures, updating Rust APIs, query return types, and build signatures. Includes comprehensive documentation, unit and integration tests for persistence and accessors, and refactors distance calculation helpers.
Replace the hardcoded `Counts` module with a generic `LayerData` trait, parameterizing `Layer` and `LayeredMap` over arbitrary payload types. This decouples read-path access from build-path logic, enabling both set membership and count-based indexing via `PersistentCompactIntVec`. Adds the `obicompactvec` dependency, implements streaming layer construction, and expands test coverage for persistence and multi-layer resolution.
Introduces PersistentBitVec, a dense, memory-mapped bit vector optimized for bulk u64-word operations and SIMD acceleration, complete with bitwise operators and Jaccard/Hamming distance metrics. Upgrades PersistentCompactIntVec to a unified .pciv format using 64-bit indices and offsets, consolidating the binary layout and updating builder/reader lifecycles accordingly. Adds corresponding documentation, updates MkDocs navigation, and implements a comprehensive test suite for persistence round-trips, edge cases, and metric accuracy.
Adds a `sum()` method to compute the aggregate of all elements using u64 arithmetic to safely handle potential overflows. Introduces a `sumadd()` method for element-wise addition into the builder, enforcing strict length equality and using `checked_add` for safe accumulation. Includes unit tests to verify correct aggregation and overflow safety.
Refactor `PersistentCompactIntVecBuilder` to replace in-memory `Vec<u8>` with `MmapMut` for primary storage. Decouples initialization from finalization by updating `finalize_pciv` to truncate, append overflow data, and overwrite the header placeholder. Adds file path tracking via a new `path()` accessor, implements `build_from` for efficient copying, and introduces arithmetic/set operations (`min`, `max`, `sum`, `diff`). Expands test coverage for persistence roundtrips, mutations, iteration, and the new vector operations.
Add an `Iter` struct that implements `Iterator` and `ExactSizeIterator` to enable idiomatic traversal of `&PersistentCompactIntVec`. The iterator maintains `slot` and `overflow_pos` state to correctly yield `u32` values from both primary and overflow memory regions. Includes three unit tests validating iteration correctness against direct indexing, accurate `len()` tracking, and proper reference-based iteration.
Refactors the storage format to separate primary and overflow data into distinct files. Introduces a cache-friendly sparse index with dynamically computed step and entry counts. Consolidates dual memory-mapped regions into a single file with explicit header parsing and validation, replacing unsafe slice casting with direct byte-offset indexing. Updates the test suite to accommodate the new file structure.
Introduce the `obicompactvec` crate, featuring a two-tier, memory-mapped integer vector that uses a primary `u8` array with a sentinel for overflow dispatch and a sparse L1-resident index for fast random access. Implement builder and reader modules with zero-copy serialization and comprehensive test coverage. Update `obilayeredmap` to replace the default hash function with a cache-line-optimized `Mphf`, adding explicit bounds checking and duplicate-slot detection. Add documentation for both modules and update project configuration files accordingly.
Replaces per-nucleotide iteration with direct bit-slice copying via `bitvec`. This eliminates per-element decoding overhead and intermediate allocations by computing the target byte length, copying the packed bit range `[start*2, end*2)` directly into a pre-allocated buffer, and constructing the result in a single pass.
Replace `ph` with `ptr_hash` and introduce `epserde` and `rayon` dependencies. Refactor MPHF construction to leverage parallel iteration, eliminating intermediate `Vec<u64>` allocations and reducing memory footprint. Add a `n_kmers` field to track and serialize total kmer counts, alongside three zero-allocation iterators for efficient chunk traversal. Include comprehensive unit tests for the new iterators and update CLAUDE.md to enforce explicit dependency validation policies.
Introduces the `obilayeredmap` crate (v0.1.0), implementing an append-only, disk-backed k-mer index using a minimal perfect hash function (MPHF). The module features memory-mapped reads, buffered writes, custom error handling, partition metadata persistence, and comprehensive unit tests. Also adds a reverse complement benchmark for `obikseq` and updates `Cargo.lock` with the new dependencies.
Updates documentation for `PackedSeqKmerIter` and `SuperKmer` iterator methods to explicitly clarify borrowing versus consuming semantics. Documents ownership transfer requirements for `into_*` variants and explains lifetime constraints that prevent borrowing forms from being used in `flat_map` closures. Highlights optimal usage contexts, particularly zero-allocation iteration for owned values.
Merge PackedSeqKmerIter and OwnedPackedSeqKmerIter into a single generic PackedSeqKmerIter<S> parameterized over the storage type. Add an AsRef<PackedSeq> implementation to PackedSeq to enable this abstraction, allowing the zero-allocation sliding-window kmer iterator to seamlessly accept both borrowed and owned sequences without code duplication.
Introduces `into_kmers()` and `into_canonical_kmers()` consuming methods to `PackedSeq` and `Superkmer`, enabling zero-allocation sliding-window k-mer extraction via bitwise operations. This complements existing borrow-based iterators by allowing direct ownership transfer. Also includes minor documentation updates, whitespace fixes, and new unit tests to verify canonical k-mer iteration counts and output sequences.
Removes the `codec` and `limits` modules, eliminating manual serialization logic and OS file descriptor limit queries. Introduces a shared `append_path_suffix` utility to standardize path manipulation across `meta.rs` and `unitig_index.rs`. Refactors the file pool to dynamically size based on available file descriptors and optimizes descriptor lifecycle to prevent leaks. Enhances `SKFileReader` with LRU eviction, consumption tracking, and seek-on-reopen support. Migrates related round-trip and EOF tests to the reader module and updates chunking logic in the unitig index.
Replaces the `write_one` loop with threshold-based draining and inlines metadata updates. Explicitly promotes accessed entries to MRU during flush and drain operations to prevent premature LRU eviction. Updates comments to clarify two-phase locking and MRU promotion semantics.
Extend `SKError` with `BadMagic`, `Truncated`, and `InvalidData` variants to replace `expect()` calls and unsafe slice indexing. Implement proper error chaining via `source()` and update `Display` formatting. Improve magic byte validation and serde error mapping for clearer debugging. Documents a broader roadmap for further crate hardening, including LRU eviction, concurrency, and benchmarking.
Introduce the `obilayeredmap` specification and persistent MPHF-based index architecture for incremental multi-dataset indexing. Implement chunked binary serialization with a fixed `u8` k-mer count limit (256) and overlapping super-kmer segments. Add memory-mapped I/O and a companion `.idx` index file for allocation-free, O(1) unitig access. Update MkDocs navigation, enhance the k-mer comparison script, and add comprehensive tests for serialization, partitioning, and file I/O pipelines.
Centralize k-mer and minimizer configuration using a thread-safe global module, and replace manual bit-packing with a memory-efficient `PackedSeq` type. Refactor core sequence and k-mer types to use compile-time length enforcement and centralized hashing. Introduce a new De Bruijn graph implementation with compact node encoding and traversal iterators. Update I/O, partitioning, and builder modules to align with the new architecture, and add the `xxhash-rust` dependency.
Adds a `_strip_think` function using `awk` to buffer stdin and track the last `</think>` tag, emitting only the subsequent content. This utility is now piped after `aichat` calls to remove AI reasoning blocks before commit message generation. Also applies minor whitespace and indentation adjustments throughout the script.
Refactors `startIter` to separate the traversal into two distinct passes: one for chain starts (nodes without left extension) and another specifically targeting cycle nodes. Also simplifies `nextUnitigKmer` by removing the redundant `_at_start_ parameter and unifying traversal direction logic. Updates `UnitigIter` to manage visited marking internally, improving encapsulation and cycle detection reliability.
- Refactored Node representation using compact bitfields for neighbor counts
and nucleotides; added count_neighbors helper to compute_degrees()
- Introduced StartIter iterator for unitig/longtigu generation with revised
traversal semantics (e.g., interior node marking)
- Added nucleotide() accessor to Kmer type for 2-bit extraction at position i
- Renamed unitig.rs → longtigs, updated CLI command and output filenames to reflect "long t ig"
- Extended extract_kmers() in scripts/compare.py with duplication statistics
```
Refactors the `start_iter` function to separate chain/cycle detection into two passes, consolidates visited marking logic internally in `next_unitig_kmer`, and streamlines the public API by removing redundant methods, updating iteration parameters to `(start_first_next)`, and eliminating external state like `at_start`.
Refactor core types to consistently use `CanonicalKMer` (lexicographically minimal of k-mer and its reverse complement) as the canonical representation, ensuring deterministic behavior in graph traversal (unitig decomposition), neighbor resolution (`unique_neighbor` with `[CanonicalKmer; 4]` input) and scatter output generation. Introduce `RoutableSuperKmer`, add `.seq_hash()` support, fix type syntax errors in unitig extraction methods and deduplication tests. Update all k-mer construction to use canonical-aware APIs, including unsafe unchecked constructors for performance-critical paths.
Introduce `. seqhash(&self)` for direct XXH3-64 sequencing of packed bytes, and remove legacy `.hash()` method that used conditional canonicalization via revcomp. Also update partitioning logic to use `sk.hashseq_hash()` and deduplicate imports.
Replace raw SuperkMer routing with a new RoutableSuperKimer type that embeds canonical sequences and precomputed minimizers, enabling direct partition routing via hash. Update the build pipeline to yield RoutableSuperKmers throughout (builder, scatterer), refactor FASTA/unitig export commands to use the new type and compressed outputs (.fasta.gz, .unitigs.fasta.zst), revise SuperKmer header to store n_kmers instead of seql (avoiding 256-byte wrap), and update documentation to reflect minimizer-based theory, two evidence-encoding strategies for unitig-MPHF indexing (global offset vs. ID+rank), and the new obipipeline library architecture with parallel workers, biased scheduling, and error handling.
- Replace WorkerPool-based pipelines with typed `Pipe` abstraction in obipipeline
- Introduce Pipe/PipeIter for composable, sourceless/sink-less pipelines
- Update partition and superkmer commands to use new Pipe API via make_pipe!
- Remove Arc<Mutex<...>> patterns; simplify state management
- Fix macOS available_memory() returning 0 by falling back to half total memory in dereplicate()
- Remove unused `format: "zstd"` field from partition.meta
- Added `use std::marker::PhantomData;` to prepare for generic scheduler implementations
- Ensures type safety and avoids unused lifetime/type parameters warnings
- Remove redundant validation logic in login handler
- Consolidate session token generation into a single utility function
- Update error handling to use consistent HTTP status codes
- Add `hash` field to MmerItem for stable, randomized minimizer ordering
- Introduce hash_mMER() using mix64 with XOR seed to avoid fixed points (e.g., poly-A/T)
- Remove is_degenerate() and minimizer_worse(), simplifying comparison to hash-only
- Update push logic: compare hashes instead of canonical values with degeneracy checks
- Replace `limits` module and raw binary I/O with a new high-level abstraction using obiskio::SKFileWriter
- Remove `niffler` dependency and compression logic (Gzip/Zstd/Lz4/Bgzf)
- Simplify PartitionManager to manage partitioned file writers based on kmer hashing
* Uses `n_partition_bits` for bitmask-based partition selection (2^n partitions)
- Add obiskio as a local dependency
Note: This is likely part of aligning with unified I/O primitives in the obiskio crate.
- Add new obiskio crate for high-performance SuperKmer serialization/deserialization
- Implement binary codec with 2-bit packed sequence encoding and raw header format (32 bits)
- Add transparent compression support via niffler: Zstd, Gzip/Bgzf/Lz4
- Implement SKFilePool with LRU-based fd management, max-concurrent-fd limiting (75% of ulimit)
- Add SKFileWriter with batched writes, configurable flush threshold (8 KiB default), and two-phase locking
- Add SKFileReader with sequential access, LRU recovery via reopen_and_seek()
+ New obikpartitionner crate: basic header/seq handling for binary super-kmer format
- Bump niffler from 2.7 to v3, add dependencies: allocator-api2, bitflags(>=1), errno/fastrand/rustix/tempfile/lru/hashbrown/bzip2/thiserror
- Update workspace members to include obikpartitionner andobiskio
- Replace crossbeam-channel-based threading model
- Introduce obipipeline crate with Stage::Transform/Flat support
- Replace single input + format detection by multiple inputs via PathIter
- Implement pipeline stages: open_chunks → normalize → build_superkmers (flat) + write_batch
- Add SharedFlatFn for 1→N transformations with delta tracking in scheduler loop
- Added `regex` dependency to obiread crate
- Replaced manual byte checks with regex-based detection for FASTA/FASTQ formats in mimetype.rs
- Switched from `once_cell::sync::Lazy` to standard library's `std:: sync :: LazyLock`
- Added generic text/plain fallback detection for ASCII-compatible content
- Updated `MimeTypeGuesser::new` constructor call syntax and simplified API usage of PeekReader's header method
- Implemented `Read trait for MimeTypeGuesser to allow transparent passthrough reading
- Introduce `obipipline` crate with multi-threaded data pipeline architecture
- Implement core types: SourceFn, SharedFun (Arc), SinkFN with biased scheduler and crossbeam channels
- Add macros: `make_source!`, `transform!/fallible`/sink!, and high-level DSL macro
- Replace old wrapper/error modules with unified scheduler module (renamed types, improved error variants)
- Update workspace: add `obipipeline` member to Cargo.toml and lockfile
- Document pipeline in docmd/implementation with full architecture, error handling & example
- Refactor sandbox_pipeline.rs to use new DSL instead of manual channel wiring
- Introduce new `obipackage` library with pipeline stages, scheduler and worker pool
- Refactor path expansion in `obiread`: replace old list_of_files with new PathIter iterator
- Add MIME type detection using `infer` crate (fastq/fasta)
- Update dependencies in Cargo.lock: add bumpalo, byteorder, cfb (with deps), fnv,
infer, js-sys/uuid/wasm-bindgen ecosystem
- Fix formatting and improve tests in SuperKmer (canonical, revcomp)
* Note: edition = "2024" in obipipeline/Cargo.toml is invalid; should be 2021
Eric Coissac