• GBWT is now silent by default; adjust with Verbosity::set() if necessary.
• GBWTBuilder (and related tools) will automatically increase buffer size if a sequence is too large for the buffer.
• Metadata improvements:
  • FullPathName: A standalone version of PathName that stores sample/contig names/ids as strings without requiring Metadata.
  • Metadata::findFragment(): Returns the path identifier of the haplotype fragment possibly covering the (sample, contig, haplotype, offset) represented by a path name.
  • FragmentMap: A helper structure for working with fragmented haplotypes.
• New functionality:
  • FastLocate::decompressSA() and FastLocate::decompressDA() for decompressing the part of the suffix array / document array corresponding to a node.

Minor patch release for the GBZ paper.

• Empty paths are fully supported (but still discouraged).
• Text input format for build_gbwt (mostly for testing).
• The broken CMake support has been removed.

• Supports 64-bit ARM.
• File format version 5:
  • Optional serialization using simple-sds structures.
  • Tags structure storing arbitrary key-value pairs.
  • Compatible with versions 1-4.
  • Uses Metadata version 2 (compatible with versions 0-1).
• inverseLF(): Follow the sequence backward in a bidirectional index.
• Serialization and loading use exceptions to handle failures.
• Requires the vgteam fork of SDSL.

• Uses C++14 and the vgteam fork of SDSL.
• Direct GBWT to DynamicGBWT conversion.
• Temporary files are now thread-safe.
• An option to use persistent phasing files for haplotype generation. These files persist when the associated object is deleted, but they are still deleted when the program exits.
• The fast GBWT merging algorithm now works with overlapping node id ranges as long as the non-empty records do not overlap.
• metadata_tool now prints metadata or removes it completely.

Major new functionality: FastLocate. An add-on structure for the compressed GBWT implementing the r-index locate() algorithm. Larger than the existing locate() structure but also much faster. Must be rebuilt whenever the GBWT is changed.

Other improvements:

• Metadata is ignored when merging empty GBWTs.
• Faster construction when the paths contain many different starting nodes.

Various minor improvements. The GBWT is now stable enough to reach v1.0.

• Option to force the phasing of homozygous variants (default on).
• CachedGBWT: A caching layer over GBWT for workloads that repeatedly access the same subset of nodes.
• Direct DynamicGBWT to GBWT conversion.
• Install script.

Proper metadata: Each path (or a combination of a path and its reverse complement in a bidirectional index) has a name that consists of four integer components: sample, contig, phase, and count. Sample and contig ids may further have strings as names.

• Extended metadata with path, sample, and contig names.
• Sample names and contig name in VCF parse.
• Create full metadata when building GBWT from a VCF parse using build_gbwt.
• Renamed metadata to metadata_tool.
• Remove sequences by sample / contig name in remove_seq.
• New functionality: GBWT::firstNode(), GBWT::empty(node).

Construction improvements. This version was used for the benchmarks in the full version of the paper.

• An algorithm for removing sequences from DynamicGBWT.
• Multiple parallel merge jobs in BWT-merge. If the temporary disk is fast enough, merging is roughly twice as fast as in v0.8.
• build_gbwt improvements: Accept file lists, write metadata when building from VCF parse.

Faster construction for datasets larger than 1000GP.

• Parallel merging algorithm for quickly merging multiple GBWTs over the same chromosome. It can reduce the index construction time for large datasets by a factor of 2 to 3.
• Optional metadata in the GBWT index.
• New functionality: GBWT::extract(position), GBWT::extract(position, max_length), DynamicGBWT::fullLF().

Various improvements to support building GBWT for larger datasets than 1000GP.

• Option to change the path identifier sampling interval.
• Save the temporary structures from haplotype generation and use them as input for build_gbwt.
• Decompress the endmarker of compressed GBWT for faster extract() queries in indexes with millions of paths.
• Bug fix: Initialize incoming edges correctly when loading DynamicGBWT if alphabet offset is non-zero.
• Support for Clang.

Full decompression of the endmarker made changes to the index file format unnecessary at the moment. The changes will be made before v1.0, though.