2.0.0

While this is a new major release with breaking changes in the internal Rust and external C-API, graphANNIS 2.x can read all existing data directories. No data migration is needed, and you should be able to just run the updated binary file. Also, there are no changes in how AQL is interpreted.

Changed

• Refactored the basic GraphStorage and AnnotationStorage APIs to handle
  errors. Previously, we used mostly main memory containers which had an API
  that could not fail. This was reflected in the GraphStorage and
  AnnotationStorage APIs, which directly returned the result or an iterator
  over the non-fallible results. With the addition of disk-based
  implementations, this API model was not possible to implement without using
  panics when repeated access to the disk failed. Some of the API that was
  changed was user visible when using the graphannis-core crate (and thus the
  C-API), so this release is not technically backwards-compatible. Adapting to
  the updated API should be restricted to handle the errors returned by the
  functions.
• The changes to the error handling also affects the C-API. These following
  functions have now a ErrorList argument:
  • annis_cs_list_node_annotations
  • annis_cs_list_edge_annotations
  • annis_cs_list_components_by_type
  • annis_cs_unload
  • annis_iter_nodeid_next
  • annis_graph_annotations_for_node
  • annis_graph_outgoing_edges
  • annis_graph_annotations_for_edge
• Renamed the Criterion-based benchmark CLI to bench_queries and synchronize
  its arguments to the current version of Criterion.

Fixed

• More efficient node path extraction in count_extra function and when sorting
  the matches.
• Avoid large memory consumption when importing GraphML files by resetting an
  internal buffer on each XML event.
• Limit the number of disk maps for the GraphUpdate so there are less issues
  with large corpora where the maximum number of open files per process might be
  reached.
• Performance improvements when importing large corpora in disk-based mode. This
  optimizes the DiskMap to use a C0 (normal in memory BTree), a C1 (on disk
  BTree) and a C2 map when serialized to disk. On compacting, the entries are
  only written to C1 in O(n*log(n)). Before, multiple on disk maps might need to
  be merged, which had a much worse complexity. The C1 file uses the
  transient-btree-index crate.
• Trim mapping entries when importing relANNIS resolver files (#222).
• Fixed schema errors in the Webservice OpenAPI file.