0.25.1

Fixed

• Inverted sort order did not reverse the corpus name list for multiple corpora
• Workaround for docs.rs problems seem to have caused other problems and graphANNIS was not recognized as library

0.25.0

Changed

• Backward incompatible: the several search functions (find, count, etc.) not take several corpus names as argument.
  This is especially important for find, where the implementation can be optimized to correctly skip over a given offset
  using the internal state.
  Such an optimization is impossible from outside when calling the API and not having access to the iterator.

Fixed

• Don't assume inverse operator has the same cost when fan-out is too different.
  Subgraph queries could be very slow for corpora with large documents due to an estimation error from this assumption
  the @ operator.

0.24.0

Changed

• The annotation storage is now a complete interface which provides all functions necessary to write and read annotations.
  To make this less dependent on the current implementation of the
  in-memory annotation storage, the annotation key symbol (an integer) has been removed.
  This annotation key symbol has been used in the Match class as well, which is now using an Arc<AnnoKey> instead. The AnnoKey contains
  the fully qualified name as String.
  Several functions of the annotation storage that used to have String parameters now take &str and resulting string values are now returned as Cow<str>. The latter change is also meant to enable more flexible implementations, that can choose to allocate new strings (e.g. from disk) or return references to existing memory locations.
• The Graph uses a boxed instance of the general AnnotationStorage trait.
  Before, this was an Arc to the specific implementation, which made it possible to simply clone the node annotation storage.
  Now, references to it must be used, e.g. in the operators. This changes a lot of things in the BinaryOperator trait, like
  the signature of get_inverse_operator() and the filter functions that are used as conditions for the node search (these
  need an argument to the node annotation storage now)
• Graph does not implement the AnnotationStorage<NodeID> trait anymore,
  but provides a getter to reference its field.
• Data source nodes are now included when querying for a subgraph with context. This is needed for parallel text support in ANNIS 4.

0.23.1

Fixed

• Deploying release artifacts by CI was broken due to invalid condition

0.23.0

Added

• Subgraph queries can now define the context using ordering relation names (segmentation)
  instead of the default context in tokens. This changes the function signature of the subgraph(...) function.

Changed

• For performance and stylistic reasons, the GraphStorage API has been changed to accept integer node IDs instead of references to integers.
• Windows DLL in releases is now created by Travis CI instead of Appveyor

0.22.0

Fixed

• Windows DLL generated by CI was empty

Changed

• Updated several dependencies
• Organize documentation topics in sub-folders.
  Previously, mdbook did not updated the images on these sites on the print.html.
  Since mdbook >0.3.1 this is fixed and we can use the better layout.

0.21.0

Changed

• C API now has an argument to return error messages when creating a corpus storage

Added

• C API now also allows to unload a corpus from the cache manually

Fixed

• CorpusStorageManager: Escape the corpus name when writing it to its disk location to support e.g. corpora with slash
  in their name.
• Quirks mode: sort matches by reversed document path (document first)
• Node names/paths where double encoded both when importing them and when executing the "find" function
• Quirks mode: use default collation of Rust for corpora imported from relANNIS 3.3

0.20.0

Deprecated

• meta:: queries are now deprecated and can only be used in quirks mode

Fixed

• Output annotations with the namespace "annis" in find function
• Quirks mode: add additional identity joins in the order as the nodes are defined in the query
• Encode ",", " " and ":" in the Salt ID output of the find(...) function
• Sort longer vectors ("more specific") before shorter ones in find(...) output

0.19.4

Changed

• Optimize parallel nested loop join by performing less copy operations

Fixed

• Quirks mode: meta-data nodes are not part of the match result anymore

0.19.2

Fixed

• Escape corpus and document paths with percent encoding when importing them from relANNIS
• Use locale aware sorting of the results in quirks mode (which depends on the system graphANNIS is executed on)
• CLI did not allow to turn quirks mode off once activated