v0.5.0
Contains many breaking changes:
- Boolean query syntax is completely ditched to follow the more conventional
+ / -
syntax. A~
operator is also supported, which translates toAND NOT word
. - Chinese tokenizer is completely rewritten without Jieba, and now uses single chinese character tokenization, relying on proximity ranking to produce relevant results. A good amount of tests are also added for it.
- Search UI design is completely redone
- Many css variables are removed/added.
- Search UI assets are now bundled into the CLI tool, allowing for semantic versioning installation of both packages.
- Some indexer options (
loader_config
,field_store_block_size
) are renamed to shorter, simpler ones - Many API changes for "Altering HTML outputs" section
Other changes:
- Stop words are now never removed at query time if present, making proximity ranking produce even more relevant results
- Refactor some internal query processing to reduce binary size by ~10%.
- Use nightly rust to build the wasm bundle with the
panic_immediate_abort
std feature, reducing binary size by ~15%. - Increase spelling correction thresholds slightly
- Reduce mobile breakpoint for using fullscreen version to
768px
- Use unsafe unchecked functions where there are strong guarantees to further reduce panic code and improve query performance
- MdBook plugin no longer requires installing the indexer package as well, it is now bundled.
- MdBook plugin no longer generates any configuration file, it is passed to the indexer directly.
- Add body scroll locking for the fullscreen UI, with an option to disable it.
- Fix/add some more default values for field configuration to minimize configuration
- Disable Gzip after more benchmarking
- Fix CLI tool indexing an empty directory
- Dependency version bumps
- Some documentation reorganizations