Releases: bosun-ai/swiftide
swiftide-v0.13.1
swiftide-v0.13.0
New features
BREAKING CHANGE: The batch size of batch transformers when indexing is
now configured on the batch transformer. If no batch size or default is
configured, a configurable default is used from the pipeline. The
default batch size is 256.
Bug fixes
BREAKING CHANGE: SupportedLanguages are now non-exhaustive. This means that matching on SupportedLanguages will now require a catch-all arm.
This change was made to allow for future languages to be added without breaking changes.
Miscellaneous
Qdrant and FastEmbed now have a default batch size, removing the need to set it manually. The default batch size is 50 and 256 respectively.
Full Changelog: 0.12.3...0.13.0
swiftide-v0.12.3
New features
Bug fixes
As learned from [#309](https://github.com/bosun-ai/swiftide/pull/309), test coverage for the refs defs transformer was
not great. There _are_ more tests in code_tree. Turns out, with the
latest treesitter update, javascript broke as it was the only language
not covered at all.
Miscellaneous
Full Changelog: 0.12.2...0.12.3
swiftide-v0.12.2
Other
Full Changelog: 0.12.1...0.12.2
swiftide-v0.12.1
Added
-
ec227d2 (indexing,query) Add concise info log with transformation name
-
081a248 (query) Improve query performance similar to indexing in 0.12
-
01cf579 (query) Add query_mut for reusable query pipelines
-
8029926 (query,indexing) Add duration in log output on pipeline completion
Fixed
-
d62b047 (ci) Update testcontainer images and fix tests
-
39b6ecb (core) Truncate long strings safely when printing debug logs
-
16e9c74 (openai) Reduce debug verbosity
-
6914d60 (qdrant) Reduce debug verbosity when storing nodes
-
3d13889 (query) Reduce and improve debugging verbosity
-
133cf1d (query) Remove verbose debug and skip self in instrumentation
-
a871c61 (uncategorized) Fmt
-
ce17981 (uncategorized) Clippy
Other
- 214ee8d (readme) Add link to latest release post
Full Changelog: 0.12.0...0.12.1
swiftide-v0.12.0
Added
-
e902cb7 (query) Add support for filters in SimilaritySingleEmbedding (#298)
-
f158960 (uncategorized) Major performance improvements (#291)
Fixed
-
45d8a57 (ci) Use llm-cov preview via nightly and improve test coverage (#289)
-
f95f806 (indexing) Debugging nodes should respect utf8 char boundaries
-
f8314cc (indexing) Limit logged chunk to max 100 chars (#292)
-
8595553 (uncategorized) Implement into_stream_boxed for all loaders
Other
Full Changelog: 0.11.1...0.12.0
swiftide-v0.11.1
swiftide-v0.11.0
swiftide-v0.10.0
BREAKING CHANGE: Indexing nodes now have their ID calculated using UUIDv3 via MD5 as the previous algorithm was unreliable and broke in 1.81. Added benefit that collision chance is even smaller. This means that when indexing again, nodes will have different IDs and upsert will not work. Backwards compatibility is non-trivial. If this is a huge issue, ping us on discord and we will look into it.
Added
Fixed
Other
Full Changelog: 0.9.2...0.10.0
swiftide-v0.9.2
Added
-
84e9bae (indexing) Add chunker for text with text_splitter (#270)
-
387fbf2 (query) Hybrid search for qdrant in query pipeline (#260)
Fixed
Other
Full Changelog: 0.9.1...0.9.2