Skip to content

Releases: rapidfuzz/rapidfuzz-cpp

Release 3.1.1

24 Oct 14:08
Compare
Choose a tag to compare

Fixed

  • Fixed incorrect score calculation for SIMD implementations of Levenshtein and OSA on 32 bit systems

Release 3.1.0

24 Oct 10:21
76befcc
Compare
Choose a tag to compare

Changed

  • split editops_apply/opcodes_apply into *_apply_str and *_apply_vec. This avoids the instantiation of
    std::basic_string for unsupported types.

Release 3.0.5

02 Jul 20:04
Compare
Choose a tag to compare

Fixed

  • the editops implementation didn't properly account for some cells in the Levenshtein matrix.
    This could lead both to incorrect results and crashes.

Release 3.0.4

07 Apr 20:06
10426d2
Compare
Choose a tag to compare

Fixed

  • fix tagged version

Release 3.0.3

06 Apr 13:41
Compare
Choose a tag to compare

Fixed

  • fix potentially incorrect results of JaroWinkler when using high prefix weights

Release 3.0.2

04 Mar 11:55
cb4b1c7
Compare
Choose a tag to compare

Fixed

  • fix assert leading to compilation failures

Release 3.0.1

03 Mar 15:11
03e2dfe
Compare
Choose a tag to compare

Fixed

  • fix doxygen warnings

Release 3.0.0

26 Dec 09:05
Compare
Choose a tag to compare

Performance

  • add banded implementation of LCS / Indel. This improves the runtime from O((|s1|/64) * |s2|) to O((score_cutoff/64) * |s2|)

Changed

  • changed many types in the interface from int64_t to size_t, since they can't be negative.

Fixed

  • fix incorrect transposition calculation in simd implementation of Jaro similarity
  • use posix_memalign on android

Release 2.2.3

02 Nov 11:12
Compare
Choose a tag to compare

Fixed

  • use _mm_malloc/_mm_free on macOS if aligned_alloc is unsupported

Release 2.2.2

01 Nov 08:32
Compare
Choose a tag to compare

Fixed

  • fix compilation failure on macOS