Giorgi / DuckDB.NET / 39571241390

93%

Build Type

push

github

Committed by Giorgi

Commit Message

Decode a GetChars value once and count its length without decoding

GetChars decoded the value rune by rune from its start on every call, so
reading a large value in chunks took time that grows with the square of
its length. It now decodes the value once with GetString, keeps it for
later calls on the same row, and copies from it. A null buffer counts
the UTF-16 length from the UTF-8 bytes without decoding them.

Measured on the benchmark VM (net10.0, 1M characters, one read from
nothing kept): chunked reads in 4096 characters went from 113 ms to
0.2 ms (ASCII) and from 184 ms to 2.8 ms (mixed widths). A single whole
read of mixed-width text is 1.38 times slower, because GetString builds
the string the old code did not.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01F9b3CGWkCMzmhoxge1QiZ8

Coverage Stats

2313 of 2564 branches covered (90.21%)

Branch coverage included in aggregate %.

38 of 42 new or added lines in 1 file covered. (90.48%)

2 existing lines in 1 file now uncovered.

4032 of 4274 relevant lines covered (94.34%)

970826.32 hits per line