| Name | Modified | Size | Downloads / Week |
|---|---|---|---|
| Parent folder | |||
| README.md | 2026-10-30 | 1.2 kB | |
| v1.2.1 source code.tar.gz | 2026-10-30 | 283.6 kB | |
| v1.2.1 source code.zip | 2026-10-30 | 311.1 kB | |
| Totals: 3 Items | 595.9 kB | 0 | |
CSV v1.2.1
Fixed
- Columns requested as
Float32orFloat16(throughtypes,typemap, or a scan'sselect) are now parsed at that precision. They were parsed asFloat64and then converted, which rounds twice, so some values came out one unit off:1.0000000596046448read asFloat32gave1.0f0instead of1.0000001f0. Reading them also uses less memory. (#1228) - A streamed
Tables.Scanfilter no longer throws when a later chunk changes a column's sampled type (for example fromBooltoString) and the filter can't compare the sampled type; the read parses the whole window first, as it already did for other type changes. (#1227) - The docs now explain that type inference only looks at the rows a read returns. (#1229)
(Release notes generated via AI.)
Merged pull requests:
- Fall back when a streamed Scan filter rejects a sampled column type (#1227) (@quinnj-bot[bot])
- Parse requested Float32 and Float16 columns at their own precision (#1228) (@quinnj-bot[bot])
- Document that type inference reads only the returned rows (#1229) (@quinnj-bot[bot])
- Bump version to 1.2.1 (#1230) (@quinnj-bot[bot])