Repository navigation

Enhancing ctable with a new utf8() string type - #677

Merged

FrancescAlted merged 20 commits into

mainfrom

enhancing-ctable3

Jul 17, 2026

Commits on Jul 16, 2026
- Add blosc2.utf8(): variable-length string columns as offsets + bytes
  
  FrancescAlted
  and
  claude
  committed
  e5bbd55
- Record P3.a implementation notes in the phase-3 plan
  
  FrancescAlted
  and
  claude
  committed
  0df0c80
- Arrow interop for utf8 columns: large_string export, sentinel-null import
  
  FrancescAlted
  and
  claude
  committed
  1ac204c
- Vectorized comparisons for utf8 columns via chunked StringDType predicates
  
  FrancescAlted
  and
  claude
  committed
  ceef758
- Allow utf8 columns as group_by() keys (correctness-first, benchmark gap recorded)
  
  FrancescAlted
  and
  claude
  committed
  c8b6ba5
- Allow sorting by utf8 columns
  
  FrancescAlted
  and
  claude
  committed
  a640b32
- Post-review fixes for utf8 columns: persistence of inplace sort/compact, fused null masks, buffer-level Arrow export
  
  FrancescAlted
  and
  claude
  committed
  0cd1ca7
- Fast factorization for utf8 groupby keys: benchmark gate passes at 2.83x
  
  FrancescAlted
  and
  claude
  committed
  77bf321
- Add bench_string_kinds.py: utf8() vs string() vs vlstring() comparison
  
  FrancescAlted
  and
  claude
  committed
  564180b
- Keep NumPy 1.26 working: vlstring fallback for Arrow string imports, skip utf8 tests
  
  FrancescAlted
  and
  claude
  committed
  507b9e9
Commits on Jul 17, 2026
- utf8: byte-level scalar comparisons (U1.a + U1.b)
  FrancescAlted
  committed
  3a7d9a3
- utf8: C-level bulk StringDType constructor for full reads (U2)
  FrancescAlted
  committed
  45ec790
- Clean NotImplementedError for create_index on utf8 columns
  
  FrancescAlted
  and
  claude
  committed
  d5bad02
- Document how to choose among string/utf8/dictionary/vlstring columns
  
  FrancescAlted
  and
  claude
  committed
  1feab86
- Self-contained NpyString declarations in utf8_ext.pyx for older numpy pxds
  
  FrancescAlted
  and
  claude
  committed
  64a832f
- utf8: chunked bulk extend, ASCII join+encode fast path, wider flush batch (I1)
  
  FrancescAlted
  and
  claude
  committed
  b672dd0
- bench_string_kinds: add --ingest-only to skip the slow read/sort/to_arrow steps
  
  FrancescAlted
  and
  claude
  committed
  89d0b26
- utf8: C-level bulk UTF-8 encode kernel for ingest (I2)
  
  FrancescAlted
  and
  claude
  committed
  45b0b4c
- utf8 plan: express speedups as Nx multipliers instead of percentages
  
  FrancescAlted
  and
  claude
  committed
  22aafbf
- utf8: address Copilot review comments on PR #677
  
  FrancescAlted
  and
  claude
  committed
  28d5f8f

Navigation Menu

Repository navigation

Enhancing ctable with a new utf8() string type - #677

Enhancing ctable with a new utf8() string type#677
FrancescAlted merged 20 commits into
mainfrom
enhancing-ctable3

Commits on Jul 16, 2026

Add blosc2.utf8(): variable-length string columns as offsets + bytes

Record P3.a implementation notes in the phase-3 plan

Arrow interop for utf8 columns: large_string export, sentinel-null import

Vectorized comparisons for utf8 columns via chunked StringDType predicates

Allow utf8 columns as group_by() keys (correctness-first, benchmark gap recorded)

Allow sorting by utf8 columns

Post-review fixes for utf8 columns: persistence of inplace sort/compact, fused null masks, buffer-level Arrow export

Fast factorization for utf8 groupby keys: benchmark gate passes at 2.83x

Add bench_string_kinds.py: utf8() vs string() vs vlstring() comparison

Keep NumPy 1.26 working: vlstring fallback for Arrow string imports, skip utf8 tests

Commits on Jul 17, 2026

utf8: byte-level scalar comparisons (U1.a + U1.b)

utf8: C-level bulk StringDType constructor for full reads (U2)

Clean NotImplementedError for create_index on utf8 columns

Document how to choose among string/utf8/dictionary/vlstring columns

Self-contained NpyString declarations in utf8_ext.pyx for older numpy pxds

utf8: chunked bulk extend, ASCII join+encode fast path, wider flush batch (I1)

bench_string_kinds: add --ingest-only to skip the slow read/sort/to_arrow steps

utf8: C-level bulk UTF-8 encode kernel for ingest (I2)

utf8 plan: express speedups as Nx multipliers instead of percentages

utf8: address Copilot review comments on PR #677