Post 4.10.0 release actions done

Author: FrancescAlted

ANNOUNCE.rst

@@ -1,35 +1,44 @@
-Announcing Python-Blosc2 4.9.1
+Announcing Python-Blosc2 4.10.0
 ==============================
 
-This is a hot-fix release for the Arrow interop work introduced in 4.9.0 —
-one real performance regression and one clearer error message, both in
-``CTable``.
-
-- **Faster dictionary-column Arrow export**: ``CTable.iter_arrow_batches()``
-  (and therefore ``to_arrow()`` and the Arrow PyCapsule interchange,
-  ``__arrow_c_stream__``) was recomputing the full live-row-position array
-  from scratch on every batch, for every dictionary-encoded string column —
-  an ``O(n_rows)`` scan repeated ``O(n_rows / batch_size)`` times. It's now
-  computed once per export call instead. 6-14x faster export for
-  dictionary columns on a 1M-row benchmark.
-
-- **Worth knowing regardless of this fix**: the Arrow PyCapsule protocol
-  (``__arrow_c_stream__``) has no column-projection pushdown — a consumer
-  that only needs two columns (DuckDB, pyarrow, Polars, pandas) still
-  triggers export of *every* column in the table, since the raw Arrow C
-  Stream interface has no way to say "I only need these." Use
-  ``CTable.select([...])`` to project down to the columns you actually
-  need before handing the table off, especially if any column holds an
-  expensive nested/list type::
-
-      sub = t.select(["company", "fare"])
-      duckdb.sql("SELECT company, avg(sub.fare) FROM sub GROUP BY company").show()
-
-- **Clearer error on ``mode="a"``**: opening a ``CTable`` with
-  ``mode="a"`` at a path that doesn't exist yet now raises a
-  ``FileNotFoundError`` explaining that ``mode="a"`` opens an existing
-  table (use ``mode="w"`` to create one), instead of silently creating a
-  new, empty table.
+This is the string-support release: string expressions and DSL kernels now
+run on miniexpr, ``utf8()`` and ``dictionary()`` columns gain full indexing
+and comparisons, and NumPy's ``StringDType`` is understood by the array
+constructors. Alongside, slicing with plain keys is up to 1.7x faster and a
+new ``blosc2.random`` module brings chunk-parallel, NumPy-quality random
+constructors.
+
+- **String expressions and DSL kernels over strings.** Concatenation,
+  ``lower``/``upper``/``strip``/``replace``/``substr``/``split_part`` and
+  friends now run on miniexpr over fixed-width ``<Un`` and bytes ``S``
+  arrays, producing string results sized by miniexpr itself. ``utf8()``
+  columns can be queried in expression form
+  (``t.where("name == 'x'")``), with scalar comparisons 5-6x faster via a
+  raw-byte scan, and new ``blosc2.utf8_array()`` builds variable-length
+  arrays directly.
+
+- **Full indexing for string columns.** ``create_index()`` now works on
+  ``utf8()`` and ``dictionary()`` columns via alphabetical ranks —
+  ``sort_by`` drops from 424 ms to 7 ms at 1M rows — and scalar comparisons
+  are served from the index.
+
+- **New ``blosc2.random`` module**: 42 of NumPy's 43 ``Generator`` methods,
+  each chunk generated in parallel with its own seeded ``PCG64`` stream
+  (~3x faster than the NumPy path on 100M elements).
+
+- **Slicing up to 1.7x faster.** Plain slice/int keys skip ndindex's general
+  machinery (it was 43% of a scattered-read loop); strided steps, ellipsis
+  and fancy indexing still use it.
+
+- **String plumbing**: ``from_utf8()``/``to_utf8()`` conversions,
+  ``CTable.add_column(values=)``, ``Column.assign()`` on variable-length
+  columns, ``StringDType`` dispatch in the array constructors, and DSL
+  operands can be native NumPy arrays or pandas ``Series``.
+
+- **Important fixes**: string column indexes returning zero rows at the
+  default column width, ``SChunk`` slices for typesizes above 255 bytes
+  (upstream, via C-Blosc2 3.3.1), scalar bools in tuple keys now matching
+  NumPy, and a batch of miniexpr correctness fixes.
 
 Install it with::
 

RELEASE_NOTES.md

@@ -1,5 +1,9 @@
 # Release notes
 
+## Changes from 4.10.0 to 4.10.1
+
+XXX version-specific blurb XXX
+
 ## Changes from 4.9.1 to 4.10.0
 
 This release is the string-support milestone: string expressions and DSL

pyproject.toml

@@ -41,7 +41,7 @@ dependencies = [
     "rich",
     "threadpoolctl; platform_machine != 'wasm32'",
 ]
-version = "4.10.0"
+version = "4.10.1.dev0"
 [project.entry-points."array_api"]
 blosc2 = "blosc2"
 

src/blosc2/version.py

@@ -1,2 +1,2 @@
-__version__ = "4.10.0"
+__version__ = "4.10.1.dev0"
 __array_api_version__ = "2024.12"