Changelog

Changelog#

All notable changes to this project are documented on this page.

[v0.6.0] - 2026-10-12#

Changed#

  • Reworked the Data Step parser and unified row execution in the Rust runtime.

  • Extended string expressions and apply() support for registered Python functions.

  • Standard string functions now support Unicode and count characters instead of bytes, similar to K functions.

  • Custom format/informat catalogs now accept dictionary mappings only.

  • Added warnings for repeated BY values across MERGE inputs.

Fixed#

  • Improved missing-value handling and preservation of Arrow values, types, and metadata.

[v0.5.0] - 2026-06-04#

v0.3.x and v0.4.x are treated as internal development milestones, and this section summarizes the changes rolled up into v0.5.0.

Added#

  • Session.transpose(...) and DatasetView.transpose(...) for minimal PROC TRANSPOSE-like reshaping with by, id, var, and out.

  • Session.assign(...) and DatasetView.assign(...) for ordered Data Step-style column creation with literals, expressions, built-in functions, and case when ... then ... else ... end, executed left-to-right through Polars column expressions for supported assignment syntax.

  • Built-in put(...) and input(...) support in helper expressions, backed by a shared session format / informat registry.

  • Session.dictionary.tables, Session.dictionary.columns, Session.dictionary(name), and DatasetView.dictionary for dynamic Arrow-metadata-backed dictionary views.

  • Structured diagnostics now carry span/label/source metadata across parse and validation failures.

Changed#

  • SubmitResult.format_log() now renders structured diagnostics through the diagnostic renderer, includes error counts, and shows source excerpts when available.

  • Case-insensitive resolution is now applied more consistently across Session/DatasetView column helpers, runtime variable lookup, and WORK.-prefixed dataset references.

  • The execution pipeline now reserves a validate stage ahead of input resolution, and rejects unresolved input datasets plus DICTIONARY.* output targets before execution.

  • Unsupported %let, %put, %*, %macro ... %mend, and proc ... run|quit; blocks are now skipped through parser-based split-stage region detection so surrounding supported code can still execute without string-scanner false positives.

  • Numeric put(...) formatting now treats w as total width, defaults d to 0, supports zw.d zero-filled output, and keeps limulus’s no-space-padding rule for short values. astype(...) / cast(...) also support alias= for writing typed results into new columns.

  • Session.sql(...) now supports DROP TABLE ... and DICTIONARY table queries in addition to read queries and CREATE TABLE ... AS ....

  • Session dataset lookup now strips a WORK. prefix consistently and normalizes through a shared dataset-key helper.

  • The Rust backend now consumes parser-provided structured AST payloads for IF / DO / ARRAY and dataset references instead of maintaining a separate subset parse path.

  • Helper pipelines now preserve Arrow metadata and numeric column types more consistently across transpose and Arrow/Polars roundtrips.

  • The Python backend now has a narrower role, and data handoff is now Arrow-based.

Fixed#

  • Multi-block Session.submit(...) now prefers datasets created earlier in the same submit call when a later SET reuses the same dataset name, preventing stale rows from pre-existing session tables from being mixed into rewritten outputs.

[v0.2.0] - 2026-03-17#

Added#

  • firstobs= and obs= source dataset options

  • Session.sql(query).

  • DatasetView.astype(mapping) for dataset-scoped type conversion in chained workflows.

  • Dataset labels in Arrow schema metadata under memlabel, and column labels in per-field Arrow metadata.

  • Session.include(path) for loading and executing external Data Step scripts from the current session.

  • DatasetView.select(columns, out=...) for explicit column selection.

  • Dictionary-based Session.set_option({...}) and Session.get_option(...) as reserved session options for future use.

Changed#

  • Source-side dataset option processing order is now keep/drop → where → rename → obs/firstobs.

  • out= is now the primary user-facing output parameter for Session/DatasetView column-oriented helpers; DatasetView no longer exposes target= and Session keeps it only as a compatibility alias.

  • The Python backend implementation was split into executor_python.py, executor_py_stage.py, executor_py_stmt.py, and executor_py_data.py.

Fixed#

  • Parsing now accepts MERGE ... END=.

[v0.1.0] - 2026-03-03#

Initial release.

Contents