substrait-io/duckdb-substrait-extension v1.5.6 on GitHub

Highlights

DuckDB 1.5.6 and Substrait v0.102.0.
Subquery support in the consumer: uncorrelated subquery expressions can now be read from Substrait plans. Correlated subqueries will follow in a later release.
Emit mappings are honored on every relation, not only ProjectRel. Before this, a mapping such as [2, 0] on another relation type could quietly return the wrong columns.
Fewer crashes on malformed or unusual plans. Several plans that used to kill the process with SIGSEGV now fail with a clear error instead. These include an unset Rel or Expression, unrecognized enum values, and enum arguments on aggregates.

NOT LIKE is now emitted as not(like(x, pattern)). It used to be emitted as not_equal, and not_equal is now always read as inequality. Plans written by earlier versions of this extension encoded NOT LIKE as not_equal, so they will now be read as !=. Those plans did not follow the spec, and the old workaround also broke real comparisons: WHERE s != 'a%c' came back as NOT LIKE. (#286)
When an aggregate has more than one grouping set, its grouping set index column is no longer included in the output, so the root names match the output columns. (#291)
Stricter emit validation. These are now rejected: mappings that are explicitly empty or whose indices are out of range, an emit on a WriteRel, and an emit on a lateral right-side FilterRel that is folded into the join condition. (#274)
Delete plans must filter the table they delete from. A plan that filters one table and names a different table as the delete target is now rejected, because it used to delete rows the plan never selected. (#274)
local_files reads bind columns to baseSchema by name, including nested fields and Hive partition columns. A mismatch with baseSchema now raises an error. (#298)

Fix local files base schema by name by @mbrobbel in #298
fix: support aggregates whose variant is selected by an enum argument (std_dev, variance, median) by @andrew-coleman in #294
fix: honor emit mappings beyond ProjectRel by @alexandrefimov in #274
fix: do not crash on a plan with an unset Rel or Expression oneof by @andrew-coleman in #296
fix: empty-table optimizer folding in get_substrait by @thgschwind in #269
fix: name an unsupported enum value instead of crashing on it by @alexandrefimov in #295
fix: emit NOT LIKE as not(like(..)) instead of not_equal by @andrew-coleman in #286
fix: preserve timestamp literal precision by @alexandrefimov in #275
fix: leave the grouping set index out of an aggregate's output by @alexandrefimov in #291
fix: dialect has missing variants for arithmetic aggregates by @andrew-coleman in #284
fix: correct two wrong DuckDB mappings in duckdb_dialect.yaml (bitwise_xor, PRECISION_TIMESTAMP_TZ) by @andrew-coleman in #283
fix: emit integer literals for TINYINT and unsigned integer types by @thgschwind in #280
fix: map substrait extract(UNIX_TIME) to date_part('epoch') by @thgschwind in #271
fix: emit TIMESTAMPTZ constants as precision_timestamp_tz literals by @thgschwind in #273

build: update duckdb to 1.5.6 by @andrew-coleman in #304
build: update Substrait to v0.102.0 (includes window bound validation and typed RANGE offset fixes) by @bvolpato in #265

test: document upcast-width intent and correct TIMETZ skip comment by @thgschwind in #281
chore: keep CodeRabbit's summary out of the PR description by @nielspardon in #299

Full Changelog: v1.5.5.1...v1.5.6