v0.3.14

boring-semantic-layer v0.3.14

Highlights

This release fixes joins on plain ibis backends (Snowflake, Databricks, BigQuery), unifies calc-measure compilation on an ibis-native classifier (ADR 0001), and includes several correctness fixes around joins and serialization.

• Fixed ValueError: Don't know how to handle type <class 'ibis.expr.types.relations.Table'> when joining models on Snowflake / Databricks / BigQuery (#222, fixes #221)
• Unified calc measures on the new ibis-native classifier (#262, ADR 0001 phases 1+2)
• Fixed grain mismatch when chaining with_measures over joined models (#261)
• Fixed serialization content hashing to include BSL metadata (#264)

Why upgrade

• If you're on Snowflake, Databricks, BigQuery, or any other backend xorq doesn't wrap, joined-model queries now execute natively instead of crashing
• The new calc-measure compiler is faster, surfaces clearer errors on type-inference failures, and correctly preserves user casts
• Catalog workflows relying on tag content hashes will now invalidate correctly when BSL metadata changes

Included changes

Fixes

• #222 — handle plain ibis backends in join backend rebinding (fixes #221)
• #261 — preserve preagg + canonical backend through with_measures on joins
• #264 — use HashingTag so BSL metadata contributes to content hash

Features / refactors

• #262 — unify calc measures on ibis-native classifier (ADR 0001 phases 1+2)
  • New calc_analyzer and calc_compiler modules
  • Inline-reduction lift and runtime orchestration
  • Windowed totals carried through aggregation (replaces cross-join totals)
  • Tighter bare-ref detection, surfaced lift failures, asserted totals substitutions

Upgrade

pip install --upgrade boring-semantic-layer==0.3.14

v0.3.13

boring-semantic-layer v0.3.13

Highlights

This release adds metadata support on model fields, expands xorq/TagHandler integration, improves MCP tooling, and fixes several round-trip and serialization correctness issues.

• Added metadata field on Dimension and Measure for arbitrary annotations (#245, #246)
• Exposed BSL TagHandler via xorq.from_tag_node entry point for catalog workflows (#235)
• Added reemit support on TagHandler for full catalog rebuild (#244)
• Added MCP period comparison workflow (#238)
• Fixed join rebind guard for non-xorq backends (#242, #243)
• Fixed multiple round-trip and serialization correctness issues in xorq boundary and calc-measure handling

Why upgrade

• If you use metadata on dimensions or measures, this is the release that adds it
• If you're integrating BSL with xorq catalog workflows, the new TagHandler entry points make that cleaner
• Several xorq serialization round-trip bugs are fixed — recommended if you serialize/deserialize semantic models

Included changes

Features

• #238 — MCP period comparison workflow
• #244 — TagHandler reemit for catalog rebuild
• #245/#246 — metadata field on Dimension and Measure
• #235 — expose BSL TagHandler via xorq.from_tag_node

Fixes

• #242/#243 — guard join rebind against non-xorq backends
• Fix: walk join branches in tag_handler.extract_metadata
• Fix: roundtrip model-level description through to_tagged/from_tagged
• Fix: preserve join cardinality on round-trip; fix pre-agg grain for derived dims
• Fix: treat UnboundTable as leaf node in _reconstruct_table
• Fix: use matching ibis module for desc() in MCP search_dimension_values
• Fix: broaden agg-method recognition, surface typos, pin ibis workaround

Refactors

• Introduce Predicate AST for filter expressions
• Tighten calc-measure AST and surface type-inference failures
• Route all xorq imports through single _xorq shim
• Convert ibis→xorq once at SemanticModel construction
• Dispatch reduction-type check on expr's ibis flavor

Install

pip install -U boring-semantic-layer==0.3.13

v0.3.12

boring-semantic-layer v0.3.12

Highlights

This release focuses on join planning and reliability.

• Added deferred join_one support for dimension lookups after aggregation (#220)
• Added grain-aware join_one behavior via is_entity dimensions (#219)
• Pruned unused dimension joins from generated queries for leaner SQL (#228)
• Fixed demo_bsl_v2.py so the example script works again (#236)
• Added regression coverage for read-only DuckDB joined dimension introspection (#237, refs #232)

Why upgrade

If you use joined semantic models, this release improves correctness and efficiency:

• better handling of dimension lookups across joins
• better join behavior when entity grain is defined explicitly
• fewer unnecessary joins in compiled queries
• better protection against regressions around read-only DuckDB joined introspection (#232)

Included changes

• #219 — grain-aware join_one via is_entity dimensions
• #220 — deferred join_one for dimension lookups after aggregation
• #228 — prune unused dimension joins from generated queries
• #232 — regression coverage for read-only DuckDB joined introspection
• #236 — repair demo_bsl_v2.py
• #237 — add regression test coverage for read-only joined dimensions

Install

pip install -U boring-semantic-layer==0.3.12

v0.3.11

What's Changed

• fix: route dimension-only queries to source table on joins (#225)
  • Dimension-only queries (no measures) on joined models now route directly to the source dimension table
  • Ensures dimension members with no matching fact rows are still returned
  • Supports filters on the target dimension table
  • Falls back to standard join path when filters reference other tables

Full Changelog: v0.3.10...v0.3.11

v0.3.10

What's New

Features

• HTTP server transport — Optional bsl serve command exposes BSL models over REST via FastAPI (pip install boring-semantic-layer[server]) (#214)
• Per-dimension time grains — time_grains parameter lets each time dimension use a different grain (e.g., {"order_date": "month", "ship_date": "quarter"}) (#214)
• Short grain names — Core query() now accepts "month", "quarter", etc. alongside "TIME_GRAIN_MONTH" (#214)
• Derived time dimensions — Auto-generate year/month/day parts from time dimensions via derived_dimensions (#215, closes #213)
• Plain ibis backend support — BSL works with any ibis backend (Databricks, ClickHouse, MySQL, etc.) without xorq wrapping (#217)
• Self-joins with aliases — Join the same model multiple times in YAML with different aliases (#205, addresses #114)
• YAML calculated_measures — Support calculated_measures and .all() in YAML definitions (#211)

Fixes

• Fix content hashing so models with different measures/dimensions produce different hashes (#209)
• Fix 15 broken links across documentation (#208)
• Fix model prefix behavior in with_dimensions/with_measures after joins (#211)

Internal

• Convert to_untagged() to return xorq-vendored ibis types consistently (#203)

v0.3.9

What's Changed

Features

• Measure filtering: Support filtering by measures in query() (#178)
• MCP tool: Add search_dimension_values MCP tool for discovering dimension values

Refactoring

• Serialization: Move serialization into dedicated package with singledispatch architecture
• xorq vendored types: Convert to_untagged() to return xorq-vendored ibis types at the leaf (#203)
• Eliminate cloudpickle: Remove cloudpickle from xorq tag serialization (v2.0)
• Deduplicate: Consolidate deserialization, join deprecation, and xorq unwrapping

Bug Fixes

• Preserve mutated group-by columns in join_many pre-aggregation (#187)
• Add join methods to SemanticFilter for from_config compatibility (#186)
• Support multi-level derived dimensions in filters (#182)
• Strip VARCHAR(N) length from dtype strings in _make_schema (#183)
• Make join_one use left outer join by default (Malloy semantics)
• Preserve unmatched left-join rows in pre-aggregation
• Resolve ambiguous field reference in multi-way joins with shared column names
• Propagate is_right_of_many through nested joins in pre-aggregation
• Compute actual ibis join depth for column rename suffix
• Resolve unprefixed measure names via suffix matching on joined models
• Split join expression in from_tagged to prevent cross-join on round-trip

Performance

• Compute COUNT DISTINCT on raw source table when possible

Full Changelog: v0.3.8...v0.3.9

v0.3.8

What's Changed

• fix: dim bridge for 3+ join_many arms by @hussainsultan in #165
• feat: deferred group by aggregate + fan/chasm trap by @hussainsultan in #164
• feat: structured resolver serialization by @hussainsultan in #170
• fix: resolve prefixed dimension/measure names on standalone models by @boringdata in #167
• feat: fix repr and require FDQN names by @hussainsultan in #171
• Fix multi-layer derived query resolution (dims, calcs, and t.all chaining) by @boringdata in #168
• fix: non-additive measure re-aggregation by @hussainsultan in #173
• fix: unwrap nested chart spec in dashboard renderer by @boringdata in #158
• fix: support coalesce and post-ops in YAML measure definitions by @Angel-RC in #160
• fix: SemanticAggregateOp.values/.schema fails with calc measures by @hussainsultan in #174
• fix: resolve .type() on derived dimensions and measures by @boringdata in #176

New Contributors

• @Angel-RC made their first contribution in #160

Full Changelog: v0.3.7...v0.3.8

v0.3.7

What's Changed

• feat: add CSS theming support via YAML frontmatter (#157)

Full Changelog: v0.3.6...v0.3.7

v0.3.6

What's Changed

• ECharts chart backend - New visualization backend with bar, line, pie, scatter, area charts and auto chart detection
• Dashboard renderer - 16-column CSS grid layout for markdown-based dashboards
• Altair remains default - ECharts available via backend="echarts"

Full Changelog: v0.3.5...v0.3.6

v0.3.5

What's Changed

• fix: resolve time_grain infinite recursion with Deferred expressions by @benjaminsicard in #152
  • Fix infinite recursion when using time_grain with Deferred dimension expressions
  • Fix timestamp comparison errors when using time_range with certain database backends
  • Use Dimension.__call__ for time_range filters (properly resolves Deferred)
  • Add regression tests for Deferred expressions with time_grain and time_range

Full Changelog: v0.3.4...v0.3.5