Crate improv_duckdb_model 

DuckDB-free core of the improv DuckDB extension (crates/duckdb_improv).

The extension reads rows through DuckDB and hands them here as plain Rust values, the same split as improv_pg_model / pg_improv:

• calc: improv_calc / improv_calc_text, a scalar formula over named JSON inputs (compiled once, cached);
• function_rows: improv_functions, the built-in function registry;
• parse_spec, bind, the *_sql builders and run: improv(source, model), a small model spec over a table or query, the SQL that reads it, and the engine’s cells as result rows.

The model spec has one item per line (blank lines and # / -- comments are skipped):

category <Name> = <column>
input    <Name> = <column> [sum|avg|min|max|count]
<Name> = <formula>

Every input ranges over every category. A category’s items are its column’s distinct non-NULL values in the column’s own order (numbers numerically, dates chronologically), so PREV and NEXT follow it.

Modules

calc

improv_calc(formula [, inputs]): one scalar Improv formula over named inputs, the engine’s own per-cell operators ([improv_engine::scalar]), as LANGUAGE improv evaluates a function body.

Structs

Bound

The spec bound to the source’s columns.

CategorySpec

A category <Name> = <column> line.

Column

A source column as DESCRIBE reports it.

DerivedSpec

A <Name> = <formula> line.

FunctionRow

One row of improv_functions().

InputSpec

An input <Name> = <column> [agg] line.

Row

One result row of improv(source, model).

Spec

A parsed model spec.

Enums

CellValue

Re-exported so the extension shell needs no direct engine dependency. The dataflow encoding of a cell value.

Error

Everything that can go wrong building or running a model.

ValueType

Re-exported so the extension shell needs no direct engine dependency. The declared type of a measure.

Functions

bind

Resolve the spec’s columns against the source (DuckDB names are case-insensitive) and type its inputs: numbers, booleans and text; sum and avg need a number, count is a number.

describe_sql

The SQL listing the source’s columns and types.

function_rows

The built-in functions in registry order; with query, those [functions::search] finds, best match first.

input_sql

The SQL reading input i’s rows: every category column as text, then the value as a DOUBLE, BOOLEAN or VARCHAR, aggregated per coordinate when the input has an aggregation.

items_sql

The SQL listing a category’s items: its column’s distinct non-NULL values as text, in the column’s own order.

parse_spec

Parse a model spec. Errors name the offending line.

run

Build the model from the spec, its categories’ items (from items_sql, per category) and its inputs’ rows (from input_sql, per input), evaluate every derived measure, and return one row per derived cell: measures in declaration order, each one’s cells in item order.

source_sql

The source in a FROM: a query (SELECT, WITH, FROM, VALUES, or anything parenthesized) as a subquery, else a table or view name as written. A trailing ; is dropped.

Type Aliases

FactRow

One fact row: the category values (in declaration order) and the value.

Result

Result alias for this crate.