Support extension syntax without GROUP BY per paper

Views can now define measures without explicit GROUP BY:

  CREATE VIEW sales_v AS
  SELECT year, region, SUM(amount) AS MEASURE revenue
  FROM sales;

Yardstick automatically adds GROUP BY ALL to make it valid SQL.
This follows Julian Hyde's paper more closely.

Author: nicosuave

README.md

@@ -49,8 +49,7 @@ SELECT
     region,
     SUM(amount) AS MEASURE revenue,
     COUNT(*) AS MEASURE order_count
-FROM sales
-GROUP BY year, region;
+FROM sales;
 
 -- Query with AGGREGATE() and AT modifiers (SEMANTIC prefix required)
 SEMANTIC SELECT
@@ -110,11 +109,10 @@ SELECT
     dimension1,
     dimension2,
     AGG(expr) AS MEASURE measure_name
-FROM table
-GROUP BY dimension1, dimension2;
+FROM table;
 ```
 
-Supported aggregations: `SUM`, `COUNT`, `AVG`, `MIN`, `MAX`
+Yardstick automatically handles the grouping. Supported aggregations: `SUM`, `COUNT`, `AVG`, `MIN`, `MAX`
 
 ### Querying Measures
 

test/sql/measures.test

@@ -23,8 +23,7 @@ INSERT INTO sales VALUES
 statement ok
 CREATE VIEW sales_v AS
 SELECT year, region, SUM(amount) AS MEASURE revenue
-FROM sales
-GROUP BY year, region;
+FROM sales;
 
 # Basic query on view with measure
 query IIR rowsort
@@ -87,8 +86,7 @@ GROUP BY year, region;
 statement ok
 CREATE VIEW sales_yearly AS
 SELECT year, SUM(amount) AS MEASURE revenue
-FROM sales
-GROUP BY year;
+FROM sales;
 
 # Prior year comparison
 query IR rowsort
@@ -181,8 +179,7 @@ SELECT
     SUM(amount) AS MEASURE total_revenue,
     COUNT(*) AS MEASURE order_count,
     AVG(amount) AS MEASURE avg_order
-FROM sales
-GROUP BY year;
+FROM sales;
 
 query IRR rowsort
 SEMANTIC SELECT year, AGGREGATE(total_revenue), AGGREGATE(avg_order) FROM orders_v GROUP BY year;
@@ -349,8 +346,7 @@ INSERT INTO products VALUES
 statement ok
 CREATE VIEW products_v AS
 SELECT year, region, category, SUM(amount) AS MEASURE revenue
-FROM products
-GROUP BY year, region, category;
+FROM products;
 
 # Grand total across all dimensions
 query IIIR rowsort
@@ -410,8 +406,7 @@ GROUP BY year, region, category;
 statement ok
 CREATE VIEW sales_minmax AS
 SELECT year, region, MIN(amount) AS MEASURE min_sale, MAX(amount) AS MEASURE max_sale
-FROM sales
-GROUP BY year, region;
+FROM sales;
 
 query IIRR rowsort
 SEMANTIC SELECT year, region, AGGREGATE(min_sale), AGGREGATE(max_sale) FROM sales_minmax GROUP BY year, region;
@@ -489,8 +484,7 @@ INSERT INTO adjustments VALUES
 statement ok
 CREATE VIEW adj_v AS
 SELECT year, region, SUM(amount) AS MEASURE adjustment
-FROM adjustments
-GROUP BY year, region;
+FROM adjustments;
 
 query IIR rowsort
 SEMANTIC SELECT year, region, AGGREGATE(adjustment) FROM adj_v GROUP BY year, region;
@@ -523,8 +517,7 @@ INSERT INTO yearly_data VALUES (2020, 100), (2021, 120), (2022, 150), (2023, 180
 statement ok
 CREATE VIEW yearly_v AS
 SELECT year, SUM(amount) AS MEASURE revenue
-FROM yearly_data
-GROUP BY year;
+FROM yearly_data;
 
 query IRR rowsort
 SEMANTIC SELECT
@@ -647,8 +640,7 @@ GROUP BY year;
 statement ok
 CREATE VIEW quarterly AS
 SELECT year, quarter, SUM(amount) AS MEASURE revenue
-FROM (VALUES (2022, 1, 100), (2022, 2, 120), (2022, 3, 90), (2022, 4, 140)) AS t(year, quarter, amount)
-GROUP BY year, quarter;
+FROM (VALUES (2022, 1, 100), (2022, 2, 120), (2022, 3, 90), (2022, 4, 140)) AS t(year, quarter, amount);
 
 query IIR rowsort
 SEMANTIC SELECT
@@ -670,8 +662,7 @@ GROUP BY year, quarter;
 statement ok
 CREATE OR REPLACE VIEW case_measure AS
 SELECT year, CASE WHEN SUM(amount) > 150 THEN 1 ELSE 0 END AS MEASURE high_value
-FROM (VALUES (2022, 100), (2022, 50), (2023, 200), (2023, 100)) AS t(year, amount)
-GROUP BY year;
+FROM (VALUES (2022, 100), (2022, 50), (2023, 200), (2023, 100)) AS t(year, amount);
 
 query II rowsort
 SEMANTIC SELECT year, AGGREGATE(high_value)
@@ -699,8 +690,7 @@ SELECT year,
     SUM(revenue) AS MEASURE revenue,
     SUM(cost) AS MEASURE cost,
     revenue - cost AS MEASURE profit
-FROM financials
-GROUP BY year;
+FROM financials;
 
 # Query basic measures
 query IRR rowsort
@@ -747,8 +737,7 @@ INSERT INTO daily_orders VALUES
 statement ok
 CREATE VIEW daily_orders_v AS
 SELECT order_date, SUM(amount) AS MEASURE revenue
-FROM daily_orders
-GROUP BY order_date;
+FROM daily_orders;
 
 # SET with expression dimension: fix to a specific month
 # This generates WHERE MONTH(_inner.order_date) = 2
@@ -790,10 +779,10 @@ INSERT INTO fact_returns VALUES (2022, 'US', 10), (2022, 'EU', 5), (2023, 'US',
 
 # Create two measure views
 statement ok
-CREATE VIEW fact_orders_v AS SELECT year, region, SUM(amount) AS MEASURE revenue FROM fact_orders GROUP BY year, region;
+CREATE VIEW fact_orders_v AS SELECT year, region, SUM(amount) AS MEASURE revenue FROM fact_orders;
 
 statement ok
-CREATE VIEW fact_returns_v AS SELECT year, region, SUM(return_amount) AS MEASURE refunds FROM fact_returns GROUP BY year, region;
+CREATE VIEW fact_returns_v AS SELECT year, region, SUM(return_amount) AS MEASURE refunds FROM fact_returns;
 
 # Query measures from both fact tables in a JOIN
 query IIRR rowsort

yardstick-rs/src/sql/measures.rs

@@ -983,9 +983,6 @@ fn extract_measures_from_sql(sql: &str) -> Result<(String, Vec<ViewMeasure>, Opt
             replacements.push((remove_start, info.name_end, String::new()));
         } else {
             // Base measure: replace "AS MEASURE name" with "AS name"
-            let pattern_start = info.name_end - " AS MEASURE ".len() - info.name.len()
-                - (sql[info.expr_start..info.name_end].len() - info.expression.len() - " AS MEASURE ".len() - info.name.len());
-            // Actually, simpler: find " AS MEASURE " before name_end
             let chunk = &sql[info.expr_start..info.name_end];
             if let Some(am_pos) = chunk.to_uppercase().find(" AS MEASURE ") {
                 let abs_start = info.expr_start + am_pos;
@@ -1010,6 +1007,32 @@ fn extract_measures_from_sql(sql: &str) -> Result<(String, Vec<ViewMeasure>, Opt
         clean_sql = format!("{}{}{}", &clean_sql[..start], replacement, &clean_sql[end..]);
     }
 
+    // If there are aggregate measures but no GROUP BY, add GROUP BY ALL
+    // This enables the "extension" syntax from the paper where views define
+    // measures without explicit grouping
+    let has_aggregate_measure = measures.iter().any(|m| find_aggregation_in_expression(&m.expression).is_some());
+    let clean_sql_upper = clean_sql.to_uppercase();
+    let has_group_by = clean_sql_upper.contains("GROUP BY");
+
+    if has_aggregate_measure && !has_group_by {
+        // Find insertion point: before ORDER BY, LIMIT, or at end
+        let insert_pos = ["ORDER BY", "LIMIT", ";"]
+            .iter()
+            .filter_map(|kw| clean_sql_upper.find(kw))
+            .min()
+            .unwrap_or(clean_sql.len());
+
+        clean_sql = format!(
+            "{} GROUP BY ALL{}",
+            clean_sql[..insert_pos].trim_end(),
+            if insert_pos < clean_sql.len() {
+                format!(" {}", clean_sql[insert_pos..].trim_start())
+            } else {
+                String::new()
+            }
+        );
+    }
+
     Ok((clean_sql, measures, view_name))
 }
 
@@ -2249,6 +2272,33 @@ mod tests {
         assert!(!result.clean_sql.contains("revenue - cost"));
     }
 
+    #[test]
+    #[serial]
+    fn test_process_create_view_without_group_by() {
+        clear_measure_views();
+
+        // Per the paper, views can define measures without GROUP BY
+        let sql = r#"CREATE VIEW orders_extended AS
+SELECT
+    id,
+    product,
+    region,
+    amount,
+    SUM(amount) AS MEASURE revenue
+FROM orders"#;
+        let result = process_create_view(sql);
+
+        eprintln!("is_measure_view: {}", result.is_measure_view);
+        eprintln!("clean_sql: {}", result.clean_sql);
+        eprintln!("measures: {:?}", result.measures);
+
+        assert!(result.is_measure_view);
+        assert_eq!(result.measures.len(), 1);
+        assert_eq!(result.measures[0].column_name, "revenue");
+        // The clean_sql should NOT contain the measure column at all for no-groupby views
+        // because it can't be evaluated without a GROUP BY
+    }
+
     #[test]
     fn test_has_at_syntax() {
         assert!(has_at_syntax(