[kvlonge]

Why

Nested CASTs that span lines printed hugged together, as CAST(CAST(( on one line, with two groups closing at once.

Changes

• feat(format): a parenthesised group that spans lines opens its inside on its own line and closes on its own line. VALUES rows, DECODE and OVER keep their layout.
• A long line breaks at || first, then + and -, then *, /, // and %. An operator that continues a group's only operand keeps the group's indentation. A tighter operator under a looser one, such as * d under + c, keeps the extra indent so the layout reads with the right precedence.
• 15 of the 1,066 format corpus entries change layout.
• Lines between a CASE and its END no longer break before AND, OR or an arithmetic operator.
• docs: the landing page and policy page examples show the new layout.

Converted SQL

The convert layout still keeps case and refuses any change other than whitespace. This diff compares origin/main with this branch using the playground's analyze at its default settings, over 2,962 inputs:

Corpus	Inputs	Changed
basics	1,842	473
everyday	615	189
everyday-models	499	263
examples	6	1

All 926 changes are whitespace only. Outcomes, diagnostics, raises and columns are identical, with 0 panics. Examples are in the first comment.

Verification

• make verify-recorded: 12,236 cases, 0 failed, 0 unproven. translation-results shows 0 changes, and sql-sizes is unchanged at 13,903,263 B.
• The crate tests for format, blocks and the convert layout pass. The playground tests pass 14 of 14.
• The docs CLI samples match the new layout.

[kvlonge]

Converted SQL before (origin/main) and after (this branch), from the harness diff:

The playground's decimal example (example:decimal):

--- old
+++ new
@@ -1,14 +1,20 @@
 SELECT
     "ORDERS"."ORDER_ID" AS "ORDER_ID",
-    CAST(CAST((
-        sign(CAST(CAST("ORDERS"."TOTAL" AS DECIMAL(38, 2)) * 100000000 AS HUGEINT))
-        * (
+    CAST(
+        CAST(
             (
-                2 * abs(CAST(CAST("ORDERS"."TOTAL" AS DECIMAL(38, 2)) * 100000000 AS HUGEINT)) + 3
-            )
-            // 6
+                sign(CAST(CAST("ORDERS"."TOTAL" AS DECIMAL(38, 2)) * 100000000 AS HUGEINT))
+                * (
+                    (
+                        2
+                        * abs(CAST(CAST("ORDERS"."TOTAL" AS DECIMAL(38, 2)) * 100000000 AS HUGEINT))
+                        + 3
+                    )
+                    // 6
+                )
+            ) AS DECIMAL(38, 0)
         )
-    ) AS DECIMAL(38, 0))
-    * 0.00000001 AS DECIMAL(18, 8)) AS "THIRD"
+        * 0.00000001 AS DECIMAL(18, 8)
+    ) AS "THIRD"
 FROM "ORDERS" AS "ORDERS"
 ORDER BY "ORDERS"."ORDER_ID" ASC NULLS LAST

everyday: a sum of two columns cast to a decimal (everyday#183):

--- old
+++ new
@@ -6,6 +6,8 @@
 
 SELECT
-    CAST((
-        CAST("T"."A" AS DECIMAL(38, 0)) + CAST("T"."B" AS DECIMAL(38, 0))
-    ) AS DECIMAL(2, 0)) AS "S"
+    CAST(
+        (
+            CAST("T"."A" AS DECIMAL(38, 0)) + CAST("T"."B" AS DECIMAL(38, 0))
+        ) AS DECIMAL(2, 0)
+    ) AS "S"
 FROM "T" AS "T"

basics: a literal added to a column, cast to a decimal (basics#481):

--- old
+++ new
@@ -1,5 +1,7 @@
 SELECT
-    CAST((
-        CAST(1 AS DECIMAL(38, 0)) + CAST("ORDERS"."TOTAL" AS DECIMAL(38, 2))
-    ) AS DECIMAL(13, 2)) AS "V"
+    CAST(
+        (
+            CAST(1 AS DECIMAL(38, 0)) + CAST("ORDERS"."TOTAL" AS DECIMAL(38, 2))
+        ) AS DECIMAL(13, 2)
+    ) AS "V"
 FROM "ORDERS" AS "ORDERS"

Not in this change: An operator placed by the earlier pass that leads multi-line operands still indents one level more than one this change places. One everyday-models query shows both, for example || '-' and || COALESCE( in a dbt surrogate key. This is cosmetic, and fixing it changes that older rule, so it is left for a separate change.