fix(pretty-print): preserve PostgreSQL syntax in five printer edge cases

crates/pgls_pretty_print/src/nodes/alter_table_stmt.rs

@@ -103,6 +103,16 @@ fn emit_alter_table_cmd_impl(e: &mut EventEmitter, cmd: &AlterTableCmd, for_type
             } else {
                 e.token(TokenKind::COLUMN_KW);
             }
+            // Dropping the flag turns an idempotent DDL into one that fails on a second run, so
+            // the AST round trip guard refuses the whole statement rather than let it through.
+            if cmd.missing_ok {
+                e.space();
+                e.token(TokenKind::IF_KW);
+                e.space();
+                e.token(TokenKind::NOT_KW);
+                e.space();
+                e.token(TokenKind::EXISTS_KW);
+            }
             if let Some(ref def) = cmd.def {
                 e.space();
                 e.indent_start();

crates/pgls_pretty_print/src/nodes/boolean_test.rs

@@ -10,7 +10,20 @@ pub(super) fn emit_boolean_test(e: &mut EventEmitter, n: &BooleanTest) {
 
     // Emit the argument
     if let Some(ref arg) = n.arg {
+        // AND, OR and NOT bind more loosely than the postfix IS test, so the grouping is lost
+        // unless it is spelled out: `(a OR b) IS TRUE` would come back as `a OR b IS TRUE`, which
+        // parses as `a OR (b IS TRUE)`. Every other argument kind binds tighter and needs nothing.
+        let needs_parens = matches!(arg.node.as_ref(), Some(pgls_query::NodeEnum::BoolExpr(_)));
+
+        if needs_parens {
+            e.token(TokenKind::L_PAREN);
+        }
+
         super::emit_node(arg, e);
+
+        if needs_parens {
+            e.token(TokenKind::R_PAREN);
+        }
     }
 
     e.line(LineType::SoftOrSpace);

crates/pgls_pretty_print/src/nodes/constraint.rs

@@ -231,6 +231,18 @@ pub(super) fn emit_constraint(e: &mut EventEmitter, n: &Constraint) {
 
             e.token(TokenKind::UNIQUE_KW);
 
+            // For a table constraint the qualifier sits between the keyword and the column list,
+            // unlike CREATE INDEX where it follows the list. Dropping it would let two NULLs
+            // coexist where the schema forbids it.
+            if n.nulls_not_distinct {
+                e.space();
+                e.token(TokenKind::NULLS_KW);
+                e.space();
+                e.token(TokenKind::NOT_KW);
+                e.space();
+                e.token(TokenKind::DISTINCT_KW);
+            }
+
             if !n.keys.is_empty() {
                 e.space();
                 e.token(TokenKind::L_PAREN);

crates/pgls_pretty_print/src/nodes/select_stmt.rs

@@ -348,6 +348,10 @@ fn emit_distinct_clause(e: &mut EventEmitter, clause: &[Node]) {
         return;
     }
 
+    // The clause has its own group so that it can stay on one line while the target list breaks,
+    // which is the usual shape of a DISTINCT ON query.
+    e.group_start(GroupKind::SelectStmt);
+
     e.space();
     e.token(TokenKind::ON_KW);
     e.space();
@@ -364,7 +368,12 @@ fn emit_distinct_clause(e: &mut EventEmitter, clause: &[Node]) {
     }
 
     e.indent_end();
+    // Symmetric with the Soft after the opening parenthesis: without it the closing parenthesis
+    // stays glued to the last expression when the list breaks.
+    e.line(LineType::Soft);
     e.token(TokenKind::R_PAREN);
+
+    e.group_end();
 }
 
 /// Determines if we need parentheses around a set operation operand.

crates/pgls_pretty_print/src/nodes/string.rs

@@ -129,6 +129,22 @@ pub(super) fn emit_identifier_maybe_quoted(e: &mut EventEmitter, value: &str) {
     }
 }
 
+/// Emits a type name part, honouring `type_case` when the name can be written unquoted.
+///
+/// A name that needs quotes keeps its own spelling: `"MyType"` and `"MYTYPE"` are two different
+/// types in PostgreSQL, so casing it would change which type is referenced.
+pub(super) fn emit_type_identifier_maybe_quoted(e: &mut EventEmitter, value: &str) {
+    if value.is_empty() {
+        return;
+    }
+
+    if needs_quoting(value) {
+        emit_identifier(e, value);
+    } else {
+        e.token(TokenKind::TYPE_IDENT(value.to_string()));
+    }
+}
+
 pub(super) fn emit_keyword(e: &mut EventEmitter, keyword: &str) {
     if let Some(token) = TokenKind::from_keyword(keyword) {
         e.token(token);
@@ -254,3 +270,35 @@ fn pick_dollar_delimiter(body: &str, hint: DollarQuoteHint) -> String {
         counter += 1;
     }
 }
+
+#[cfg(test)]
+mod tests {
+    use super::*;
+    use crate::renderer::{KeywordCase, RenderConfig, Renderer};
+
+    fn render(value: &str, type_case: KeywordCase) -> String {
+        let mut emitter = EventEmitter::new();
+        emit_type_identifier_maybe_quoted(&mut emitter, value);
+
+        let mut output = String::new();
+        let config = RenderConfig {
+            type_case,
+            ..Default::default()
+        };
+        let mut renderer = Renderer::new(&mut output, config);
+        renderer.render(emitter.events).expect("render");
+        output
+    }
+
+    #[test]
+    fn an_unquoted_type_name_follows_the_type_case() {
+        assert_eq!(render("object_id", KeywordCase::Upper), "OBJECT_ID");
+        assert_eq!(render("object_id", KeywordCase::Lower), "object_id");
+    }
+
+    #[test]
+    fn a_type_name_needing_quotes_keeps_its_own_case() {
+        assert_eq!(render("MyType", KeywordCase::Upper), "\"MyType\"");
+        assert_eq!(render("OBJECT_ID", KeywordCase::Lower), "\"OBJECT_ID\"");
+    }
+}

crates/pgls_pretty_print/src/nodes/type_name.rs

@@ -5,7 +5,7 @@ use crate::{
 };
 use pgls_query::protobuf::{self, TypeName};
 
-use super::string::emit_identifier_maybe_quoted;
+use super::string::{emit_identifier_maybe_quoted, emit_type_identifier_maybe_quoted};
 
 const INTERVAL_MASK_MONTH: i32 = 1 << 1;
 const INTERVAL_MASK_YEAR: i32 = 1 << 2;
@@ -69,7 +69,7 @@ fn emit_normalized_type_name(e: &mut EventEmitter, name_parts: &[String]) {
     if let Some(words) = builtin_type_keywords(name_parts) {
         emit_keyword_sequence(e, words);
     } else if !name_parts.is_empty() {
-        emit_dot_separated_name(e, name_parts);
+        emit_dot_separated_type_name(e, name_parts);
     } else {
         e.token(TokenKind::IDENT("<?>".to_string()));
     }
@@ -93,6 +93,15 @@ fn emit_dot_separated_name(e: &mut EventEmitter, name_parts: &[String]) {
     }
 }
 
+fn emit_dot_separated_type_name(e: &mut EventEmitter, name_parts: &[String]) {
+    for (index, part) in name_parts.iter().enumerate() {
+        if index > 0 {
+            e.token(TokenKind::DOT);
+        }
+        emit_type_identifier_maybe_quoted(e, part);
+    }
+}
+
 fn builtin_type_keywords(name_parts: &[String]) -> Option<&'static [&'static str]> {
     if name_parts.is_empty() {
         return None;

crates/pgls_pretty_print/tests/data/single/alter_table_add_column_if_not_exists.sql

@@ -0,0 +1,3 @@
+ALTER TABLE s.t
+	ADD COLUMN IF NOT EXISTS row_id TEXT,
+	ADD COLUMN IF NOT EXISTS designation_source TEXT;

crates/pgls_pretty_print/tests/data/single/boolean_test_bool_expr.sql

@@ -0,0 +1,4 @@
+SELECT id
+	FROM s.units
+	WHERE (units.has_history OR units.has_shares OR units.has_calls) IS TRUE
+	  AND (units.is_active AND units.is_visible) IS NOT FALSE;

crates/pgls_pretty_print/tests/data/single/distinct_on_long_list.sql

@@ -0,0 +1,3 @@
+SELECT DISTINCT ON (addresses.hash) addresses.hash AS origin_hash, addresses.entity_type, addresses.entity_fk
+FROM normalization.addresses
+ORDER BY addresses.hash, addresses.entity_type;

crates/pgls_pretty_print/tests/data/single/qualified_type_cast.sql

@@ -0,0 +1,5 @@
+SELECT
+	CAST(t.id AS public.object_id),
+	CAST(t.ids AS public.object_id[]),
+	CAST(t.n AS bigint)
+FROM s.t;

crates/pgls_pretty_print/tests/data/single/unique_constraint_nulls_not_distinct.sql

@@ -0,0 +1,5 @@
+CREATE TABLE s.expense_types (
+	expense_type_legacy_id TEXT,
+	legacy_branch_code TEXT,
+	UNIQUE NULLS NOT DISTINCT (expense_type_legacy_id, legacy_branch_code)
+);

crates/pgls_pretty_print/tests/snapshots/multi/tests__groupingsets_100.snap

@@ -1472,9 +1472,7 @@ select not a from (values (true)) as t (a) group by rollup (not a) having not no
 
 select not a from (values (true)) as t (a) group by rollup (not a) having not not a;
 
-select distinct on (
-  a,
-  b)
+select distinct on (a, b)
   a,
   b
 from
@@ -1485,9 +1483,7 @@ group by grouping sets ((a, b), a)
 order by a,
   b;
 
-select distinct on (
-  a,
-  b)
+select distinct on (a, b)
   a,
   b
 from
@@ -1498,9 +1494,7 @@ group by grouping sets ((a, b), a)
 order by a,
   b;
 
-select distinct on (
-  a,
-  b + 1)
+select distinct on (a, b + 1)
   a,
   b + 1
 from
@@ -1511,9 +1505,7 @@ group by grouping sets ((a, b + 1), a)
 order by a,
   b + 1;
 
-select distinct on (
-  a,
-  b + 1)
+select distinct on (a, b + 1)
   a,
   b + 1
 from

crates/pgls_pretty_print/tests/snapshots/multi/tests__groupingsets_80.snap

@@ -1933,9 +1933,7 @@ group by rollup (not a)
 having
   not not a;
 
-select distinct on (
-  a,
-  b)
+select distinct on (a, b)
   a,
   b
 from
@@ -1946,9 +1944,7 @@ group by grouping sets ((a, b), a)
 order by a,
   b;
 
-select distinct on (
-  a,
-  b)
+select distinct on (a, b)
   a,
   b
 from
@@ -1959,9 +1955,7 @@ group by grouping sets ((a, b), a)
 order by a,
   b;
 
-select distinct on (
-  a,
-  b + 1)
+select distinct on (a, b + 1)
   a,
   b + 1
 from
@@ -1972,9 +1966,7 @@ group by grouping sets ((a, b + 1), a)
 order by a,
   b + 1;
 
-select distinct on (
-  a,
-  b + 1)
+select distinct on (a, b + 1)
   a,
   b + 1
 from

crates/pgls_pretty_print/tests/snapshots/multi/tests__memoize_100.snap

@@ -1,7 +1,6 @@
 ---
 source: crates/pgls_pretty_print/tests/tests.rs
 input_file: crates/pgls_pretty_print/tests/data/multi/memoize.sql
-snapshot_kind: text
 ---
 create function explain_memoize(query text, hide_hitmiss boolean)
 returns setof text
@@ -431,8 +430,7 @@ from
   tab_anti as t1
   left outer join
     lateral (
-      select distinct on (
-        a)
+      select distinct on (a)
         a,
         b,
         t1.a as x

crates/pgls_pretty_print/tests/snapshots/multi/tests__memoize_80.snap

@@ -1,7 +1,6 @@
 ---
 source: crates/pgls_pretty_print/tests/tests.rs
 input_file: crates/pgls_pretty_print/tests/data/multi/memoize.sql
-snapshot_kind: text
 ---
 create function explain_memoize(query text, hide_hitmiss boolean)
 returns setof text
@@ -457,8 +456,7 @@ from
   tab_anti as t1
   left outer join
     lateral (
-      select distinct on (
-        a)
+      select distinct on (a)
         a,
         b,
         t1.a as x

crates/pgls_pretty_print/tests/snapshots/multi/tests__select_distinct_on_100.snap

@@ -1,10 +1,8 @@
 ---
 source: crates/pgls_pretty_print/tests/tests.rs
 input_file: crates/pgls_pretty_print/tests/data/multi/select_distinct_on.sql
-snapshot_kind: text
 ---
-select distinct on (
-  string4)
+select distinct on (string4)
   string4,
   two,
   ten
@@ -14,9 +12,7 @@ order by string4 using <,
   two using >,
   ten using <;
 
-select distinct on (
-  string4,
-  ten)
+select distinct on (string4, ten)
   string4,
   two,
   ten
@@ -26,9 +22,7 @@ order by string4 using <,
   two using <,
   ten using <;
 
-select distinct on (
-  string4,
-  ten)
+select distinct on (string4, ten)
   string4,
   ten,
   two
@@ -74,9 +68,7 @@ select distinct on (y, x) x, y from distinct_on_tbl order by y;
 
 select distinct on (y, x) x, y from distinct_on_tbl order by y;
 
-select distinct on (
-  y,
-  x)
+select distinct on (y, x)
   x,
   y
 from
@@ -88,9 +80,7 @@ order by y,
   x,
   z;
 
-select distinct on (
-  y,
-  x)
+select distinct on (y, x)
   x,
   y
 from