chore: more work done and planning with AI.

pkg/format/body.go

@@ -92,10 +92,11 @@ func formatBodyInner(inner string, st config.Style) string {
 	var b strings.Builder
 
 	// Emit verbatim up to and including DECLARE (keyword-cased).
+	// Normalize CRLF in trivia so the output always uses st.Newline.
 	for i := 0; i <= declareIdx; i++ {
 		t := sig[i]
 		for _, tr := range t.Lead {
-			b.WriteString(tr.Text)
+			b.WriteString(strings.ReplaceAll(tr.Text, "\r\n", nl))
 		}
 		if i == declareIdx {
 			b.WriteString(applyCase(t.Tok.Text, st.KeywordCase))
@@ -122,57 +123,30 @@ func formatDeclareVars(b *strings.Builder, toks []cst.Tok, st config.Style) {
 	nl := st.Newline
 	indent := st.Indent
 	depth := 0
+	var decls [][]cst.Tok
 	var cur []cst.Tok
-	var preComments []string
+	var preCommentSets [][]string
+	var curPreComments []string
 
-	emit := func() {
+	collect := func() {
 		if len(cur) == 0 {
 			return
 		}
-		for _, c := range preComments {
-			b.WriteString(indent)
-			b.WriteString(c)
-			b.WriteString(nl)
-		}
-		preComments = nil
-
-		// Graceful degradation: mid-declaration comments stay verbatim.
-		if anyComment(cur[1:]) {
-			b.WriteString(indent)
-			b.WriteString(verbatimSpan(cur))
-			b.WriteString(nl)
-			cur = nil
-			return
-		}
-
-		body := cur
-		hasSemi := len(body) > 0 && body[len(body)-1].Tok.Kind == lexer.Semicolon
-		if hasSemi {
-			body = body[:len(body)-1]
-		}
-		b.WriteString(indent)
-		for i, t := range body {
-			if i > 0 && needSpace(body[i-1].Tok, t.Tok) {
-				b.WriteByte(' ')
-			}
-			b.WriteString(caseText(t.Tok, st))
-		}
-		if hasSemi {
-			b.WriteString(";")
-		}
-		b.WriteString(nl)
+		decls = append(decls, cur)
+		preCommentSets = append(preCommentSets, curPreComments)
 		cur = nil
+		curPreComments = nil
 	}
 
 	for _, t := range toks {
 		if len(cur) == 0 {
 			for _, tr := range t.Lead {
 				if tr.Kind == lexer.LineComment || tr.Kind == lexer.BlockComment {
-					preComments = append(preComments, strings.TrimRight(tr.Text, " \t"))
+					text := strings.TrimRight(strings.ReplaceAll(tr.Text, "\r", ""), " \t")
+					curPreComments = append(curPreComments, text)
 				}
 			}
 		}
-
 		switch t.Tok.Kind {
 		case lexer.LParen, lexer.LBracket:
 			depth++
@@ -181,14 +155,195 @@ func formatDeclareVars(b *strings.Builder, toks []cst.Tok, st config.Style) {
 				depth--
 			}
 		}
-
 		cur = append(cur, t)
-
 		if t.Tok.Kind == lexer.Semicolon && depth == 0 {
-			emit()
+			collect()
+		}
+	}
+	collect()
+
+	// Compute alignment widths when requested.
+	var nameColW, typeColW int
+	if st.PlpgsqlDeclareAlignType || st.PlpgsqlDeclareAlignEq {
+		for _, decl := range decls {
+			if anyComment(decl[1:]) {
+				continue
+			}
+			body := decl
+			if len(body) > 0 && body[len(body)-1].Tok.Kind == lexer.Semicolon {
+				body = body[:len(body)-1]
+			}
+			nw, tw := declareNameTypeWidth(body, st)
+			if nw > nameColW {
+				nameColW = nw
+			}
+			if tw > typeColW {
+				typeColW = tw
+			}
+		}
+	}
+
+	for i, cur := range decls {
+		for _, c := range preCommentSets[i] {
+			b.WriteString(indent)
+			b.WriteString(c)
+			b.WriteString(nl)
+		}
+		// Graceful degradation: mid-declaration comments stay verbatim.
+		if anyComment(cur[1:]) {
+			b.WriteString(indent)
+			b.WriteString(verbatimSpan(cur))
+			b.WriteString(nl)
+			continue
+		}
+
+		body := cur
+		hasSemi := len(body) > 0 && body[len(body)-1].Tok.Kind == lexer.Semicolon
+		if hasSemi {
+			body = body[:len(body)-1]
+		}
+		b.WriteString(indent)
+		if (st.PlpgsqlDeclareAlignType || st.PlpgsqlDeclareAlignEq) && nameColW > 0 {
+			writeDeclareAligned(b, body, st, nameColW, typeColW)
+		} else {
+			for j, t := range body {
+				if j > 0 && needSpace(body[j-1].Tok, t.Tok) {
+					b.WriteByte(' ')
+				}
+				b.WriteString(caseText(t.Tok, st))
+			}
+		}
+		if hasSemi {
+			b.WriteString(";")
+		}
+		b.WriteString(nl)
+	}
+}
+
+// declareNameTypeWidth returns the rendered width of the name and type portions
+// of a DECLARE variable declaration (without the default assignment).
+// Format is: [name type [:= default]] or [name type [DEFAULT default]].
+func declareNameTypeWidth(body []cst.Tok, st config.Style) (nameW, typeW int) {
+	if len(body) < 2 {
+		return 0, 0
+	}
+	// name is always the first token.
+	name := caseText(body[0].Tok, st)
+	nameW = len(name)
+
+	// type runs from body[1] until we hit := / DEFAULT / = at depth 0.
+	var typeTokens []cst.Tok
+	depth := 0
+	for _, t := range body[1:] {
+		switch t.Tok.Kind {
+		case lexer.LParen, lexer.LBracket:
+			depth++
+		case lexer.RParen, lexer.RBracket:
+			if depth > 0 {
+				depth--
+			}
+		}
+		if depth == 0 {
+			low := lowerASCII(t.Tok.Text)
+			if t.Tok.Kind == lexer.Operator && (t.Tok.Text == ":=" || t.Tok.Text == "=") {
+				break
+			}
+			if t.Tok.Kind == lexer.Ident && low == "default" {
+				break
+			}
+		}
+		typeTokens = append(typeTokens, t)
+	}
+	var tb strings.Builder
+	for j, t := range typeTokens {
+		if j > 0 && needSpace(typeTokens[j-1].Tok, t.Tok) {
+			tb.WriteByte(' ')
+		}
+		tb.WriteString(caseText(t.Tok, st))
+	}
+	typeW = len(tb.String())
+	return nameW, typeW
+}
+
+// writeDeclareAligned writes a single DECLARE variable with aligned columns.
+func writeDeclareAligned(b *strings.Builder, body []cst.Tok, st config.Style, nameColW, typeColW int) {
+	if len(body) == 0 {
+		return
+	}
+	name := caseText(body[0].Tok, st)
+	b.WriteString(name)
+	if len(body) == 1 {
+		return
+	}
+
+	// Pad name to nameColW if align_type is requested.
+	if st.PlpgsqlDeclareAlignType {
+		pad := nameColW - len(name)
+		for k := 0; k < pad; k++ {
+			b.WriteByte(' ')
+		}
+	}
+
+	// Collect type tokens.
+	var typeTokens, restTokens []cst.Tok
+	depth := 0
+	pastType := false
+	for _, t := range body[1:] {
+		switch t.Tok.Kind {
+		case lexer.LParen, lexer.LBracket:
+			depth++
+		case lexer.RParen, lexer.RBracket:
+			if depth > 0 {
+				depth--
+			}
+		}
+		if !pastType && depth == 0 {
+			low := lowerASCII(t.Tok.Text)
+			if (t.Tok.Kind == lexer.Operator && (t.Tok.Text == ":=" || t.Tok.Text == "=")) ||
+				(t.Tok.Kind == lexer.Ident && low == "default") {
+				pastType = true
+				restTokens = append(restTokens, t)
+				continue
+			}
+		}
+		if pastType {
+			restTokens = append(restTokens, t)
+		} else {
+			typeTokens = append(typeTokens, t)
+		}
+	}
+
+	var typeStr strings.Builder
+	for j, t := range typeTokens {
+		if j > 0 && needSpace(typeTokens[j-1].Tok, t.Tok) {
+			typeStr.WriteByte(' ')
+		}
+		typeStr.WriteString(caseText(t.Tok, st))
+	}
+	typeRendered := typeStr.String()
+
+	b.WriteByte(' ')
+	b.WriteString(typeRendered)
+
+	if len(restTokens) > 0 {
+		// Pad type to typeColW if align_eq is requested.
+		if st.PlpgsqlDeclareAlignEq {
+			pad := typeColW - len(typeRendered)
+			for k := 0; k < pad; k++ {
+				b.WriteByte(' ')
+			}
+		}
+		for j, t := range restTokens {
+			prev := restTokens[0].Tok
+			if j > 0 {
+				prev = restTokens[j-1].Tok
+			}
+			if j == 0 || needSpace(prev, t.Tok) {
+				b.WriteByte(' ')
+			}
+			b.WriteString(caseText(t.Tok, st))
 		}
 	}
-	emit()
 }
 
 // bline is one logical line within an accumulated statement.
@@ -208,12 +363,17 @@ type bline struct {
 //  3. After EXCEPTION the formatter switches to verbatim-indent mode (original
 //     leading whitespace is preserved) to avoid conflicts between styles that
 //     put WHEN at col-0 vs indented.
-//  4. Blank-line counts from the original are preserved.
+//  4. Blank-line counts from the original are preserved (capped by PlpgsqlMaxBlankLines).
 func formatBodyStatements(text string, st config.Style) string {
 	nl := st.Newline
 	normalised := strings.ReplaceAll(text, "\r\n", "\n")
 	rawLines := strings.Split(normalised, "\n")
 
+	maxBlanks := st.PlpgsqlMaxBlankLines
+	if maxBlanks < 0 {
+		maxBlanks = 0
+	}
+
 	var (
 		result        strings.Builder
 		stmt          []bline
@@ -228,7 +388,11 @@ func formatBodyStatements(text string, st config.Style) string {
 		if len(stmt) == 0 {
 			return
 		}
-		for i := 0; i < pendingBlanks; i++ {
+		blanks := pendingBlanks
+		if blanks > maxBlanks {
+			blanks = maxBlanks
+		}
+		for i := 0; i < blanks; i++ {
 			result.WriteString(nl)
 		}
 		pendingBlanks = 0
@@ -259,8 +423,6 @@ func formatBodyStatements(text string, st config.Style) string {
 			}
 			effectiveDepth = blockDepth
 		case "else", "elsif", "elseif":
-			// Emit at one level up; blockDepth unchanged so the body continues
-			// at the same depth (THEN will re-apply depthInc for elsif).
 			if blockDepth > 0 {
 				effectiveDepth = blockDepth - 1
 			}
@@ -271,7 +433,14 @@ func formatBodyStatements(text string, st config.Style) string {
 
 		baseIndent := strings.Repeat(st.Indent, effectiveDepth)
 
-		for i, ll := range stmt {
+		// plpgsql_if_then_newline: when false, THEN stays on the same line as
+		// the condition. When true (default) it's already on its own logical line.
+		stmtLines := stmt
+		if !st.PlpgsqlIfThenNewline && fw == "if" {
+			stmtLines = joinThenToCondition(stmt)
+		}
+
+		for i, ll := range stmtLines {
 			if i == 0 || ll.indent == "" {
 				result.WriteString(baseIndent)
 			} else {
@@ -301,8 +470,6 @@ func formatBodyStatements(text string, st config.Style) string {
 
 		fw := lowerASCII(firstBodyKeyword(stripped))
 		isColZero := indent == ""
-		// Only join when we are at paren-depth 0; content inside parens (e.g.
-		// inside a CTE subquery) should not be merged across lines.
 		joinToPrev := isColZero && parenDepth == 0 && len(stmt) > 0 && !sqlClauseKw[fw]
 
 		if joinToPrev {
@@ -312,7 +479,6 @@ func formatBodyStatements(text string, st config.Style) string {
 			stmt = append(stmt, bline{text: stripped, indent: indent})
 		}
 
-		// Scan tokens to track paren depth and detect flush triggers.
 		var lastD0Kw string
 		for _, tok := range lexer.Lex(stripped) {
 			if tok.IsTrivia() || tok.Kind == lexer.EOF {
@@ -327,6 +493,13 @@ func formatBodyStatements(text string, st config.Style) string {
 				}
 			case lexer.Semicolon:
 				if parenDepth == 0 {
+					// plpgsql_loop_collapse: fold empty FOR … LOOP END LOOP; to one line.
+					if st.PlpgsqlLoopCollapse && len(stmt) > 0 {
+						collapsed, ok := tryCollapseLoop(stmt, st)
+						if ok {
+							stmt = []bline{{text: collapsed, indent: ""}}
+						}
+					}
 					flush()
 				}
 			}
@@ -335,13 +508,9 @@ func formatBodyStatements(text string, st config.Style) string {
 			}
 		}
 
-		// Structural keywords at the end of a line (paren depth 0) trigger a
-		// flush and possibly a block-depth change.
 		if parenDepth == 0 && len(stmt) > 0 {
 			switch lastD0Kw {
 			case "then", "loop", "begin":
-				// ELSIF/ELSEIF headers end with THEN but must NOT increment depth
-				// (blockDepth is already at the right level for the body).
 				fw0 := lowerASCII(firstBodyKeyword(stmt[0].text))
 				if fw0 != "elsif" && fw0 != "elseif" {
 					depthInc = true
@@ -357,6 +526,52 @@ func formatBodyStatements(text string, st config.Style) string {
 	return result.String()
 }
 
+// joinThenToCondition merges a THEN line (on its own bline) into the preceding
+// condition line when plpgsql_if_then_newline is false.
+func joinThenToCondition(lines []bline) []bline {
+	out := make([]bline, 0, len(lines))
+	for i, ll := range lines {
+		if i > 0 && strings.EqualFold(strings.TrimSpace(ll.text), "then") {
+			out[len(out)-1].text = strings.TrimRight(out[len(out)-1].text, " \t") + " THEN"
+		} else {
+			out = append(out, ll)
+		}
+	}
+	return out
+}
+
+// tryCollapseLoop tries to collapse an empty loop body to one line.
+// Detects: FOR … LOOP\n  (empty or only blanks)\nEND LOOP;
+// Returns the collapsed line and true on success.
+func tryCollapseLoop(lines []bline, st config.Style) (string, bool) {
+	if len(lines) < 2 {
+		return "", false
+	}
+	first := strings.TrimSpace(lines[0].text)
+	last := strings.TrimSpace(lines[len(lines)-1].text)
+	firstLow := lowerASCII(first)
+	lastLow := lowerASCII(last)
+
+	// Check last line is END LOOP; or LOOP (for WHILE/FOR empty bodies that end with LOOP).
+	if !strings.HasPrefix(lastLow, "end loop") && lastLow != "end loop;" {
+		return "", false
+	}
+	// Check middle lines are all empty.
+	for _, mid := range lines[1 : len(lines)-1] {
+		if strings.TrimSpace(mid.text) != "" {
+			return "", false
+		}
+	}
+	// Check first line ends with LOOP.
+	if !strings.HasSuffix(firstLow, "loop") {
+		return "", false
+	}
+	_ = st
+	_ = firstLow
+	// Collapse to: <header> END LOOP;
+	return strings.TrimRight(first, " \t") + " " + strings.ToUpper(last), true
+}
+
 // firstBodyKeyword returns the text of the first identifier token in s
 // (lowercased), or "" if the first significant token is not an identifier.
 func firstBodyKeyword(s string) string {