@ttsc/lint 0.14.0-dev.20260529.2 → 0.14.0

package/linthost/rules_format_statement_split.go

@@ -0,0 +1,368 @@
+package linthost
+
+import (
+  "strings"
+
+  shimast "github.com/microsoft/typescript-go/shim/ast"
+  shimscanner "github.com/microsoft/typescript-go/shim/scanner"
+)
+
+// formatStatementSplit puts every statement in a statement list on its
+// own physical line, mirroring Prettier's "one statement per line"
+// layout. Prettier never leaves two statements sharing a source line,
+// `const a = 1; let b = 2;` becomes two lines, and this rule is the
+// always-on equivalent.
+//
+// Scope is every statement list the language produces: the SourceFile
+// body, every Block and ModuleBlock, and the statement lists of `case`
+// and `default` clauses. The rule registers for KindSourceFile and
+// walks the subtree itself (like `format/sort-imports` and
+// `format/jsdoc`), emitting one finding that may carry many TextEdits.
+//
+// Per-statement decision:
+//
+//  1. Find the statement's first non-trivia byte.
+//  2. If it is the first non-whitespace byte on its physical line, the
+//     statement already starts its own line, abstain (that is
+//     `format/indent`'s surface).
+//  3. Otherwise another statement precedes it on the same line. Replace
+//     the whitespace run immediately before the statement (back to the
+//     previous non-whitespace char) with EOL + indent(depth).
+//
+// Safety: abstain for a statement whose inter-statement gap holds a
+// `//` or `/*` comment. Re-emitting EOL + indent over that gap would
+// delete the comment, so the rule leaves the bytes untouched and a
+// later pass (or the user) handles it.
+//
+// Idempotent: once each statement is alone on its line, step 2 abstains
+// for all of them and the rule emits nothing.
+type formatStatementSplit struct{}
+
+// formatStatementSplitOptions carries the indentation + EOL settings the
+// rule needs to synthesize the inserted line break. The JSON tags match
+// the `format` block keys the config layer mirrors in (see
+// `expandFormatBlock` in config_format.go).
+type formatStatementSplitOptions struct {
+  TabWidth  *int    `json:"tabWidth"`
+  UseTabs   *bool   `json:"useTabs"`
+  EndOfLine *string `json:"endOfLine"`
+}
+
+func (formatStatementSplit) Name() string   { return "format/statement-split" }
+func (formatStatementSplit) IsFormat() bool { return true }
+func (formatStatementSplit) Visits() []shimast.Kind {
+  return []shimast.Kind{shimast.KindSourceFile}
+}
+
+func (formatStatementSplit) Check(ctx *Context, node *shimast.Node) {
+  if ctx == nil || ctx.File == nil {
+    return
+  }
+  layout := loadFormatLayout(ctx)
+  src := ctx.File.Text()
+  var edits []TextEdit
+  forEachStatementInList(ctx.File, func(stmt *shimast.Node, depth int) {
+    // Empty statements (`;`) carry no content. Splitting each one onto
+    // its own line only multiplies blank-ish noise, so abstain.
+    if stmt.Kind == shimast.KindEmptyStatement {
+      return
+    }
+    start := shimscanner.SkipTrivia(src, stmt.Pos())
+    if start <= 0 || start > len(src) {
+      return
+    }
+    // Walk back over the whitespace run that precedes the statement.
+    ws := start
+    for ws > 0 && (src[ws-1] == ' ' || src[ws-1] == '\t') {
+      ws--
+    }
+    if ws == 0 || src[ws-1] == '\n' {
+      // First non-whitespace byte on its line already; that is
+      // `format/indent`'s job, not this rule's.
+      return
+    }
+    // Do not split a statement off a leading-semicolon ASI guard
+    // (`;(expr)`): the lone `;` before it is an empty-statement guard
+    // that format/orphan-semi intentionally merged onto this line, and
+    // re-splitting it would oscillate against that rule. The `;` is a
+    // guard (not a terminator like `foo();bar()`) when only whitespace
+    // precedes it back to the start of its line.
+    if src[ws-1] == ';' {
+      k := ws - 1
+      for k > 0 && (src[k-1] == ' ' || src[k-1] == '\t') {
+        k--
+      }
+      if k == 0 || src[k-1] == '\n' {
+        return
+      }
+    }
+    // A block that opens right after its own `case`/`default` label
+    // (`case 2: {`) is not sharing a line with a preceding statement;
+    // the only thing before it is the clause label, and Prettier keeps
+    // the brace on the label line. Abstain so the rule does not break
+    // the block off into `case 2:\n{`.
+    if stmt.Kind == shimast.KindBlock && firstStatementAfterCaseLabel(stmt) {
+      return
+    }
+    // The gap between the previous statement and this one must be pure
+    // whitespace. A `//` or `/*` anywhere from the previous statement's
+    // end to this one would be eaten by the replacement, so abstain. The
+    // scan starts at the previous statement boundary, not at `ws`: a
+    // comment is non-whitespace, so a scan that begins at `ws` (the end
+    // of the immediate whitespace run) can never see it.
+    if gapHasComment(src, prevStatementEnd(src, ws), start) {
+      return
+    }
+    edits = append(edits, TextEdit{
+      Pos:  ws,
+      End:  start,
+      Text: layout.eol + layout.indent(depth),
+    })
+  })
+  if len(edits) == 0 {
+    return
+  }
+  ctx.ReportRangeFix(
+    edits[0].Pos,
+    edits[0].End,
+    "Each statement must begin on its own line.",
+    edits...,
+  )
+}
+
+// gapHasComment reports whether the byte range [start, end) contains the
+// opening bytes of a `//` or `/*` comment. The split rule abstains when
+// the inter-statement gap carries a comment so its line-break insertion
+// never deletes the comment.
+func gapHasComment(src string, start, end int) bool {
+  if start < 0 {
+    start = 0
+  }
+  for i := start; i+1 < end && i+1 < len(src); i++ {
+    if src[i] == '/' && (src[i+1] == '/' || src[i+1] == '*') {
+      return true
+    }
+  }
+  return false
+}
+
+// prevStatementEnd returns the offset where the content preceding `stmt`
+// on its line ends, used as the lower bound of the inter-statement gap
+// scanned for comments. It locates `stmt` in its parent statement list
+// and returns the previous sibling's end; for the first statement of a
+// list it returns the parent owner's start (the `{`/`case:` boundary).
+// The fallback walks back from `floor` to the previous non-whitespace
+// byte so a comment-free caller still gets a sane lower bound.
+func prevStatementEnd(src string, floor int) int {
+  // floor sits just past the whitespace run before the statement, i.e.
+  // immediately after the previous content's last byte. Walking back one
+  // byte lands inside that content; the gap to scan is everything from
+  // the start of the current line up to floor, but a comment belongs to
+  // this gap only when it follows the previous statement's terminator on
+  // the same line. Scanning from the line start would also catch a
+  // comment that belongs to an earlier statement, so bound the scan at
+  // the nearest preceding statement terminator (`;`, `}`) or label
+  // (`:`), or the line start, whichever comes first.
+  i := floor - 1
+  for i >= 0 {
+    c := src[i]
+    if c == '\n' {
+      return i + 1
+    }
+    if c == ';' || c == '}' || c == '{' || c == ':' {
+      return i + 1
+    }
+    i--
+  }
+  return 0
+}
+
+// firstStatementAfterCaseLabel reports whether `stmt` is the first
+// statement of a `case`/`default` clause. Such a statement follows only
+// the clause label on its line, not a preceding statement, so the split
+// rule must not break a block off its label (`case 2: {` stays intact).
+func firstStatementAfterCaseLabel(stmt *shimast.Node) bool {
+  if stmt == nil || stmt.Parent == nil {
+    return false
+  }
+  parent := stmt.Parent
+  if parent.Kind != shimast.KindCaseClause && parent.Kind != shimast.KindDefaultClause {
+    return false
+  }
+  clause := parent.AsCaseOrDefaultClause()
+  if clause == nil || clause.Statements == nil || len(clause.Statements.Nodes) == 0 {
+    return false
+  }
+  return clause.Statements.Nodes[0] == stmt
+}
+
+// formatLayout is the resolved indentation + EOL snapshot the structural
+// format rules share. Defaults match Prettier: 2-space indent, spaces
+// (not tabs), LF newlines.
+type formatLayout struct {
+  tabWidth int
+  useTabs  bool
+  eol      string
+}
+
+// indent returns the indentation string for `depth` nesting levels:
+// `depth` tab characters when useTabs, otherwise `depth * tabWidth`
+// spaces.
+func (l formatLayout) indent(depth int) string {
+  if depth <= 0 {
+    return ""
+  }
+  if l.useTabs {
+    return strings.Repeat("\t", depth)
+  }
+  return strings.Repeat(" ", depth*l.tabWidth)
+}
+
+// loadFormatLayout decodes the shared tabWidth/useTabs/endOfLine options
+// used by `format/statement-split` and `format/indent` and applies the
+// Prettier defaults. Both rules carry the identical option struct shape,
+// so the decode is funneled through the statement-split option struct.
+func loadFormatLayout(ctx *Context) formatLayout {
+  var opts formatStatementSplitOptions
+  _ = ctx.DecodeOptions(&opts)
+  layout := formatLayout{tabWidth: 2, useTabs: false, eol: "\n"}
+  if opts.TabWidth != nil && *opts.TabWidth > 0 {
+    layout.tabWidth = *opts.TabWidth
+  }
+  if opts.UseTabs != nil {
+    layout.useTabs = *opts.UseTabs
+  }
+  if opts.EndOfLine != nil && *opts.EndOfLine == "crlf" {
+    layout.eol = "\r\n"
+  }
+  return layout
+}
+
+// forEachStatementInList walks the file's AST and invokes `fn` once for
+// every statement that lives directly inside a statement list, the
+// SourceFile body, a Block, a ModuleBlock, or a case/default clause.
+// `depth` is the nesting level used to compute indentation: one level
+// per enclosing Block or ModuleBlock plus one per enclosing case/default
+// clause. Top-level statements are depth 0.
+//
+// The SourceFile body's statements are visited at depth 0, then the walk
+// recurses through the whole subtree via `ForEachChild`, bumping the
+// depth on each statement-list owner it descends into. The rule's
+// KindSourceFile registration therefore sees every nested statement
+// without the engine dispatching one node kind at a time.
+func forEachStatementInList(file *shimast.SourceFile, fn func(stmt *shimast.Node, depth int)) {
+  if file == nil || file.Statements == nil {
+    return
+  }
+  for _, stmt := range file.Statements.Nodes {
+    if stmt == nil {
+      continue
+    }
+    fn(stmt, 0)
+  }
+  walkStatementLists(file.AsNode(), file.Text(), 0, fn)
+}
+
+// blockStartsOwnLine reports whether a brace block opens on its own line, the
+// `{` being the first non-whitespace byte of its physical line. A case/default
+// body block written as `case X: {` shares the label's line (false); one
+// written as `case X:` then `{` on the next line starts its own line (true).
+// Prettier indents the former like a braceless `case X: stmt` (no extra level)
+// but the latter like an ordinary nested block (one level deeper).
+func blockStartsOwnLine(src string, block *shimast.Node) bool {
+  if block == nil {
+    return false
+  }
+  pos := shimscanner.SkipTrivia(src, block.Pos())
+  if pos < 0 || pos > len(src) {
+    return false
+  }
+  lineStart := lineStartOffset(src, pos)
+  for i := lineStart; i < pos; i++ {
+    if src[i] != ' ' && src[i] != '\t' {
+      return false
+    }
+  }
+  return true
+}
+
+// walkStatementLists recurses through `node`'s children. When it
+// descends into a statement-list owner (Block, ModuleBlock, case/default
+// clause) it bumps `depth` and invokes `fn` for each statement that
+// owner directly holds, so a statement is always reported at the depth
+// of the list it belongs to. The SourceFile body is visited by the
+// caller, so this function only handles the nested owners.
+//
+// Some nodes are descend-only +1 frames: they are not statement lists
+// themselves but their child statement lists nest one column deeper than
+// the node. Two cases need this:
+//
+//   - KindCaseBlock wraps a switch's clauses
+//     (SwitchStatement -> CaseBlock -> CaseClause -> statements). The
+//     braces sit one level under `switch`, and the clause body sits one
+//     level under `case:`, so a case-body statement is switchDepth+2.
+//     The clause's own +1 stacks on top of the CaseBlock's +1.
+//   - A class/interface/object-type-literal/object-literal body is not a
+//     statement list, but a member's method/constructor Block nests
+//     inside it. Without counting the body frame a method-body statement
+//     would land one column short (the member Block's +1 only), so
+//     already-correct 4-space class bodies would be rewritten to 2.
+func walkStatementLists(node *shimast.Node, src string, depth int, fn func(stmt *shimast.Node, depth int)) {
+  if node == nil {
+    return
+  }
+  node.ForEachChild(func(child *shimast.Node) bool {
+    if child == nil {
+      return false
+    }
+    childDepth := depth
+    switch child.Kind {
+    case shimast.KindBlock, shimast.KindModuleBlock:
+      childDepth = depth + 1
+      // A same-line `case X: { stmt }` block does not add an indent level:
+      // Prettier indents it exactly like a braceless `case X: stmt`, the
+      // clause having already contributed the level. A block written on its
+      // OWN line under the clause (`case X:` then `{` on the next line) is an
+      // ordinary nested block and keeps its extra level, so only collapse the
+      // same-line form.
+      if child.Kind == shimast.KindBlock && child.Parent != nil &&
+        (child.Parent.Kind == shimast.KindCaseClause ||
+          child.Parent.Kind == shimast.KindDefaultClause) &&
+        !blockStartsOwnLine(src, child) {
+        childDepth = depth
+      }
+      for _, stmt := range child.Statements() {
+        if stmt == nil {
+          continue
+        }
+        fn(stmt, childDepth)
+      }
+    case shimast.KindCaseClause, shimast.KindDefaultClause:
+      childDepth = depth + 1
+      clause := child.AsCaseOrDefaultClause()
+      if clause != nil && clause.Statements != nil {
+        for _, stmt := range clause.Statements.Nodes {
+          if stmt == nil {
+            continue
+          }
+          fn(stmt, childDepth)
+        }
+      }
+    case shimast.KindCaseBlock,
+      shimast.KindClassDeclaration,
+      shimast.KindClassExpression,
+      shimast.KindInterfaceDeclaration,
+      shimast.KindTypeLiteral,
+      shimast.KindObjectLiteralExpression:
+      // Descend-only +1 frame: not a visited statement list, but its
+      // nested statement lists sit one column deeper.
+      childDepth = depth + 1
+    }
+    walkStatementLists(child, src, childDepth, fn)
+    return false
+  })
+}
+
+func init() {
+  Register(formatStatementSplit{})
+}

package/linthost/rules_format_ternary_nullish_parens.go

@@ -0,0 +1,77 @@
+package linthost
+
+import (
+  shimast "github.com/microsoft/typescript-go/shim/ast"
+  shimscanner "github.com/microsoft/typescript-go/shim/scanner"
+)
+
+// formatTernaryNullishParens parenthesizes a `??` expression used as the
+// condition, consequent, or alternate of a conditional expression,
+// matching Prettier 3. TypeScript permits an unparenthesized `??` in
+// those positions (unlike `??` mixed with `||`/`&&`), but Prettier 3
+// always adds the parens for clarity:
+//
+//  cond ? a ?? b : c   ->   cond ? (a ?? b) : c
+//  a ?? b ? c : d      ->   (a ?? b) ? c : d
+//
+// Only `??` is wrapped; `||` and `&&` operands are left bare. The wrap is
+// width-independent (it applies even when the ternary fits on one line),
+// so it is a normalization rule rather than part of the print-width
+// reflow. Idempotent: a wrapped operand parses as a parenthesized
+// expression, not a bare `??`, so the next pass leaves it alone.
+type formatTernaryNullishParens struct{}
+
+func (formatTernaryNullishParens) Name() string   { return "format/ternary-nullish-parens" }
+func (formatTernaryNullishParens) IsFormat() bool { return true }
+
+func (formatTernaryNullishParens) Visits() []shimast.Kind {
+  return []shimast.Kind{shimast.KindConditionalExpression}
+}
+
+func (formatTernaryNullishParens) Check(ctx *Context, node *shimast.Node) {
+  if ctx == nil || ctx.File == nil || node == nil {
+    return
+  }
+  cond := node.AsConditionalExpression()
+  if cond == nil {
+    return
+  }
+  src := ctx.File.Text()
+  var edits []TextEdit
+  for _, operand := range []*shimast.Node{cond.Condition, cond.WhenTrue, cond.WhenFalse} {
+    if !isBareNullishCoalescing(operand) {
+      continue
+    }
+    start := shimscanner.SkipTrivia(src, operand.Pos())
+    end := operand.End()
+    if start < 0 || end < start || end > len(src) {
+      continue
+    }
+    edits = append(edits, TextEdit{Pos: start, End: end, Text: "(" + src[start:end] + ")"})
+  }
+  if len(edits) == 0 {
+    return
+  }
+  ctx.ReportRangeFix(
+    edits[0].Pos,
+    edits[0].End,
+    "Parenthesize a nullish-coalescing operand of a conditional expression.",
+    edits...,
+  )
+}
+
+// isBareNullishCoalescing reports whether `node` is an unparenthesized
+// `a ?? b` binary expression.
+func isBareNullishCoalescing(node *shimast.Node) bool {
+  if node == nil || node.Kind != shimast.KindBinaryExpression {
+    return false
+  }
+  bin := node.AsBinaryExpression()
+  return bin != nil &&
+    bin.OperatorToken != nil &&
+    bin.OperatorToken.Kind == shimast.KindQuestionQuestionToken
+}
+
+func init() {
+  Register(formatTernaryNullishParens{})
+}

package/linthost/rules_format_trailing_comma.go

@@ -5,7 +5,7 @@ import (
 )
 
 // formatTrailingComma adds trailing commas to multi-line lists. Mirrors
-// prettier's `trailingComma: "all"` default — *not* a tunable.
+// prettier's `trailingComma: "all"` default, *not* a tunable.
 //
 // Scope (intentionally narrower than the closing-brace surface of TS):
 //
@@ -109,6 +109,14 @@ func (formatTrailingComma) Check(ctx *Context, node *shimast.Node) {
     if call == nil {
       return
     }
+    if isDynamicImportCall(call) {
+      // Prettier never emits a trailing comma inside a dynamic
+      // `import(...)` argument list, even under trailingComma:"all", a
+      // documented exception because the import() spec historically
+      // rejected one. The printer half (print_nodes_call.go) honors the
+      // same exception so the two rules agree on the reflowed shape.
+      return
+    }
     considerTrailingComma(ctx, call.Arguments, node.End()-1)
   case shimast.KindNewExpression:
     if mode == "es5" {
@@ -268,12 +276,23 @@ func (formatTrailingComma) Check(ctx *Context, node *shimast.Node) {
   }
 }
 
+// isDynamicImportCall reports whether a CallExpression is a dynamic
+// `import(...)`: its callee is the `import` keyword itself, mirroring
+// typescript-go's ast.IsImportCall. Prettier never emits a trailing
+// comma inside a dynamic-import argument list, even under
+// trailingComma:"all".
+func isDynamicImportCall(call *shimast.CallExpression) bool {
+  return call != nil &&
+    call.Expression != nil &&
+    call.Expression.Kind == shimast.KindImportKeyword
+}
+
 // considerTrailingComma reports a fix when the bracket-delimited list is
 // multi-line and missing its trailing comma. `closeBracketPos` points at
 // the closing punctuation byte itself (e.g. the `]` of an array literal).
 //
 // "Multi-line" means the close bracket sits on a different line from the
-// last element's end — not just "the list contains a newline somewhere".
+// last element's end, not just "the list contains a newline somewhere".
 // Prettier's `trailingComma: "all"` omits the comma whenever the close
 // bracket is adjacent to the last element on the same line, even if the
 // element itself is internally multi-line. The canonical shape is