[wadler] Prettier handling of trivia

Split the rules for pre- and post- trivia, understand when we want to
do either, handle multi-line-break (in an unsatisfying way, I guess)
but otherwise lay the groundwork for thinking about it better.

Also now we don't generate lazy "Text" nodes because I thought I might
want to actually look at the newlines in the source but I don't yet.
I *can* now, though. (I can also detect EOF so there's that.)

grammar.py

@@ -24,7 +24,7 @@ class FineGrammar(Grammar):
     # generator = parser.GenerateLR1
     start = "File"
 
-    trivia = ["BLANKS", "LINE_BREAKS", "COMMENT"]
+    trivia = ["BLANKS", "LINE_BREAK", "COMMENT"]
 
     pretty_indent = "  "
 
@@ -426,7 +426,7 @@ class FineGrammar(Grammar):
         return self.IDENTIFIER | group(self.IDENTIFIER, self.COLON, indent(sp, self.expression))
 
     BLANKS = Terminal(Re.set(" ", "\t").plus())
-    LINE_BREAKS = Terminal(Re.set("\r", "\n").plus(), trivia_mode=TriviaMode.NewLine)
+    LINE_BREAK = Terminal(Re.set("\r", "\n"), trivia_mode=TriviaMode.NewLine)
     COMMENT = Terminal(
         Re.seq(Re.literal("//"), Re.set("\n").invert().star()),
         highlight=highlight.comment.line,

harness.py

@@ -371,7 +371,7 @@ class Harness:
 
             printer = self.load_printer()
             if self.tree is not None:
-                self.document = printer.convert_tree_to_document(self.tree)
+                self.document = printer.convert_tree_to_document(self.tree, self.source)
             else:
                 self.document = None
 
@@ -541,12 +541,6 @@ class Harness:
                 append(f"indent {doc.amount}")
                 self.format_document(lines, doc.doc, indent + 1)
 
-            case wadler.Text(start, end):
-                if self.source is not None:
-                    append(f"< {repr(self.source[start:end])}")
-                else:
-                    append(f"< ??? {start}:{end}")
-
             case wadler.Literal(text):
                 append(f"literal {repr(text)}")
 

parser/parser.py

@@ -2109,6 +2109,10 @@ class Re:
 UNICODE_MAX_CP = 1114112
 
 
+def _str_repr(x: int) -> str:
+    return repr(chr(x))[1:-1]
+
+
 @dataclasses.dataclass
 class ReSet(Re):
     values: list[Span]
@@ -2165,12 +2169,12 @@ class ReSet(Re):
         if len(self.values) == 1:
             span = self.values[0]
             if len(span) == 1:
-                return chr(span.lower)
+                return _str_repr(span.lower)
 
         ranges = []
         for span in self.values:
-            start = chr(span.lower)
+            start = _str_repr(span.lower)
-            end = chr(span.upper - 1)
+            end = _str_repr(span.upper - 1)
             if start == end:
                 ranges.append(start)
             else:
@@ -2736,7 +2740,7 @@ class TriviaMode(enum.Enum):
     pretty-printing. Attach this to a "trivia_mode" property on a Terminal
     definition.
 
-    - Ignore means that the trivia should be ignored. (This is the default.)
+    - Blank means that the trivia represents blank space. (This is the default.)
 
     - NewLine means that the trivia is a line break. This is important for
       other modes, specifically...
@@ -2748,7 +2752,7 @@ class TriviaMode(enum.Enum):
       a forced break.
     """
 
-    Ignore = 0
+    Blank = 0
     NewLine = 1
     LineComment = 2
 

parser/wadler.py

@@ -32,12 +32,6 @@ class Indent:
     doc: "Document"
 
 
-@dataclasses.dataclass(frozen=True)
-class Text:
-    start: int
-    end: int
-
-
 @dataclasses.dataclass(frozen=True)
 class Literal:
     text: str
@@ -69,13 +63,11 @@ class Lazy:
         return self.value
 
     @classmethod
-    def from_tree(cls, tree: runtime.Tree, printer: "Printer") -> "Lazy":
+    def from_tree(cls, tree: runtime.Tree, src: str, printer: "Printer") -> "Lazy":
-        return Lazy(lambda: printer.convert_tree_to_document(tree))
+        return Lazy(lambda: printer.convert_tree_to_document(tree, src))
 
 
-Document = (
+Document = None | Literal | NewLine | ForceBreak | Cons | Indent | Group | Trivia | Marker | Lazy
-    None | Text | Literal | NewLine | ForceBreak | Cons | Indent | Group | Trivia | Marker | Lazy
-)
 
 
 def cons(*documents: Document) -> Document:
@@ -207,9 +199,6 @@ def layout_document(doc: Document, width: int, indent: str) -> DocumentLayout:
                 case None:
                     pass
 
-                case Text(start, end):
-                    remaining -= end - start
-
                 case Literal(text):
                     remaining -= len(text)
 
@@ -268,10 +257,6 @@ def layout_document(doc: Document, width: int, indent: str) -> DocumentLayout:
             case None:
                 pass
 
-            case Text(start, end):
-                output.append((start, end))
-                column += end - start
-
             case Literal(text):
                 output.append(text)
                 column += len(text)
@@ -337,7 +322,7 @@ def resolve_document(doc: Document) -> Document:
         case Trivia(child):
             return Trivia(resolve_document(child))
 
-        case Text() | Literal() | NewLine() | ForceBreak() | Indent() | None:
+        case Literal() | NewLine() | ForceBreak() | Indent() | None:
             return doc
 
         case _:
@@ -358,7 +343,12 @@ class Matcher:
     newline_replace: dict[str, str]
     trivia_mode: dict[str, parser.TriviaMode]
 
-    def match(self, printer: "Printer", items: list[runtime.Tree | runtime.TokenValue]) -> Document:
+    def match(
+        self,
+        printer: "Printer",
+        items: list[runtime.Tree | runtime.TokenValue],
+        src: str,
+    ) -> Document:
         stack: list[tuple[int, Document]] = [(0, None)]
         table = self.table
 
@@ -434,10 +424,13 @@ class Matcher:
                     value = current_token[1]
 
                     if isinstance(value, runtime.Tree):
-                        child = Lazy.from_tree(value, printer)
+                        child = Lazy.from_tree(value, src, printer)
                     else:
-                        child = Text(value.start, value.end)
+                        child = cons(
-                        child = cons(child, self.apply_trivia(value.post_trivia))
+                            trivia(self.apply_pre_trivia(value.pre_trivia, src)),
+                            Literal(src[value.start : value.end]),
+                            trivia(self.apply_post_trivia(value.post_trivia, src)),
+                        )
 
                     stack.append((action.state, child))
                     input_index += 1
@@ -445,46 +438,100 @@ class Matcher:
                 case parser.Error():
                     raise Exception("How did I get a parse error here??")
 
-    def apply_trivia(self, trivia_tokens: list[runtime.TokenValue]) -> Document:
+    def slice_pre_post_trivia(self, trivia_tokens: list[runtime.TokenValue], src: str) -> tuple[
-        has_newline = False
+        list[tuple[parser.TriviaMode, runtime.TokenValue]],
+        list[tuple[parser.TriviaMode, runtime.TokenValue]],
+    ]:
+        tokens = [
+            (self.trivia_mode.get(token.kind, parser.TriviaMode.Blank), token)
+            for token in trivia_tokens
+        ]
+
+        for index, (mode, token) in enumerate(tokens):
+            if token.start == 0:
+                # Everything is pre-trivia if we're at the start of the file.
+                return (tokens, [])
+
+            if mode == parser.TriviaMode.NewLine:
+                # This is the first newline; it belongs with the post-trivia.
+                return (tokens[index + 1 :], tokens[: index + 1])
+
+        # If we never found a new line then it's all post-trivia.
+        return ([], tokens)
+
+    def apply_pre_trivia(self, trivia_tokens: list[runtime.TokenValue], src: str) -> Document:
+        pre_trivia, _ = self.slice_pre_post_trivia(trivia_tokens, src)
+        if len(pre_trivia) == 0:
+            return None
+
+        at_start_of_file = pre_trivia[0][1].start == 0
+
         trivia_doc = None
-        for token in trivia_tokens:
+        new_line_count = 0
-            mode = self.trivia_mode.get(token.kind, parser.TriviaMode.Ignore)
+        for mode, token in pre_trivia:
             match mode:
-                case parser.TriviaMode.Ignore:
+                case parser.TriviaMode.LineComment:
+                    trivia_doc = cons(
+                        trivia_doc,
+                        Literal(src[token.start : token.end]),
+                        ForceBreak(False),
+                    )
+                    new_line_count = 0  # There will be a newline after this.
+                    at_start_of_file = False
+
+                case parser.TriviaMode.Blank:
                     pass
 
                 case parser.TriviaMode.NewLine:
-                    # We ignore line breaks because obviously
+                    new_line_count += 1
-                    # we expect the pretty-printer to put the
+                    if new_line_count == 2 and not at_start_of_file:
-                    # line breaks in where they belong *but*
+                        trivia_doc = cons(
-                    # we track if they happened to influence
+                            trivia_doc,
-                    # the layout.
+                            ForceBreak(False),
-                    has_newline = True
+                            ForceBreak(False),
-
+                        )
-                case parser.TriviaMode.LineComment:
-                    if has_newline:
-                        # This line comment is all alone on
-                        # its line, so we need to maintain
-                        # that.
-                        line_break = NewLine("")
-                    else:
-                        # This line comment is attached to
-                        # something to the left, reduce it to
-                        # a space.
-                        line_break = Literal(" ")
-
-                    trivia_doc = cons(
-                        trivia_doc,
-                        line_break,
-                        Text(token.start, token.end),
-                        ForceBreak(True),  # This is probably the wrong place for this!
-                    )
 
                 case _:
                     typing.assert_never(mode)
 
-        return trivia(trivia_doc)
+        return trivia_doc
+
+    def apply_post_trivia(self, trivia_tokens: list[runtime.TokenValue], src: str) -> Document:
+        _, post_trivia = self.slice_pre_post_trivia(trivia_tokens, src)
+        if len(post_trivia) == 0:
+            return None
+
+        trivia_doc = None
+        for mode, token in post_trivia:
+            match mode:
+                case parser.TriviaMode.Blank:
+                    pass
+
+                case parser.TriviaMode.NewLine:
+                    # Anything after a line break is not processed as post
+                    # trivia.
+                    break
+
+                case parser.TriviaMode.LineComment:
+                    # Because this is post-trivia, we know there's something
+                    # to our left, and we can force the space.
+                    trivia_doc = cons(
+                        Literal(" "),
+                        Literal(src[token.start : token.end]),
+                        ForceBreak(True),  # And the line needs to end.
+                    )
+                    break
+
+                case _:
+                    typing.assert_never(mode)
+
+        if len(trivia_tokens) > 0 and trivia_tokens[-1].end == len(src):
+            # As a special case, if we're post trivia at the end of the file
+            # then we also need to be pre-trivia too, for the hypthetical EOF
+            # token that we never see.
+            trivia_doc = cons(trivia_doc, self.apply_pre_trivia(trivia_tokens, src))
+
+        return trivia_doc
 
 
 class Printer:
@@ -686,19 +733,19 @@ class Printer:
 
         return result
 
-    def convert_tree_to_document(self, tree: runtime.Tree) -> Document:
+    def convert_tree_to_document(self, tree: runtime.Tree, src: str) -> Document:
         name = tree.name
         assert name is not None, "Cannot format a tree if it still has transparent nodes inside"
 
         rule = self.lookup_nonterminal(name)
         matcher = self.rule_to_matcher(rule)
-        m = matcher.match(self, list(tree.children))
+        m = matcher.match(self, list(tree.children), src)
         if m is None:
             raise ValueError(
                 f"Could not match a valid tree for {tree.name} with {len(tree.children)} children:\n{tree.format()}"
             )
         return resolve_document(m)
 
-    def format_tree(self, tree: runtime.Tree, width: int) -> DocumentLayout:
+    def format_tree(self, tree: runtime.Tree, src: str, width: int) -> DocumentLayout:
-        doc = self.convert_tree_to_document(tree)
+        doc = self.convert_tree_to_document(tree, src)
         return layout_document(doc, width, self._indent)

tests/test_wadler.py

@@ -14,6 +14,7 @@ from parser.parser import (
     sp,
     nl,
     br,
+    TriviaMode,
 )
 
 import parser.runtime as runtime
@@ -72,6 +73,7 @@ class JsonGrammar(Grammar):
         )
 
     BLANKS = Terminal(Re.set(" ", "\t", "\r", "\n").plus())
+
     LCURLY = Terminal("{")
     RCURLY = Terminal("}")
     COMMA = Terminal(",")
@@ -118,8 +120,6 @@ def flatten_document(doc: wadler.Document, src: str) -> list:
             return ["<forced break>"]
         case wadler.Indent():
             return [[f"<indent {doc.amount}>", flatten_document(doc.doc, src)]]
-        case wadler.Text(start, end):
-            return [src[start:end]]
         case wadler.Literal(text):
             return [text]
         case wadler.Group():
@@ -149,7 +149,7 @@ def test_convert_tree_to_document():
     assert tree is not None
 
     printer = wadler.Printer(JSON)
-    doc = flatten_document(printer.convert_tree_to_document(tree), text)
+    doc = flatten_document(printer.convert_tree_to_document(tree, text), text)
 
     assert doc == [
         [
@@ -212,7 +212,7 @@ def test_layout_basic():
     assert tree is not None
 
     printer = wadler.Printer(JSON)
-    result = printer.format_tree(tree, 50).apply_to_source(text)
+    result = printer.format_tree(tree, text, 50).apply_to_source(text)
 
     assert (
         result
@@ -226,38 +226,44 @@ def test_layout_basic():
     )
 
 
+class TG(Grammar):
+    start = "root"
+    trivia = ["BLANKS", "LINE_BREAK", "COMMENT"]
+
+    @rule
+    def root(self):
+        return self._expression
+
+    @rule
+    def _expression(self):
+        return self.word | self.list
+
+    @rule
+    def list(self):
+        return group(self.LPAREN, indent(nl, self._expressions), nl, self.RPAREN)
+
+    @rule
+    def _expressions(self):
+        return self._expression | seq(self._expressions, sp, self._expression)
+
+    @rule
+    def word(self):
+        return self.OK | seq(self.BREAK, br, self.BREAK)
+
+    LPAREN = Terminal("(")
+    RPAREN = Terminal(")")
+    OK = Terminal("ok")
+    BREAK = Terminal("break")
+
+    BLANKS = Terminal(Re.set(" ", "\t").plus())
+    LINE_BREAK = Terminal(Re.set("\r", "\n"), trivia_mode=TriviaMode.NewLine)
+    COMMENT = Terminal(
+        Re.seq(Re.literal(";"), Re.set("\n").invert().star()),
+        trivia_mode=TriviaMode.LineComment,
+    )
+
+
 def test_forced_break():
-    class TG(Grammar):
-        start = "root"
-        trivia = ["BLANKS"]
-
-        @rule
-        def root(self):
-            return self._expression
-
-        @rule
-        def _expression(self):
-            return self.word | self.list
-
-        @rule
-        def list(self):
-            return group(self.LPAREN, indent(nl, self._expressions), nl, self.RPAREN)
-
-        @rule
-        def _expressions(self):
-            return self._expression | seq(self._expressions, sp, self._expression)
-
-        @rule
-        def word(self):
-            return self.OK | seq(self.BREAK, br, self.BREAK)
-
-        LPAREN = Terminal("(")
-        RPAREN = Terminal(")")
-        OK = Terminal("ok")
-        BREAK = Terminal("break")
-
-        BLANKS = Terminal(Re.set(" ", "\t", "\r", "\n").plus())
-
     g = TG()
     g_lexer = g.compile_lexer()
     g_parser = runtime.Parser(g.build_table())
@@ -269,7 +275,7 @@ def test_forced_break():
     assert tree is not None
 
     printer = wadler.Printer(g)
-    result = printer.format_tree(tree, 200).apply_to_source(text)
+    result = printer.format_tree(tree, text, 200).apply_to_source(text)
 
     assert (
         result