DeCarabas/lrparsers
1
0
Fork 0
Code Issues Pull requests Projects Releases Packages Wiki Activity

[wadler] Refactor: data and runtime split

Browse source 
Now we convert the grammar into data for a pretty-printer, so in
theoryw e could write the pretty-printer in a different language.
This commit is contained in:
John Doty 2024-09-21 06:44:53 -07:00
parent e4585170d8
commit 1f84752538
2 changed files with 252 additions and 215 deletions
Download patch file Download diff file Expand all files Collapse all files

451
parser/wadler.py
View file

@ -360,13 +360,243 @@ def slice_pre_post_trivia(
return ([], tokens) return ([], tokens)
############################################################################
# Data to Drive the Pretty Printer
############################################################################
@dataclasses.dataclass @dataclasses.dataclass
class Matcher: class MatcherTable:
"""Information necessary to create a document from a concrete parse tree,
as generated by the parser.
(In order to do this we need to re-parse the children of the tree, in
order to recover structure added by transparent rules. That's why each
MatcherTable has an associated ParseTable!)
"""
# Parse table to recover the node into a document
table: parser.ParseTable table: parser.ParseTable
# Mapping from the name of i_ rules to indent counts
indent_amounts: dict[str, int] indent_amounts: dict[str, int]
# Mapping from the names of n_ rules to the text they flatten to
newline_replace: dict[str, str] newline_replace: dict[str, str]
def _compile_nonterminal_matcher(
grammar: parser.Grammar,
nonterminals: dict[str, parser.NonTerminal],
rule: parser.NonTerminal,
) -> MatcherTable:
generated_grammar: list[typing.Tuple[str, list[str]]] = []
visited: set[str] = set()
# In order to generate groups, indents, and newlines we need to
# synthesize new productions. And it happens sometimes that we get
# duplicates, repeated synthetic productions. It's important to
# de-duplicate productions, otherwise we'll wind up with ambiguities
# in the parser.
#
# These dictionaries track the synthetic rules: the keys are
# production and also the parameter (if any), and the values are the
# names of the productions that produce the effect.
#
groups: dict[tuple[str, ...], str] = {}
indents: dict[tuple[tuple[str, ...], int], str] = {}
newlines: dict[tuple[tuple[str, ...], str], str] = {}
prefix_count: int = 0
final_newlines: dict[str, str] = {}
def compile_nonterminal(name: str, rule: parser.NonTerminal):
if name not in visited:
visited.add(name)
for production in rule.fn(grammar).flatten(with_metadata=True):
trans_prod = compile_production(production)
generated_grammar.append((name, trans_prod))
def compile_production(production: parser.FlattenedWithMetadata) -> list[str]:
nonlocal groups
nonlocal indents
nonlocal newlines
nonlocal prefix_count
nonlocal final_newlines
prefix_stack: list[str] = []
result = []
for item in production:
if isinstance(item, str):
nt = nonterminals[item]
if nt.transparent:
# If it's transparent then we make a new set of
# productions that covers the contents of the
# transparent nonterminal.
name = "xxx_" + nt.name
compile_nonterminal(name, nt)
result.append(name)
else:
# Otherwise it's a "token" in our input, named
# "tree_{whatever}".
result.append(f"tree_{item}")
elif isinstance(item, parser.Terminal):
# If it's a terminal it will appear in our input as
# "token_{whatever}".
result.append(f"token_{item.name}")
else:
meta, children = item
tx_children = compile_production(children)
pretty = meta.get("format")
if isinstance(pretty, parser.FormatMeta):
if pretty.group:
# Make a fake rule.
child_key = tuple(tx_children)
rule_name = groups.get(child_key)
if rule_name is None:
rule_name = f"g_{len(groups)}"
groups[child_key] = rule_name
generated_grammar.append((rule_name, tx_children))
tx_children = [rule_name]
if pretty.indent:
child_key = (tuple(tx_children), pretty.indent)
rule_name = indents.get(child_key)
if rule_name is None:
rule_name = f"i_{len(indents)}"
indents[child_key] = rule_name
generated_grammar.append((rule_name, tx_children))
tx_children = [rule_name]
if pretty.newline is not None:
if len(tx_children) == 0:
tx_children = result
result = []
if len(tx_children) > 0:
# n == postfix newline
child_key = (tuple(tx_children), pretty.newline)
rule_name = newlines.get(child_key)
if rule_name is None:
rule_name = f"n_{len(newlines)}"
newlines[child_key] = rule_name
generated_grammar.append((rule_name, tx_children))
tx_children = [rule_name]
else:
# p == prefix newline
rule_name = f"p_{prefix_count}"
prefix_count += 1
final_newlines[rule_name] = pretty.newline
prefix_stack.append(rule_name)
if pretty.forced_break:
if len(tx_children) == 0:
tx_children = result
result = []
if len(tx_children) > 0:
# f == postfix forced break
rule_name = f"f_{prefix_count}"
prefix_count += 1
generated_grammar.append((rule_name, tx_children))
tx_children = [rule_name]
else:
# d == prefix forced break (to the right of 'f' on my kbd)
rule_name = f"d_{prefix_count}"
prefix_count += 1
prefix_stack.append(rule_name)
# If it turned out to have formatting meta then we will
# have replaced or augmented the translated children
# appropriately. Otherwise, if it's highlighting meta or
# something else, we'll have ignored it and the
# translated children should just be inserted inline.
result.extend(tx_children)
# OK so we might have some prefix newlines. They should contain... things.
while len(prefix_stack) > 0:
rule_name = prefix_stack.pop()
generated_grammar.append((rule_name, result))
result = [rule_name]
return result
start_name = f"yyy_{rule.name}"
compile_nonterminal(start_name, rule)
gen = grammar._generator(start_name, generated_grammar)
parse_table = gen.gen_table()
for (_, replacement), rule_name in newlines.items():
final_newlines[rule_name] = replacement
indent_amounts = {rule_name: amount for ((_, amount), rule_name) in indents.items()}
return MatcherTable(
parse_table,
indent_amounts,
final_newlines,
)
@dataclasses.dataclass
class PrettyTable:
"""Information necessary to convert a parsed tree into a wadler-style
pretty document, where it can then be formatted.
This is basically a bunch of "MatcherTables", one for each kind of tree,
that tell us how to recover document structure from the tree node.
"""
indent: str
trivia_modes: dict[str, parser.TriviaMode]
matchers: dict[str, MatcherTable]
def compile_pretty_table(grammar: parser.Grammar, indent: str | None = None) -> PrettyTable:
nonterminals = {nt.name: nt for nt in grammar.non_terminals()}
matchers = {}
if indent is None:
indent = getattr(grammar, "pretty_indent", None)
if indent is None:
indent = " "
trivia_mode = {}
for t in grammar.terminals():
mode = t.meta.get("trivia_mode")
if t.name is not None and isinstance(mode, parser.TriviaMode):
trivia_mode[t.name] = mode
for name, rule in nonterminals.items():
matchers[name] = _compile_nonterminal_matcher(grammar, nonterminals, rule)
return PrettyTable(
indent,
trivia_mode,
matchers,
)
############################################################################
# The Actual Pretty Printer
############################################################################
class Matcher:
table: MatcherTable
trivia_mode: dict[str, parser.TriviaMode] trivia_mode: dict[str, parser.TriviaMode]
def __init__(self, table: MatcherTable, trivia_mode: dict[str, parser.TriviaMode]):
self.table = table
self.trivia_mode = trivia_mode
def match( def match(
self, self,
printer: "Printer", printer: "Printer",
@ -374,7 +604,7 @@ class Matcher:
src: str, src: str,
) -> Document: ) -> Document:
stack: list[tuple[int, Document]] = [(0, None)] stack: list[tuple[int, Document]] = [(0, None)]
table = self.table table = self.table.table
# eof_trivia = [] # eof_trivia = []
# if len(items) > 0: # if len(items) > 0:
@ -420,15 +650,15 @@ class Matcher:
child = group(child) child = group(child)
elif name[0] == "i": elif name[0] == "i":
amount = self.indent_amounts[name] amount = self.table.indent_amounts[name]
child = Indent(amount, child) child = Indent(amount, child)
elif name[0] == "n": elif name[0] == "n":
replace = self.newline_replace[name] replace = self.table.newline_replace[name]
child = cons(child, NewLine(replace)) child = cons(child, NewLine(replace))
elif name[0] == "p": elif name[0] == "p":
replace = self.newline_replace[name] replace = self.table.newline_replace[name]
child = cons(NewLine(replace), child) child = cons(NewLine(replace), child)
elif name[0] == "f": elif name[0] == "f":
@ -464,8 +694,6 @@ class Matcher:
def apply_pre_trivia(self, trivia_tokens: list[runtime.TokenValue], src: str) -> Document: def apply_pre_trivia(self, trivia_tokens: list[runtime.TokenValue], src: str) -> Document:
pre_trivia, _ = slice_pre_post_trivia(self.trivia_mode, trivia_tokens) pre_trivia, _ = slice_pre_post_trivia(self.trivia_mode, trivia_tokens)
# print(f"PRE:\n{pre_trivia}")
if len(pre_trivia) == 0: if len(pre_trivia) == 0:
return None return None
@ -474,7 +702,6 @@ class Matcher:
trivia_doc = None trivia_doc = None
new_line_count = 0 new_line_count = 0
for mode, token in pre_trivia: for mode, token in pre_trivia:
# print(f"PRE {mode:25} {token.kind:30} ({new_line_count})")
match mode: match mode:
case parser.TriviaMode.LineComment: case parser.TriviaMode.LineComment:
trivia_doc = cons( trivia_doc = cons(
@ -509,7 +736,6 @@ class Matcher:
trivia_doc = None trivia_doc = None
for mode, token in post_trivia: for mode, token in post_trivia:
# print(f"POST {mode:25} {token.kind:30}")
match mode: match mode:
case parser.TriviaMode.Blank: case parser.TriviaMode.Blank:
pass pass
@ -573,210 +799,23 @@ class Matcher:
class Printer: class Printer:
# TODO: Pre-generate the matcher tables for a grammar, to make it table: PrettyTable
# possible to do codegen in other languages. matchers: dict[str, Matcher]
grammar: parser.Grammar
_matchers: dict[str, Matcher]
_nonterminals: dict[str, parser.NonTerminal]
_indent: str
_trivia_mode: dict[str, parser.TriviaMode]
def __init__(self, grammar: parser.Grammar, indent: str | None = None): def __init__(self, table: PrettyTable):
self.grammar = grammar self.table = table
self._nonterminals = {nt.name: nt for nt in grammar.non_terminals()} self.matchers = {
self._matchers = {} name: Matcher(value, self.table.trivia_modes) for name, value in table.matchers.items()
}
if indent is None:
indent = getattr(self.grammar, "pretty_indent", None)
if indent is None:
indent = " "
self._indent = indent
trivia_mode = {}
for t in grammar.terminals():
mode = t.meta.get("trivia_mode")
if t.name is not None and isinstance(mode, parser.TriviaMode):
trivia_mode[t.name] = mode
self._trivia_mode = trivia_mode
def indent(self) -> str: def indent(self) -> str:
return self._indent return self.table.indent
def lookup_nonterminal(self, name: str) -> parser.NonTerminal:
return self._nonterminals[name]
def compile_rule(self, rule: parser.NonTerminal) -> Matcher:
generated_grammar: list[typing.Tuple[str, list[str]]] = []
visited: set[str] = set()
# In order to generate groups, indents, and newlines we need to
# synthesize new productions. And it happens sometimes that we get
# duplicates, repeated synthetic productions. It's important to
# de-duplicate productions, otherwise we'll wind up with ambiguities
# in the parser.
#
# These dictionaries track the synthetic rules: the keys are
# production and also the parameter (if any), and the values are the
# names of the productions that produce the effect.
#
groups: dict[tuple[str, ...], str] = {}
indents: dict[tuple[tuple[str, ...], int], str] = {}
newlines: dict[tuple[tuple[str, ...], str], str] = {}
prefix_count: int = 0
final_newlines: dict[str, str] = {}
def compile_nonterminal(name: str, rule: parser.NonTerminal):
if name not in visited:
visited.add(name)
for production in rule.fn(self.grammar).flatten(with_metadata=True):
trans_prod = compile_production(production)
generated_grammar.append((name, trans_prod))
def compile_production(production: parser.FlattenedWithMetadata) -> list[str]:
nonlocal groups
nonlocal indents
nonlocal newlines
nonlocal prefix_count
nonlocal final_newlines
prefix_stack: list[str] = []
result = []
for item in production:
if isinstance(item, str):
nt = self._nonterminals[item]
if nt.transparent:
# If it's transparent then we make a new set of
# productions that covers the contents of the
# transparent nonterminal.
name = "xxx_" + nt.name
compile_nonterminal(name, nt)
result.append(name)
else:
# Otherwise it's a "token" in our input, named
# "tree_{whatever}".
result.append(f"tree_{item}")
elif isinstance(item, parser.Terminal):
# If it's a terminal it will appear in our input as
# "token_{whatever}".
result.append(f"token_{item.name}")
else:
meta, children = item
tx_children = compile_production(children)
pretty = meta.get("format")
if isinstance(pretty, parser.FormatMeta):
if pretty.group:
# Make a fake rule.
child_key = tuple(tx_children)
rule_name = groups.get(child_key)
if rule_name is None:
rule_name = f"g_{len(groups)}"
groups[child_key] = rule_name
generated_grammar.append((rule_name, tx_children))
tx_children = [rule_name]
if pretty.indent:
child_key = (tuple(tx_children), pretty.indent)
rule_name = indents.get(child_key)
if rule_name is None:
rule_name = f"i_{len(indents)}"
indents[child_key] = rule_name
generated_grammar.append((rule_name, tx_children))
tx_children = [rule_name]
if pretty.newline is not None:
if len(tx_children) == 0:
tx_children = result
result = []
if len(tx_children) > 0:
# n == postfix newline
child_key = (tuple(tx_children), pretty.newline)
rule_name = newlines.get(child_key)
if rule_name is None:
rule_name = f"n_{len(newlines)}"
newlines[child_key] = rule_name
generated_grammar.append((rule_name, tx_children))
tx_children = [rule_name]
else:
# p == prefix newline
rule_name = f"p_{prefix_count}"
prefix_count += 1
final_newlines[rule_name] = pretty.newline
prefix_stack.append(rule_name)
if pretty.forced_break:
if len(tx_children) == 0:
tx_children = result
result = []
if len(tx_children) > 0:
# f == postfix forced break
rule_name = f"f_{prefix_count}"
prefix_count += 1
generated_grammar.append((rule_name, tx_children))
tx_children = [rule_name]
else:
# d == prefix forced break (to the right of 'f' on my kbd)
rule_name = f"d_{prefix_count}"
prefix_count += 1
prefix_stack.append(rule_name)
# If it turned out to have formatting meta then we will
# have replaced or augmented the translated children
# appropriately. Otherwise, if it's highlighting meta or
# something else, we'll have ignored it and the
# translated children should just be inserted inline.
result.extend(tx_children)
# OK so we might have some prefix newlines. They should contain... things.
while len(prefix_stack) > 0:
rule_name = prefix_stack.pop()
generated_grammar.append((rule_name, result))
result = [rule_name]
return result
start_name = f"yyy_{rule.name}"
compile_nonterminal(start_name, rule)
gen = self.grammar._generator(start_name, generated_grammar)
parse_table = gen.gen_table()
for (_, replacement), rule_name in newlines.items():
final_newlines[rule_name] = replacement
indent_amounts = {rule_name: amount for ((_, amount), rule_name) in indents.items()}
return Matcher(
parse_table,
indent_amounts,
final_newlines,
self._trivia_mode,
)
def rule_to_matcher(self, rule: parser.NonTerminal) -> Matcher:
result = self._matchers.get(rule.name)
if result is None:
result = self.compile_rule(rule)
self._matchers[rule.name] = result
return result
def convert_tree_to_document(self, tree: runtime.Tree, src: str) -> Document: def convert_tree_to_document(self, tree: runtime.Tree, src: str) -> Document:
name = tree.name name = tree.name
assert name is not None, "Cannot format a tree if it still has transparent nodes inside" assert name is not None, "Cannot format a tree if it still has transparent nodes inside"
rule = self.lookup_nonterminal(name) matcher = self.matchers[name]
matcher = self.rule_to_matcher(rule)
m = matcher.match(self, list(tree.children), src) m = matcher.match(self, list(tree.children), src)
if m is None: if m is None:
raise ValueError( raise ValueError(
@ -786,4 +825,4 @@ class Printer:
def format_tree(self, tree: runtime.Tree, src: str, width: int) -> DocumentLayout: def format_tree(self, tree: runtime.Tree, src: str, width: int) -> DocumentLayout:
doc = self.convert_tree_to_document(tree, src) doc = self.convert_tree_to_document(tree, src)
return layout_document(doc, width, self._indent) return layout_document(doc, width, self.table.indent)

16
tests/test_wadler.py
View file

@ -149,7 +149,7 @@ def test_convert_tree_to_document():
assert [] == errors assert [] == errors
assert tree is not None assert tree is not None
printer = wadler.Printer(JSON) printer = wadler.Printer(wadler.compile_pretty_table(JSON))
doc = flatten_document(printer.convert_tree_to_document(tree, text), text) doc = flatten_document(printer.convert_tree_to_document(tree, text), text)
assert doc == [ assert doc == [
@ -216,7 +216,7 @@ def test_layout_basic():
assert [] == errors assert [] == errors
assert tree is not None assert tree is not None
printer = wadler.Printer(JSON) printer = wadler.Printer(wadler.compile_pretty_table(JSON))
result = printer.format_tree(tree, text, 50).apply_to_source(text) result = printer.format_tree(tree, text, 50).apply_to_source(text)
assert result == _output( assert result == _output(
@ -278,7 +278,7 @@ def test_forced_break():
assert errors == [] assert errors == []
assert tree is not None assert tree is not None
printer = wadler.Printer(g) printer = wadler.Printer(wadler.compile_pretty_table(g))
result = printer.format_tree(tree, text, 200).apply_to_source(text) result = printer.format_tree(tree, text, 200).apply_to_source(text)
assert result == _output( assert result == _output(
@ -318,7 +318,7 @@ def test_maintaining_line_breaks():
assert errors == [] assert errors == []
assert tree is not None assert tree is not None
printer = wadler.Printer(g) printer = wadler.Printer(wadler.compile_pretty_table(g))
result = printer.format_tree(tree, text, 200).apply_to_source(text) result = printer.format_tree(tree, text, 200).apply_to_source(text)
assert result == _output( assert result == _output(
@ -352,7 +352,7 @@ def test_trailing_trivia():
assert errors == [] assert errors == []
assert tree is not None assert tree is not None
printer = wadler.Printer(g) printer = wadler.Printer(wadler.compile_pretty_table(g))
result = printer.format_tree(tree, text, 200).apply_to_source(text) result = printer.format_tree(tree, text, 200).apply_to_source(text)
assert result == _output( assert result == _output(
@ -378,7 +378,7 @@ def test_trailing_trivia_two():
assert errors == [] assert errors == []
assert tree is not None assert tree is not None
printer = wadler.Printer(g) printer = wadler.Printer(wadler.compile_pretty_table(g))
result = printer.format_tree(tree, text, 200).apply_to_source(text) result = printer.format_tree(tree, text, 200).apply_to_source(text)
assert result == _output( assert result == _output(
@ -432,9 +432,7 @@ def test_trailing_trivia_split():
print(f"{mode:25} {t.kind:10} {repr(text[t.start:t.end])}") print(f"{mode:25} {t.kind:10} {repr(text[t.start:t.end])}")
trivia_doc = wadler.Matcher( trivia_doc = wadler.Matcher(
ParseTable([], [], set()), wadler.MatcherTable(ParseTable([], [], set()), {}, {}),
{},
{},
TRIVIA_MODES, TRIVIA_MODES,
).apply_post_trivia( ).apply_post_trivia(
token.post_trivia, token.post_trivia,