This commit is contained in:
Michael Zhang 2021-06-13 18:28:28 -05:00
parent ac37c8d491
commit 8c687c4d1c
Signed by: michael
GPG key ID: BDA47A31A3C8EE6B
9 changed files with 213 additions and 122 deletions

View file

@ -1,2 +1,2 @@
watch:
watchexec -ce py,lark,ag -i gen 'mypy *.py && python gen.py && mypy gen/*.py'
watchexec -ce py,lark,ag -i gen 'mypy *.py && python agmain.py && mypy gen/*.py'

View file

@ -3,28 +3,50 @@ from lark import Transformer, Tree, Token
T = TypeVar("T")
class Ast:
# def __new__(cls: Type[Ast], name: str, bases: Tuple[type], namespace: Dict[str, Any]) -> Ast:
# x = super().__new__(cls, name, bases, namespace)
# x.id = cls.__gen()
# return x
id: str
n = 0
@classmethod
def __gen(cls, name: str = "") -> str:
newid = cls.n
cls.n += 1
return f"_a{newid}{name}"
def __init__(self) -> None:
self.id = self.__gen()
class Decl:
name: str
class Iface(Decl):
def __init__(self, name: str):
self.name = name
class IfaceField: pass
class IfaceRef: pass
class Expr: pass
class IfaceRef(str): pass
class IfaceField:
def __init__(self, name: str, ty: str):
self.name = name
self.ty = ty
class Iface(Decl):
def __init__(self, name: str, fields: List[IfaceField]):
self.name = name
self.fields = fields
class Expr(Ast):
def __init__(self) -> None:
super().__init__()
class NodeRef: pass
class NodeRefByName(NodeRef):
class NodeRefByName(NodeRef, str):
def __init__(self, name: str):
self.name = name
def __repr__(self) -> str: return f"NodeRefByName({self.name})"
class Sym: pass
class SymRename(Sym):
def __init__(self, name: str, node_ref: NodeRef):
def __init__(self, name: str, ty: NodeRef):
self.name = name
self.node_ref = node_ref
def __repr__(self) -> str: return f"SymRename({self.name} : {self.node_ref})"
self.ty = ty
def __repr__(self) -> str: return f"SymRename({self.name} : {self.ty})"
class Equation:
def __init__(self, lhs: Expr, rhs: Expr):
self.lhs = lhs
@ -45,26 +67,31 @@ class Node(Decl):
class ExprDot(Expr):
def __init__(self, left: Expr, right: str):
super().__init__()
self.left = left
self.right = right
def __repr__(self) -> str: return f"{self.left}.{self.right}"
class ExprAdd(Expr):
def __init__(self, left: Expr, right: Expr):
super().__init__()
self.left = left
self.right = right
def __repr__(self) -> str: return f"{self.left} + {self.right}"
class ExprMul(Expr):
def __init__(self, left: Expr, right: Expr):
super().__init__()
self.left = left
self.right = right
def __repr__(self) -> str: return f"{self.left} * {self.right}"
class ExprCall(Expr):
def __init__(self, func: Expr, args: List[Expr]):
super().__init__()
self.func = func
self.args = args
def __repr__(self) -> str: return f"{self.func}({self.args})"
class ExprName(Expr):
def __init__(self, name: str):
super().__init__()
self.name = name
def __repr__(self) -> str: return f"{self.name}"
@ -74,10 +101,10 @@ class Parser(Transformer[List[Decl]]):
# interfaces
def iface(self, items: List[Any]) -> Iface:
[name, fields] = items
return Iface(name)
return Iface(name, fields)
def iface_field(self, items: List[str]) -> IfaceField:
[name, ty] = items
return IfaceField()
return IfaceField(name, ty)
def iface_ref(self, items: List[str]) -> str: return items[0]
def iface_refs(self, items: List[IfaceRef]) -> List[IfaceRef]: return items
@ -98,7 +125,7 @@ class Parser(Transformer[List[Decl]]):
# equations
def equations(self, items: List[Equation]) -> List[Equation]: return items
def equation_(self, items: List[Equation]) -> Equation: return items[0]
def equation_semi(self, items: List[Equation]) -> Equation: return items[0]
def equation(self, items: List[Expr]) -> Equation: return Equation(items[0], items[1])
# expr

140
aggen.py Normal file
View file

@ -0,0 +1,140 @@
from typing import *
import textwrap
import re
import copy
from agast import *
global i
i = 0
class GenResult:
def __init__(self, pd: str = "", ex: str = ""):
self.parser_data = pd
self.extra = ex
def gen(program: List[Decl]) -> GenResult:
res = GenResult()
def gen(prefix: str = "", suffix: str = "") -> str:
global i
presan = re.sub("[^0-9a-zA-Z]+", "_", prefix)
sufsan = re.sub("[^0-9a-zA-Z]+", "_", suffix)
i += 1
return f"{presan}{i}{sufsan}"
def v(name: str) -> str:
return f"__ag_{name}"
# builtins
builtins: Dict[str, str] = {
"parseInt": "",
}
# collect a list of name -> iface declarations
ifaces: Dict[str, Iface] = dict(
map(lambda c: (c.name, cast(Iface, c)),
filter(lambda c: isinstance(c, Iface),
program)))
# list of node -> iface mappings
what_ifaces: Dict[str, Set[str]] = dict()
what_fields: Dict[str, Dict[str, str]] = dict()
for node in filter(lambda c: isinstance(c, Node), program):
node = cast(Node, node)
# all_fields = dict()
what_ifaces[node.name] = set(node.ifaces)
this_fields = dict()
for iface in node.ifaces:
fields = ifaces[iface].fields
for field in fields:
if field.name in this_fields:
raise Exception("duplicate field name")
this_fields[field.name] = field.ty
what_fields[node.name] = this_fields
print("what_ifaces:", what_ifaces)
print("what_fields:", what_fields)
# a high-level dictionary of productions; this has sub-productions
# that should be further expanded at a later step before converting
# into lark code
productions_hi: Dict[str, Union[str, List[str]]] = dict()
# TODO: this should probably not be inlined here, but i'll move it
# out once i get more info into the 'env'
def collect_required_thunks(env: List[Tuple[str, NodeRef]], expr: Expr) -> Dict[str, str]:
names = dict(env)
print(f"collect_required_thunks({expr})", expr.__class__)
if isinstance(expr, ExprDot):
return collect_required_thunks(env, expr.left)
elif isinstance(expr, ExprMul):
a = collect_required_thunks(env, expr.left)
b = collect_required_thunks(env, expr.right)
a.update(b)
return a
elif isinstance(expr, ExprAdd):
a = collect_required_thunks(env, expr.left)
b = collect_required_thunks(env, expr.right)
a.update(b)
return a
elif isinstance(expr, ExprCall):
return collect_required_thunks(env, expr.func)
elif isinstance(expr, ExprName):
if expr.name not in names and expr.name not in builtins:
raise Exception(f"unbound name '{expr.name}'")
return dict()
raise Exception(f"unhandled {expr.__class__}")
for node in filter(lambda c: isinstance(c, Node), program):
node = cast(Node, node)
n_class_name = gen(node.name)
class_decl = textwrap.dedent(f"""
class {v(n_class_name)}: pass
""")
res.extra += class_decl
print(node.name, node.ifaces)
for variant in node.variants:
v_class_name = gen(f"{n_class_name}_var")
class_decl = textwrap.dedent(f"""
class {v(v_class_name)}({v(n_class_name)}):
''' '''
pass
""")
res.extra += class_decl
prod_name = gen(node.name)
print(prod_name)
# create an environment for checking the equations based on
# the production
env: List[Tuple[str, NodeRef]] = list()
for sym in variant.prod:
if isinstance(sym, SymRename):
env.append((sym.name, sym.ty))
print(env)
# for each of the equations, find out what the equation is
# trying to compute, and generate a thunk corresponding to
# that value.
for eq in variant.equations:
eq_name = gen(f"eq_{node.name}")
thunk_name = gen(f"thunk_{node.name}")
print("RHS", eq.rhs, eq.rhs.id)
collect_required_thunks(copy.deepcopy(env), eq.rhs)
func_impl = textwrap.dedent(f"""
def {eq_name}() -> None:
''' {repr(eq)} '''
pass
def {thunk_name}() -> Thunk[None]:
return Thunk({eq_name})
""")
print(f"```py\n{func_impl}\n```")
res.extra += func_impl
# this is a "type alias" that connects it to one of the generated
# names above
res.extra += f"{node.name} = {v(n_class_name)}"
return res

View file

@ -4,7 +4,7 @@ import importlib
from lark import Lark
from agast import *
from agtypeck import *
from aggen import *
p = Lark(open("grammar.lark").read(), start="program", parser="lalr")
@ -13,22 +13,24 @@ if __name__ == "__main__":
data = f.read()
cst = p.parse(data)
# print("cst", cst)
trans = Parser()
ast = trans.transform(cst)
print("ast", ast)
res = typecheck(ast)
res = gen(ast)
if not os.path.exists("gen"):
os.makedirs("gen")
with open("gen/arith.py", "w") as f:
fmt_str = textwrap.dedent("""
__all__ = ["parse"]
from typing import Generic, TypeVar, Optional, Callable
from typing import Generic, TypeVar, Optional, Callable, Dict, Any
from lark import Lark, Transformer
T = TypeVar('T')
builtins: Dict[str, Any] = {{
"parseInt": lambda s: int(s)
}}
class Thunk(Generic[T]):
''' A thunk represents a value that may be computed lazily. '''
value: Optional[T]

View file

@ -1,85 +0,0 @@
from typing import *
import textwrap
import re
from agast import *
global i
i = 0
class TypecheckResult:
def __init__(self, pd: str = "", ex: str = ""):
self.parser_data = pd
self.extra = ex
def typecheck(program: List[Decl]) -> TypecheckResult:
res = TypecheckResult()
def gen(prefix: str = "", suffix: str = "") -> str:
global i
presan = re.sub("[^0-9a-zA-Z]+", "_", prefix)
sufsan = re.sub("[^0-9a-zA-Z]+", "_", suffix)
i += 1
return f"{presan}{i}{sufsan}"
def v(name: str) -> str:
return f"__ag_{name}"
# collect a list of name -> iface declarations
ifaces: Dict[str, Decl] = dict(
map(lambda c: (c.name, c),
filter(lambda c: isinstance(c, Iface),
program)))
print(ifaces)
# a high-level dictionary of productions; this has sub-productions
# that should be further expanded at a later step before converting
# into lark code
productions_hi: Dict[str, Union[str, List[str]]] = dict()
for node in filter(lambda c: isinstance(c, Node), program):
node = cast(Node, node)
n_class_name = gen(node.name)
class_decl = textwrap.dedent(f"""
class {v(n_class_name)}: pass
""")
res.extra += class_decl
print(node.name, node.ifaces)
for variant in node.variants:
v_class_name = gen(f"{n_class_name}_var")
class_decl = textwrap.dedent(f"""
class {v(v_class_name)}({v(n_class_name)}):
''' '''
pass
""")
res.extra += class_decl
prod_name = gen(node.name)
print(prod_name)
for sym in variant.prod:
print("sym", sym)
# for each of the equations, find out what the equation is
# trying to compute, and generate a thunk corresponding to
# that value.
for eq in variant.equations:
print("--eq", eq)
print(eq.lhs)
eq_name = gen(f"eq_{node.name}")
thunk_name = gen(f"thunk_{node.name}")
func_impl = textwrap.dedent(f"""
def {eq_name}() -> None:
''' {repr(eq)} '''
pass
def {thunk_name}() -> Thunk[None]:
return Thunk({eq_name})
""")
print(f"```py\n{func_impl}\n```")
res.extra += func_impl
# this is a "type alias" that connects it to one of the generated
# names above
res.extra += f"{node.name} = {v(n_class_name)}"
return res

View file

@ -4,7 +4,7 @@ iface HasValue {
node Expr : HasValue {
<l:Expr> "+" <r:Expr> => {
self.val = l.val + r.val;
self.val = l.val + r.val * l.val;
}
<l:Expr> "*" <r:Expr> => {
self.val = l.val * r.val;

View file

@ -1,8 +1,8 @@
program: decl*
?decl: iface
| node
| func
| node
| func
sep_trail{item, punc}: item (punc item)? punc?
@ -17,28 +17,35 @@ node: "node" ident ":" iface_refs "{" variants "}"
variants: variant*
variant: prod "=>" "{" equations "}"
prod: sym*
sym: sym_rename
| STRING
?sym: sym_rename
| STRING
sym_rename: "<" ident ":" node_ref ">"
node_ref: node_ref_name
| STRING
?node_ref: node_ref_name
| STRING
node_ref_name: ident
equations: equation_*
equation_: equation ";"
equations: equation_semi*
equation_semi: equation ";"
// TODO: the left side should really be a separate type
// called lvalue, and should NOT include literals
equation: expr "=" expr
?expr: expr_dot
| expr_add
| expr_mul
| expr_call
| expr_name
expr_dot: expr "." expr
expr_add: expr "+" expr
expr_mul: expr "*" expr
expr_call: expr "(" args ")"
// Expressions
?expr: expr2
| expr_add
expr_add: expr "+" expr2
?expr2: expr3
| expr_mul
| expr_call
expr_mul: expr2 "*" expr3
expr_call: expr2 "(" args ")"
?expr3: "(" expr ")"
| expr_dot
| expr_name
expr_dot: expr3 "." ident
expr_name: ident
args: sep_trail{expr, ","}
ty: ident

View file

View file

@ -12,5 +12,5 @@ function Invoke-Call {
}
Invoke-Call -ScriptBlock {mypy (get-item *.py) } -ErrorAction Stop
Invoke-Call -ScriptBlock {python gen.py } -ErrorAction Stop
Invoke-Call -ScriptBlock {python agmain.py } -ErrorAction Stop
Invoke-Call -ScriptBlock {mypy (get-item gen/*.py) } -ErrorAction Stop