1982 lines
79 KiB
Nim
1982 lines
79 KiB
Nim
# Copyright 2022 Mattia Giambirtone & All Contributors
|
|
#
|
|
# Licensed under the Apache License, Version 2.0 (the "License");
|
|
# you may not use this file except in compliance with the License.
|
|
# You may obtain a copy of the License at
|
|
#
|
|
# http://www.apache.org/licenses/LICENSE-2.0
|
|
#
|
|
# Unless required by applicable law or agreed to in writing, software
|
|
# distributed under the License is distributed on an "AS IS" BASIS,
|
|
# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
|
|
# See the License for the specific language governing permissions and
|
|
# limitations under the License.
|
|
import meta/token
|
|
import meta/ast
|
|
import meta/errors
|
|
import ../util/multibyte
|
|
import ../util/symbols
|
|
import lexer as l
|
|
import parser as p
|
|
|
|
import std/tables
|
|
import std/strformat
|
|
import std/algorithm
|
|
import std/parseutils
|
|
import std/strutils
|
|
import std/sequtils
|
|
import std/sets
|
|
import std/os
|
|
|
|
|
|
export ast
|
|
export token
|
|
export multibyte
|
|
|
|
|
|
type
|
|
TypeKind = enum
|
|
## An enumeration of compile-time
|
|
## types
|
|
Int8, UInt8, Int16, UInt16, Int32,
|
|
UInt32, Int64, UInt64, Float32, Float64,
|
|
Char, Byte, String, Function, CustomType,
|
|
Nil, Nan, Bool, Inf, Typevar, Generic,
|
|
Reference, Pointer
|
|
Any # Any is used internally in a few cases,
|
|
# for example when looking for operators
|
|
# when only the type of the arguments is of
|
|
# interest
|
|
Type = ref object
|
|
## A wrapper around
|
|
## compile-time types
|
|
mutable: bool
|
|
case kind: TypeKind:
|
|
of Function:
|
|
name: string
|
|
isLambda: bool
|
|
isGenerator: bool
|
|
isCoroutine: bool
|
|
args: seq[tuple[name: string, kind: Type]]
|
|
returnType: Type
|
|
isBuiltinFunction: bool
|
|
builtinOp: string
|
|
fun: FunDecl
|
|
isClosure: bool
|
|
closureBounds: tuple[start, stop: int]
|
|
childFunc: Type
|
|
of Reference, Pointer:
|
|
value: Type
|
|
of Generic:
|
|
node: IdentExpr
|
|
else:
|
|
discard
|
|
|
|
# This way we don't have recursive dependency issues
|
|
import meta/bytecode
|
|
export bytecode
|
|
|
|
|
|
type
|
|
Name = ref object
|
|
## A compile-time wrapper around
|
|
## statically resolved names
|
|
|
|
# Name of the identifier
|
|
name: IdentExpr
|
|
# Owner of the identifier (module)
|
|
owner: string
|
|
# Scope depth
|
|
depth: int
|
|
# Is this name private?
|
|
isPrivate: bool
|
|
# Is this a constant?
|
|
isConst: bool
|
|
# Can this name's value be mutated?
|
|
isLet: bool
|
|
# The name's type
|
|
valueType: Type
|
|
# For functions, this marks where the function's
|
|
# code begins. For variables, this stores where
|
|
# their StoreVar/StoreHeap instruction was emitted
|
|
codePos: int
|
|
# Is the name closed over (i.e. used in a closure)?
|
|
isClosedOver: bool
|
|
# The function that owns this variable (may be nil!)
|
|
belongsTo: Name
|
|
# Is this a function argument?
|
|
isFunctionArgument: bool
|
|
# Where is this node declared in the file?
|
|
line: int
|
|
# Is this a function declaration or a variable
|
|
# with a function as value? (The distinction *is*
|
|
# important! Check emitFunction())
|
|
isFunDecl: bool
|
|
Loop = object
|
|
## A "loop object" used
|
|
## by the compiler to emit
|
|
## appropriate jump offsets
|
|
## for continue and break
|
|
## statements
|
|
|
|
# Position in the bytecode where the loop starts
|
|
start: int
|
|
# Scope depth where the loop is located
|
|
depth: int
|
|
# Absolute jump offsets into our bytecode that we need to
|
|
# patch. Used for break statements
|
|
breakPos: seq[int]
|
|
|
|
Compiler* = ref object
|
|
## A wrapper around the Peon compiler's state
|
|
|
|
# The bytecode chunk where we write code to
|
|
chunk: Chunk
|
|
# The output of our parser (AST)
|
|
ast: seq[Declaration]
|
|
# The current AST node we're looking at
|
|
current: int
|
|
# The current file being compiled (used only for
|
|
# error reporting)
|
|
file: string
|
|
# Compile-time "simulation" of the stack at
|
|
# runtime to load variables that have stack
|
|
# behavior more efficiently
|
|
names: seq[Name]
|
|
# The current scope depth. If > 0, we're
|
|
# in a local scope, otherwise it's global
|
|
scopeDepth: int
|
|
# The current function being compiled
|
|
currentFunction: Name
|
|
# Are optimizations turned on?
|
|
enableOptimizations: bool
|
|
# The current loop being compiled (used to
|
|
# keep track of where to jump)
|
|
currentLoop: Loop
|
|
# Are we in REPL mode? If so, Pop instructions
|
|
# for expression statements at the top level are
|
|
# swapped for a special PopRepl instruction that
|
|
# prints the result of the expression once it is
|
|
# evaluated
|
|
replMode: bool
|
|
# The current module being compiled
|
|
# (used to restrict access to statically
|
|
# defined variables at compile time)
|
|
currentModule: string
|
|
# Each time a defer statement is
|
|
# compiled, its code is emitted
|
|
# here. Later, if there is any code
|
|
# to defer in the current function,
|
|
# funDecl will wrap the function's code
|
|
# inside an implicit try/finally block
|
|
# and add this code in the finally branch.
|
|
# This sequence is emptied each time a
|
|
# function declaration is compiled and stores only
|
|
# deferred code for the current function (may
|
|
# be empty)
|
|
deferred: seq[uint8]
|
|
# List of closed-over variables
|
|
closedOver: seq[tuple[name: Name, count: int]]
|
|
# Keeps track of stack frames
|
|
frames: seq[int]
|
|
# Compiler procedures called by pragmas
|
|
compilerProcs: TableRef[string, proc (self: Compiler, pragma: Pragma, node: ASTNode)]
|
|
# Stores line data
|
|
lines: seq[tuple[start, stop: int]]
|
|
# The source of the current module
|
|
source: string
|
|
# Currently imported modules
|
|
modules: HashSet[string]
|
|
# TODO
|
|
scopes: seq[Type]
|
|
CompileError* = ref object of PeonException
|
|
compiler*: Compiler
|
|
node*: ASTNode
|
|
file*: string
|
|
module*: string
|
|
|
|
|
|
## Forward declarations
|
|
proc compile*(self: Compiler, ast: seq[Declaration], file: string, lines: seq[tuple[start, stop: int]], source: string, chunk: Chunk = nil,
|
|
terminateScope: bool = true, incremental: bool = false): Chunk
|
|
proc expression(self: Compiler, node: Expression)
|
|
proc statement(self: Compiler, node: Statement)
|
|
proc declaration(self: Compiler, node: Declaration)
|
|
proc peek(self: Compiler, distance: int = 0): ASTNode
|
|
proc identifier(self: Compiler, node: IdentExpr)
|
|
proc varDecl(self: Compiler, node: VarDecl)
|
|
proc inferType(self: Compiler, node: LiteralExpr): Type
|
|
proc inferType(self: Compiler, node: Expression): Type
|
|
proc findByName(self: Compiler, name: string): seq[Name]
|
|
proc findByType(self: Compiler, name: string, kind: Type, depth: int = -1): seq[Name]
|
|
proc compareTypes(self: Compiler, a, b: Type): bool
|
|
proc patchReturnAddress(self: Compiler, pos: int)
|
|
proc handleMagicPragma(self: Compiler, pragma: Pragma, node: ASTnode)
|
|
proc handlePurePragma(self: Compiler, pragma: Pragma, node: ASTnode)
|
|
proc dispatchPragmas(self: Compiler, node: ASTnode)
|
|
proc funDecl(self: Compiler, node: FunDecl, fn: Name = nil, args: seq[Expression] = @[])
|
|
## End of forward declarations
|
|
|
|
|
|
proc newCompiler*(enableOptimizations: bool = true, replMode: bool = false): Compiler =
|
|
## Initializes a new Compiler object
|
|
new(result)
|
|
result.ast = @[]
|
|
result.current = 0
|
|
result.file = ""
|
|
result.names = @[]
|
|
result.scopeDepth = 0
|
|
result.frames = @[0]
|
|
result.lines = @[]
|
|
result.currentFunction = nil
|
|
result.enableOptimizations = enableOptimizations
|
|
result.replMode = replMode
|
|
result.currentModule = ""
|
|
result.compilerProcs = newTable[string, proc (self: Compiler, pragma: Pragma, node: ASTNode)]()
|
|
result.compilerProcs["magic"] = handleMagicPragma
|
|
result.compilerProcs["pure"] = handlePurePragma
|
|
result.source = ""
|
|
|
|
proc compileModule(self: Compiler, filename: string)
|
|
|
|
|
|
## Public getter for nicer error formatting
|
|
proc getCurrentNode*(self: Compiler): ASTNode = (if self.current >=
|
|
self.ast.len(): self.ast[^1] else: self.ast[self.current - 1])
|
|
proc getCurrentFunction*(self: Compiler): Declaration {.inline.} = (if self.currentFunction.valueType.isNil(): nil else: self.currentFunction.valueType.fun)
|
|
proc getFile*(self: Compiler): string {.inline.} = self.file
|
|
proc getModule*(self: Compiler): string {.inline.} = self.currentModule
|
|
proc getLines*(self: Compiler): seq[tuple[start, stop: int]] = self.lines
|
|
proc getSource*(self: Compiler): string = self.source
|
|
proc getRelPos*(self: Compiler, line: int): tuple[start, stop: int] = self.lines[line - 1]
|
|
|
|
## Utility functions
|
|
|
|
proc peek(self: Compiler, distance: int = 0): ASTNode =
|
|
## Peeks at the AST node at the given distance.
|
|
## If the distance is out of bounds, the last
|
|
## AST node in the tree is returned. A negative
|
|
## distance may be used to retrieve previously
|
|
## consumed AST nodes
|
|
if self.ast.high() == -1 or self.current + distance > self.ast.high() or self.current + distance < 0:
|
|
result = self.ast[^1]
|
|
else:
|
|
result = self.ast[self.current + distance]
|
|
|
|
|
|
proc done(self: Compiler): bool {.inline.} =
|
|
## Returns true if the compiler is done
|
|
## compiling, false otherwise
|
|
result = self.current > self.ast.high()
|
|
|
|
|
|
proc error(self: Compiler, message: string, node: ASTNode = nil) {.raises: [CompileError], inline.} =
|
|
## Raises a CompileError exception
|
|
raise CompileError(msg: message, node: if node.isNil(): self.getCurrentNode() else: node, file: self.file, module: self.currentModule, compiler: self)
|
|
|
|
|
|
proc step(self: Compiler): ASTNode {.inline.} =
|
|
## Steps to the next node and returns
|
|
## the consumed one
|
|
result = self.peek()
|
|
if not self.done():
|
|
self.current += 1
|
|
|
|
|
|
proc emitByte(self: Compiler, byt: OpCode | uint8) {.inline.} =
|
|
## Emits a single byte, writing it to
|
|
## the current chunk being compiled
|
|
self.chunk.write(uint8 byt, self.peek().token.line)
|
|
|
|
|
|
proc emitBytes(self: Compiler, bytarr: openarray[OpCode | uint8]) {.inline.} =
|
|
## Handy helper method to write arbitrary bytes into
|
|
## the current chunk, calling emitByte on each of its
|
|
## elements
|
|
for b in bytarr:
|
|
self.emitByte(b)
|
|
|
|
|
|
proc makeConstant(self: Compiler, val: Expression, typ: Type): array[3, uint8] =
|
|
## Adds a constant to the current chunk's constant table
|
|
## and returns its index as a 3-byte array of uint8s
|
|
var lit: string
|
|
if typ.kind in [UInt8, Int8, Int16, UInt16, Int32, UInt32, Int64, UInt64]:
|
|
lit = val.token.lexeme
|
|
if "'" in lit:
|
|
var idx = lit.high()
|
|
while lit[idx] != '\'':
|
|
lit = lit[0..^2]
|
|
dec(idx)
|
|
lit = lit[0..^2]
|
|
case typ.kind:
|
|
of UInt8, Int8:
|
|
result = self.chunk.writeConstant([uint8(parseInt(lit))])
|
|
of Int16, UInt16:
|
|
result = self.chunk.writeConstant(parseInt(lit).toDouble())
|
|
of Int32, UInt32:
|
|
result = self.chunk.writeConstant(parseInt(lit).toQuad())
|
|
of Int64:
|
|
result = self.chunk.writeConstant(parseInt(lit).toLong())
|
|
of UInt64:
|
|
result = self.chunk.writeConstant(parseBiggestUInt(lit).toLong())
|
|
of String:
|
|
result = self.chunk.writeConstant(val.token.lexeme[1..^1].toBytes())
|
|
of Float32:
|
|
var f: float = 0.0
|
|
discard parseFloat(val.token.lexeme, f)
|
|
result = self.chunk.writeConstant(cast[array[4, uint8]](float32(f)))
|
|
of Float64:
|
|
var f: float = 0.0
|
|
discard parseFloat(val.token.lexeme, f)
|
|
result = self.chunk.writeConstant(cast[array[8, uint8]](f))
|
|
else:
|
|
discard
|
|
|
|
|
|
proc emitConstant(self: Compiler, obj: Expression, kind: Type) =
|
|
## Emits a constant instruction along
|
|
## with its operand
|
|
case kind.kind:
|
|
of Int64:
|
|
self.emitByte(LoadInt64)
|
|
of UInt64:
|
|
self.emitByte(LoadUInt64)
|
|
of Int32:
|
|
self.emitByte(LoadInt32)
|
|
of UInt32:
|
|
self.emitByte(LoadUInt32)
|
|
of Int16:
|
|
self.emitByte(LoadInt16)
|
|
of UInt16:
|
|
self.emitByte(LoadUInt16)
|
|
of Int8:
|
|
self.emitByte(LoadInt8)
|
|
of UInt8:
|
|
self.emitByte(LoadUInt8)
|
|
of String:
|
|
self.emitByte(LoadString)
|
|
let str = LiteralExpr(obj).literal.lexeme
|
|
if str.len() >= 16777216:
|
|
self.error("string constants cannot be larger than 16777215 bytes")
|
|
self.emitBytes((str.len() - 2).toTriple())
|
|
of Float32:
|
|
self.emitByte(LoadFloat32)
|
|
of Float64:
|
|
self.emitByte(LoadFloat64)
|
|
else:
|
|
discard # TODO
|
|
self.emitBytes(self.makeConstant(obj, kind))
|
|
|
|
|
|
proc emitJump(self: Compiler, opcode: OpCode): int =
|
|
## Emits a dummy jump offset to be patched later
|
|
## and returns the absolute index into the chunk's
|
|
## bytecode array where the given placeholder
|
|
## instruction was written
|
|
self.emitByte(opcode)
|
|
self.emitBytes(0.toTriple())
|
|
result = self.chunk.code.len() - 4
|
|
|
|
|
|
proc patchJump(self: Compiler, offset: int) =
|
|
## Patches a previously emitted relative
|
|
## jump using emitJump
|
|
var jump: int = self.chunk.code.len() - offset
|
|
if jump > 16777215:
|
|
self.error("cannot jump more than 16777215 instructions")
|
|
let offsetArray = (jump - 4).toTriple()
|
|
self.chunk.code[offset + 1] = offsetArray[0]
|
|
self.chunk.code[offset + 2] = offsetArray[1]
|
|
self.chunk.code[offset + 3] = offsetArray[2]
|
|
|
|
|
|
proc resolve(self: Compiler, name: IdentExpr,
|
|
depth: int = self.scopeDepth): Name =
|
|
## Traverses self.names backwards and returns the
|
|
## first name object with the given name. Returns
|
|
## nil when the name can't be found. This function
|
|
## has no concept of scope depth, because getStackPos
|
|
## does that job. Note that private names declared in
|
|
## other modules will not be resolved!
|
|
for obj in reversed(self.names):
|
|
if obj.name.token.lexeme == name.token.lexeme:
|
|
if obj.isPrivate and obj.owner != self.currentModule:
|
|
continue # There may be a name in the current module that
|
|
# matches, so we skip this
|
|
return obj
|
|
return nil
|
|
|
|
|
|
proc getStackPos(self: Compiler, name: Name): int =
|
|
## Returns the predicted call stack position of a given name, relative
|
|
## to the current frame
|
|
var found = false
|
|
result = 2
|
|
for variable in self.names:
|
|
if variable.isFunDecl or variable.valueType.kind in {CustomType, Generic}:
|
|
continue
|
|
if name == variable:
|
|
found = true
|
|
break
|
|
inc(result)
|
|
if not found:
|
|
return -1
|
|
|
|
|
|
proc getClosurePos(self: Compiler, name: Name): int =
|
|
## Iterates the internal list of declared closure names backwards and
|
|
## returns the predicted closure array position of a given name.
|
|
## Returns -1 if the name can't be found (this includes names that
|
|
## are private in other modules)
|
|
result = self.closedOver.high()
|
|
var found = false
|
|
for variable in reversed(self.closedOver):
|
|
if name == variable.name:
|
|
found = true
|
|
break
|
|
dec(result)
|
|
if not found:
|
|
return -1
|
|
|
|
|
|
proc resolve(self: Compiler, name: string,
|
|
depth: int = self.scopeDepth): Name =
|
|
## Traverses self.names backwards and returns the
|
|
## first name object with the given name. Returns
|
|
## nil when the name can't be found. This function
|
|
## has no concept of scope depth, because getStackPos
|
|
## does that job. Note that private names declared in
|
|
## other modules will not be resolved!
|
|
for obj in reversed(self.names):
|
|
if obj.name.token.lexeme == name:
|
|
if obj.isPrivate and obj.owner != self.currentModule:
|
|
continue # There may be a name in the current module that
|
|
# matches, so we skip this
|
|
return obj
|
|
return nil
|
|
|
|
|
|
proc detectClosureVariable(self: Compiler, name: var Name, depth: int = self.scopeDepth) =
|
|
## Detects if the given name is used in a local scope deeper
|
|
## than the given one and modifies the code emitted for it
|
|
## to store it as a closure variable if it is. Does nothing if the name
|
|
## hasn't been declared yet or is unreachable (for example if it's
|
|
## declared as private in another module). This function must be called
|
|
## each time a name is referenced in order for closed-over variables
|
|
## to be emitted properly, otherwise the runtime may behave
|
|
## unpredictably or crash
|
|
if name.isNil() or name.depth == 0 or name.isClosedOver:
|
|
return
|
|
elif name.depth < depth and self.scopes[name.depth - 1] != self.scopes[depth - 1]:
|
|
# Ding! The given name is closed over in another function:
|
|
# we need to change the Jump instruction that self.declareName
|
|
# put in place for us into a StoreClosure. We also update
|
|
# the name's isClosedOver field so that self.identifier()
|
|
# can emit a LoadClosure instruction instead of a LoadVar
|
|
# once this name is referenced in the future
|
|
self.closedOver.add((name, 0))
|
|
name.isClosedOver = true
|
|
if not self.currentFunction.valueType.isClosure:
|
|
self.currentFunction.valueType.isClosure = true
|
|
self.currentFunction.valueType.closureBounds.start = self.closedOver.high()
|
|
self.currentFunction.valueType.closureBounds.stop = self.closedOver.high()
|
|
if self.closedOver.len() >= 16777216:
|
|
self.error("too many consecutive closed-over variables (max is 16777215)")
|
|
if not name.isFunctionArgument:
|
|
self.chunk.code[name.codePos] = StoreClosure.uint8()
|
|
for i, b in self.closedOver.high().toTriple():
|
|
self.chunk.code[name.codePos + i + 1] = b
|
|
else:
|
|
self.chunk.code[name.codePos] = LiftArgument.uint8()
|
|
for i, b in self.getStackPos(name).toTriple():
|
|
self.chunk.code[name.codePos + i + 1] = b
|
|
|
|
|
|
proc compareTypes(self: Compiler, a, b: Type): bool =
|
|
## Compares two type objects
|
|
## for equality (works with nil!)
|
|
# The nil code here is for void functions (when
|
|
# we compare their return types)
|
|
if a.isNil():
|
|
return b.isNil() or b.kind == Any
|
|
elif b.isNil():
|
|
return a.isNil() or a.kind == Any
|
|
elif a.kind == Any or b.kind == Any:
|
|
# This is needed internally: user code
|
|
# cannot generate code for matching
|
|
# arbitrary types, but we need it for
|
|
# function calls and stuff like that
|
|
# since peon doesn't have return type
|
|
# inference
|
|
return true
|
|
elif a.kind == Generic or b.kind == Generic:
|
|
# Matching generic argument types
|
|
return true
|
|
elif a.kind != b.kind:
|
|
# Next, we see the type discriminant:
|
|
# If they're different, then they can't
|
|
# be the same type!
|
|
return false
|
|
case a.kind:
|
|
# If all previous checks pass, it's time
|
|
# to go through each possible type peon
|
|
# supports and compare it
|
|
of Int8, UInt8, Int16, UInt16, Int32,
|
|
UInt32, Int64, UInt64, Float32, Float64,
|
|
Char, Byte, String, Nil, Nan, Bool, Inf:
|
|
# A value type's type is always equal to
|
|
# another one's
|
|
return true
|
|
of Reference, Pointer:
|
|
# Here we already know that both
|
|
# a and b are of either of the two
|
|
# types in this branch, so we just need
|
|
# to compare their values
|
|
return self.compareTypes(a.value, b.value)
|
|
of Function:
|
|
# Functions are a bit trickier
|
|
if a.args.len() != b.args.len():
|
|
return false
|
|
elif not self.compareTypes(a.returnType, b.returnType):
|
|
return false
|
|
for (argA, argB) in zip(a.args, b.args):
|
|
if not self.compareTypes(argA.kind, argB.kind):
|
|
return false
|
|
return true
|
|
else:
|
|
# TODO: Custom types
|
|
discard
|
|
|
|
|
|
proc toIntrinsic(name: string): Type =
|
|
## Converts a string to an intrinsic
|
|
## type if it is valid and returns nil
|
|
## otherwise
|
|
if name in ["int", "int64", "i64"]:
|
|
return Type(kind: Int64)
|
|
elif name in ["uint64", "u64", "uint"]:
|
|
return Type(kind: UInt64)
|
|
elif name in ["int32", "i32"]:
|
|
return Type(kind: Int32)
|
|
elif name in ["uint32", "u32"]:
|
|
return Type(kind: UInt32)
|
|
elif name in ["int16", "i16", "short"]:
|
|
return Type(kind: Int16)
|
|
elif name in ["uint16", "u16"]:
|
|
return Type(kind: UInt16)
|
|
elif name in ["int8", "i8"]:
|
|
return Type(kind: Int8)
|
|
elif name in ["uint8", "u8"]:
|
|
return Type(kind: UInt8)
|
|
elif name in ["f64", "float", "float64"]:
|
|
return Type(kind: Float64)
|
|
elif name in ["f32", "float32"]:
|
|
return Type(kind: Float32)
|
|
elif name in ["byte", "b"]:
|
|
return Type(kind: Byte)
|
|
elif name in ["char", "c"]:
|
|
return Type(kind: Char)
|
|
elif name == "nan":
|
|
return Type(kind: Nan)
|
|
elif name == "nil":
|
|
return Type(kind: Nil)
|
|
elif name == "inf":
|
|
return Type(kind: Inf)
|
|
elif name == "bool":
|
|
return Type(kind: Bool)
|
|
elif name == "typevar":
|
|
return Type(kind: Typevar)
|
|
elif name == "string":
|
|
return Type(kind: String)
|
|
else:
|
|
return nil
|
|
|
|
|
|
proc inferType(self: Compiler, node: LiteralExpr): Type =
|
|
## Infers the type of a given literal expression
|
|
if node.isNil():
|
|
return nil
|
|
case node.kind:
|
|
of intExpr, binExpr, octExpr, hexExpr:
|
|
let size = node.token.lexeme.split("'")
|
|
if len(size) notin 1..2:
|
|
self.error("invalid state: inferValueType -> invalid size specifier (This is an internal error and most likely a bug!)")
|
|
if size.len() == 1:
|
|
return Type(kind: Int64)
|
|
let typ = size[1].toIntrinsic()
|
|
if not self.compareTypes(typ, nil):
|
|
return typ
|
|
else:
|
|
self.error(&"invalid type specifier '{size[1]}' for int")
|
|
of floatExpr:
|
|
let size = node.token.lexeme.split("'")
|
|
if len(size) notin 1..2:
|
|
self.error("invalid state: inferValueType -> invalid size specifier (This is an internal error and most likely a bug!)")
|
|
if size.len() == 1 or size[1] == "f64":
|
|
return Type(kind: Float64)
|
|
let typ = size[1].toIntrinsic()
|
|
if not self.compareTypes(typ, nil):
|
|
return typ
|
|
else:
|
|
self.error(&"invalid type specifier '{size[1]}' for float")
|
|
of nilExpr:
|
|
return Type(kind: Nil)
|
|
of trueExpr:
|
|
return Type(kind: Bool)
|
|
of falseExpr:
|
|
return Type(kind: Bool)
|
|
of nanExpr:
|
|
return Type(kind: TypeKind.Nan)
|
|
of infExpr:
|
|
return Type(kind: TypeKind.Inf)
|
|
of strExpr:
|
|
return Type(kind: String)
|
|
else:
|
|
discard # TODO
|
|
|
|
proc matchImpl(self: Compiler, name: string, kind: Type): Name
|
|
|
|
|
|
proc inferType(self: Compiler, node: Expression): Type =
|
|
## Infers the type of a given expression and
|
|
## returns it
|
|
if node.isNil():
|
|
return nil
|
|
case node.kind:
|
|
of identExpr:
|
|
let node = IdentExpr(node)
|
|
let name = self.resolve(node)
|
|
if not name.isNil():
|
|
result = name.valueType
|
|
else:
|
|
result = node.name.lexeme.toIntrinsic()
|
|
of unaryExpr:
|
|
let node = UnaryExpr(node)
|
|
return self.matchImpl(node.operator.lexeme, Type(kind: Function, returnType: Type(kind: Any), args: @[("", self.inferType(node.a))])).valueType.returnType
|
|
of binaryExpr:
|
|
let node = BinaryExpr(node)
|
|
return self.matchImpl(node.operator.lexeme, Type(kind: Function, returnType: Type(kind: Any), args: @[("", self.inferType(node.a)), ("", self.inferType(node.b))])).valueType.returnType
|
|
of {intExpr, hexExpr, binExpr, octExpr,
|
|
strExpr, falseExpr, trueExpr, infExpr,
|
|
nanExpr, floatExpr, nilExpr
|
|
}:
|
|
return self.inferType(LiteralExpr(node))
|
|
of lambdaExpr:
|
|
var node = LambdaExpr(node)
|
|
result = Type(kind: Function, returnType: nil, args: @[], isLambda: true)
|
|
if not node.returnType.isNil():
|
|
result.returnType = self.inferType(node.returnType)
|
|
for argument in node.arguments:
|
|
result.args.add((argument.name.token.lexeme, self.inferType(argument.valueType)))
|
|
of callExpr:
|
|
var node = CallExpr(node)
|
|
case node.callee.kind:
|
|
of identExpr:
|
|
let resolved = self.resolve(IdentExpr(node.callee))
|
|
if not resolved.isNil():
|
|
result = resolved.valueType.returnType
|
|
else:
|
|
result = nil
|
|
of lambdaExpr:
|
|
result = self.inferType(LambdaExpr(node.callee).returnType)
|
|
of callExpr:
|
|
result = self.inferType(CallExpr(node.callee))
|
|
if not result.isNil():
|
|
result = result.returnType
|
|
else:
|
|
discard # Unreachable
|
|
of varExpr:
|
|
result = self.inferType(Var(node).value)
|
|
result.mutable = true
|
|
of refExpr:
|
|
result = Type(kind: Reference, value: self.inferType(Ref(node).value))
|
|
of ptrExpr:
|
|
result = Type(kind: Pointer, value: self.inferType(Ptr(node).value))
|
|
of groupingExpr:
|
|
result = self.inferType(GroupingExpr(node).expression)
|
|
else:
|
|
discard # Unreachable
|
|
|
|
|
|
proc inferType(self: Compiler, node: Declaration, strictMutable: bool = true): Type =
|
|
## Infers the type of a given declaration
|
|
## and returns it
|
|
if node.isNil():
|
|
return nil
|
|
case node.kind:
|
|
of NodeKind.funDecl:
|
|
var node = FunDecl(node)
|
|
let resolved = self.resolve(node.name)
|
|
if not resolved.isNil():
|
|
return resolved.valueType
|
|
of NodeKind.varDecl:
|
|
var node = VarDecl(node)
|
|
let resolved = self.resolve(node.name)
|
|
if not resolved.isNil():
|
|
return resolved.valueType
|
|
else:
|
|
return self.inferType(node.value, strictMutable)
|
|
else:
|
|
return # Unreachable
|
|
|
|
|
|
proc typeToStr(self: Compiler, typ: Type): string =
|
|
## Returns the string representation of a
|
|
## type object
|
|
if typ.isNil():
|
|
return "nil"
|
|
case typ.kind:
|
|
of Int8, UInt8, Int16, UInt16, Int32,
|
|
UInt32, Int64, UInt64, Float32, Float64,
|
|
Char, Byte, String, Nil, TypeKind.Nan, Bool,
|
|
TypeKind.Inf:
|
|
result &= ($typ.kind).toLowerAscii()
|
|
of Pointer:
|
|
result &= &"ptr {self.typeToStr(typ.value)}"
|
|
of Reference:
|
|
result &= &"ref {self.typeToStr(typ.value)}"
|
|
of Function:
|
|
result &= "fn ("
|
|
for i, (argName, argType) in typ.args:
|
|
result &= &"{argName}: "
|
|
if argType.mutable:
|
|
result &= "var "
|
|
result &= self.typeToStr(argType)
|
|
if i < typ.args.len() - 1:
|
|
result &= ", "
|
|
result &= ")"
|
|
if not typ.returnType.isNil():
|
|
result &= &": {self.typeToStr(typ.returnType)}"
|
|
of Generic:
|
|
result = typ.node.name.lexeme
|
|
else:
|
|
discard
|
|
|
|
|
|
proc findByName(self: Compiler, name: string): seq[Name] =
|
|
## Looks for objects that have been already declared
|
|
## with the given name. Returns all objects that apply
|
|
for obj in reversed(self.names):
|
|
if obj.name.token.lexeme == name:
|
|
if obj.isPrivate and obj.owner != self.currentModule:
|
|
continue
|
|
result.add(obj)
|
|
|
|
|
|
proc findByType(self: Compiler, name: string, kind: Type, depth: int = -1): seq[Name] =
|
|
## Looks for objects that have already been declared
|
|
## with the given name and type. If depth is not -1,
|
|
## it also compares the name's scope depth
|
|
for obj in self.findByName(name):
|
|
if self.compareTypes(obj.valueType, kind) and depth == -1 or depth == obj.depth:
|
|
result.add(obj)
|
|
|
|
#[
|
|
proc findAtDepth(self: Compiler, name: string, depth: int): seq[Name] =
|
|
## Looks for objects that have been already declared
|
|
## with the given name at the given scope depth.
|
|
## Returns all objects that apply
|
|
for obj in self.findByName(name):
|
|
if obj.depth == depth:
|
|
result.add(obj)
|
|
]#
|
|
|
|
|
|
proc matchImpl(self: Compiler, name: string, kind: Type): Name =
|
|
## Tries to find a matching function implementation
|
|
## compatible with the given type and returns its
|
|
## name object
|
|
let impl = self.findByType(name, kind)
|
|
if impl.len() == 0:
|
|
var msg = &"cannot find a suitable implementation for '{name}'"
|
|
let names = self.findByName(name)
|
|
if names.len() > 0:
|
|
msg &= &", found {len(names)} candidate"
|
|
if names.len() > 1:
|
|
msg &= "s"
|
|
msg &= ": "
|
|
for name in names:
|
|
msg &= &"\n - in module '{name.owner}' at line {name.name.token.line} of type '{self.typeToStr(name.valueType)}'"
|
|
if name.valueType.kind != Function:
|
|
msg &= ", not a callable"
|
|
elif kind.args.len() != name.valueType.args.len():
|
|
msg &= &", wrong number of arguments ({name.valueType.args.len()} expected, got {kind.args.len()})"
|
|
else:
|
|
for i, arg in kind.args:
|
|
if name.valueType.args[i].kind.mutable and not arg.kind.mutable:
|
|
msg &= &", first mismatch at position {i + 1}: {name.valueType.args[i].name} is immutable, not 'var'"
|
|
break
|
|
elif not self.compareTypes(arg.kind, name.valueType.args[i].kind):
|
|
msg &= &", first mismatch at position {i + 1}: expected argument of type '{self.typeToStr(name.valueType.args[i].kind)}', got '{self.typeToStr(arg.kind)}' instead"
|
|
break
|
|
self.error(msg)
|
|
elif impl.len() > 1:
|
|
var msg = &"multiple matching implementations of '{name}' found:\n"
|
|
for fn in reversed(impl):
|
|
msg &= &"- '{fn.name.token.lexeme}' at line {fn.line} of type {self.typeToStr(fn.valueType)}\n"
|
|
self.error(msg)
|
|
return impl[0]
|
|
|
|
|
|
proc check(self: Compiler, term: Expression, kind: Type) =
|
|
## Checks the type of term against a known type.
|
|
## Raises an error if appropriate and returns
|
|
## otherwise
|
|
let k = self.inferType(term)
|
|
if k.isNil():
|
|
if term.kind == identExpr:
|
|
self.error(&"reference to undeclared name '{term.token.lexeme}'", term)
|
|
elif term.kind == callExpr and CallExpr(term).callee.kind == identExpr:
|
|
self.error(&"call to undeclared function '{CallExpr(term).callee.token.lexeme}'", term)
|
|
self.error(&"expecting value of type '{self.typeToStr(kind)}', but expression has no type", term)
|
|
elif not self.compareTypes(k, kind):
|
|
self.error(&"expecting value of type '{self.typeToStr(kind)}', got '{self.typeToStr(k)}' instead", term)
|
|
|
|
|
|
|
|
proc emitFunction(self: Compiler, name: Name) =
|
|
## Wrapper to emit LoadFunction instructions
|
|
if name.isFunDecl:
|
|
self.emitByte(LoadFunction)
|
|
self.emitBytes(name.codePos.toTriple())
|
|
# If we're not loading a statically declared
|
|
# function, then it must be a function object
|
|
# created by previous LoadFunction instructions
|
|
# that is now bound to some variable, so we just
|
|
# load it
|
|
elif not name.isClosedOver:
|
|
self.emitByte(LoadVar)
|
|
self.emitBytes(self.getStackPos(name).toTriple())
|
|
else:
|
|
self.emitByte(LoadClosure)
|
|
self.emitBytes(self.getClosurePos(name).toTriple())
|
|
|
|
## End of utility functions
|
|
|
|
|
|
proc literal(self: Compiler, node: ASTNode) =
|
|
## Emits instructions for literals such
|
|
## as singletons, strings and numbers
|
|
case node.kind:
|
|
of trueExpr:
|
|
self.emitByte(LoadTrue)
|
|
of falseExpr:
|
|
self.emitByte(LoadFalse)
|
|
of nilExpr:
|
|
self.emitByte(LoadNil)
|
|
of infExpr:
|
|
self.emitByte(LoadInf)
|
|
of nanExpr:
|
|
self.emitByte(LoadNan)
|
|
of strExpr:
|
|
self.emitConstant(LiteralExpr(node), Type(kind: String))
|
|
of intExpr:
|
|
let y = IntExpr(node)
|
|
let kind = self.inferType(y)
|
|
if kind.kind in [Int64, Int32, Int16, Int8]:
|
|
var x: int
|
|
try:
|
|
discard parseInt(y.literal.lexeme, x)
|
|
except ValueError:
|
|
self.error("integer value out of range")
|
|
else:
|
|
var x: uint64
|
|
try:
|
|
discard parseBiggestUInt(y.literal.lexeme, x)
|
|
except ValueError:
|
|
self.error("integer value out of range")
|
|
self.emitConstant(y, kind)
|
|
of hexExpr:
|
|
var x: int
|
|
var y = HexExpr(node)
|
|
try:
|
|
discard parseHex(y.literal.lexeme, x)
|
|
except ValueError:
|
|
self.error("integer value out of range")
|
|
let node = newIntExpr(Token(lexeme: $x, line: y.token.line,
|
|
pos: (start: y.token.pos.start,
|
|
stop: y.token.pos.start + len($x))
|
|
)
|
|
)
|
|
self.emitConstant(node, self.inferType(y))
|
|
of binExpr:
|
|
var x: int
|
|
var y = BinExpr(node)
|
|
try:
|
|
discard parseBin(y.literal.lexeme, x)
|
|
except ValueError:
|
|
self.error("integer value out of range")
|
|
let node = newIntExpr(Token(lexeme: $x, line: y.token.line,
|
|
pos: (start: y.token.pos.start,
|
|
stop: y.token.pos.start + len($x))
|
|
)
|
|
)
|
|
self.emitConstant(node, self.inferType(y))
|
|
of octExpr:
|
|
var x: int
|
|
var y = OctExpr(node)
|
|
try:
|
|
discard parseOct(y.literal.lexeme, x)
|
|
except ValueError:
|
|
self.error("integer value out of range")
|
|
let node = newIntExpr(Token(lexeme: $x, line: y.token.line,
|
|
pos: (start: y.token.pos.start,
|
|
stop: y.token.pos.start + len($x))
|
|
)
|
|
)
|
|
self.emitConstant(node, self.inferType(y))
|
|
of floatExpr:
|
|
var x: float
|
|
var y = FloatExpr(node)
|
|
try:
|
|
discard parseFloat(y.literal.lexeme, x)
|
|
except ValueError:
|
|
self.error("floating point value out of range")
|
|
self.emitConstant(y, self.inferType(y))
|
|
of awaitExpr:
|
|
var y = AwaitExpr(node)
|
|
self.expression(y.expression)
|
|
self.emitByte(OpCode.Await)
|
|
else:
|
|
self.error(&"invalid AST node of kind {node.kind} at literal(): {node} (This is an internal error and most likely a bug!)")
|
|
|
|
|
|
|
|
proc handleBuiltinFunction(self: Compiler, fn: Name, args: seq[Expression]) =
|
|
## Emits instructions for builtin functions
|
|
## such as addition or subtraction
|
|
if fn.valueType.builtinOp notin ["LogicalOr", "LogicalAnd"]:
|
|
if len(args) == 2:
|
|
self.expression(args[1])
|
|
self.expression(args[0])
|
|
elif len(args) == 1:
|
|
self.expression(args[0])
|
|
const codes: Table[string, OpCode] = {"Negate": Negate,
|
|
"NegateFloat32": NegateFloat32,
|
|
"NegateFloat64": NegateFloat64,
|
|
"Add": Add,
|
|
"Subtract": Subtract,
|
|
"Divide": Divide,
|
|
"Multiply": Multiply,
|
|
"SignedDivide": SignedDivide,
|
|
"AddFloat64": AddFloat64,
|
|
"SubtractFloat64": SubtractFloat64,
|
|
"DivideFloat64": DivideFloat64,
|
|
"MultiplyFloat64": MultiplyFloat64,
|
|
"AddFloat32": AddFloat32,
|
|
"SubtractFloat32": SubtractFloat32,
|
|
"DivideFloat32": DivideFloat32,
|
|
"MultiplyFloat32": MultiplyFloat32,
|
|
"Pow": Pow,
|
|
"SignedPow": SignedPow,
|
|
"PowFloat32": PowFloat32,
|
|
"PowFloat64": PowFloat64,
|
|
"Mod": Mod,
|
|
"SignedMod": SignedMod,
|
|
"ModFloat32": ModFloat32,
|
|
"ModFloat64": ModFloat64,
|
|
"Or": Or,
|
|
"And": And,
|
|
"Xor": Xor,
|
|
"Not": Not,
|
|
"LShift": LShift,
|
|
"RShift": RShift,
|
|
"Equal": Equal,
|
|
"NotEqual": NotEqual,
|
|
"LessThan": LessThan,
|
|
"GreaterThan": GreaterThan,
|
|
"LessOrEqual": LessOrEqual,
|
|
"GreaterOrEqual": GreaterOrEqual,
|
|
"PrintInt64": PrintInt64,
|
|
"PrintUInt64": PrintUInt64,
|
|
"PrintInt32": PrintInt32,
|
|
"PrintUInt32": PrintUInt32,
|
|
"PrintInt16": PrintInt16,
|
|
"PrintUInt16": PrintUInt16,
|
|
"PrintInt8": PrintInt8,
|
|
"PrintUInt8": PrintUInt8,
|
|
"PrintFloat64": PrintFloat64,
|
|
"PrintFloat32": PrintFloat32,
|
|
"PrintHex": PrintHex,
|
|
"PrintBool": PrintBool,
|
|
"PrintNan": PrintNan,
|
|
"PrintInf": PrintInf,
|
|
"PrintString": PrintString,
|
|
"SysClock64": SysClock64
|
|
}.to_table()
|
|
if fn.valueType.builtinOp in codes:
|
|
self.emitByte(codes[fn.valueType.builtinOp])
|
|
return
|
|
# Some builtin operations are slightly more complex
|
|
# so we handle them separately
|
|
case fn.valueType.builtinOp:
|
|
of "LogicalOr":
|
|
self.expression(args[0])
|
|
let jump = self.emitJump(JumpIfTrue)
|
|
self.expression(args[1])
|
|
self.patchJump(jump)
|
|
of "LogicalAnd":
|
|
self.expression(args[0])
|
|
var jump = self.emitJump(JumpIfFalseOrPop)
|
|
self.expression(args[1])
|
|
self.patchJump(jump)
|
|
else:
|
|
self.error(&"unknown built-in: '{fn.valueType.builtinOp}'", fn.valueType.fun)
|
|
|
|
|
|
proc generateCall(self: Compiler, fn: Name, args: seq[Expression], onStack: bool = false) =
|
|
## Small wrapper that abstracts emitting a call instruction
|
|
## for a given function
|
|
if fn.valueType.isBuiltinFunction:
|
|
# Builtins map to individual instructions
|
|
# (usually 1, but some use more) so we handle
|
|
# them differently
|
|
self.handleBuiltinFunction(fn, args)
|
|
return
|
|
if not onStack:
|
|
self.emitFunction(fn)
|
|
self.emitByte(LoadReturnAddress)
|
|
let pos = self.chunk.code.len()
|
|
# We initially emit a dummy return
|
|
# address. It is patched later
|
|
self.emitBytes(0.toQuad())
|
|
for argument in reversed(args):
|
|
# We pass the arguments in reverse
|
|
# because of how stack semantics
|
|
# work. They'll be fixed at runtime
|
|
self.expression(argument)
|
|
var f = fn.valueType
|
|
while not f.isNil():
|
|
if f.isClosure:
|
|
for i in f.closureBounds.start..f.closureBounds.stop:
|
|
self.closedOver[i].count += 1
|
|
f = f.childFunc
|
|
# Creates a new call frame and jumps
|
|
# to the function's first instruction
|
|
# in the code
|
|
self.emitByte(Call)
|
|
self.emitBytes(fn.valueType.args.len().toTriple())
|
|
self.patchReturnAddress(pos)
|
|
|
|
|
|
proc callUnaryOp(self: Compiler, fn: Name, op: UnaryExpr) =
|
|
## Emits the code to call a unary operator
|
|
self.generateCall(fn, @[op.a])
|
|
|
|
|
|
proc callBinaryOp(self: Compiler, fn: Name, op: BinaryExpr) =
|
|
## Emits the code to call a binary operator
|
|
self.generateCall(fn, @[op.a, op.b])
|
|
|
|
|
|
proc unary(self: Compiler, node: UnaryExpr) =
|
|
## Compiles unary expressions such as decimal
|
|
## and bitwise negation
|
|
let valueType = self.inferType(node.a)
|
|
let funct = self.matchImpl(node.token.lexeme, Type(kind: Function, returnType: Type(kind: Any), args: @[("", valueType)]))
|
|
self.callUnaryOp(funct, node)
|
|
|
|
|
|
proc binary(self: Compiler, node: BinaryExpr) =
|
|
## Compiles all binary expressions
|
|
let typeOfA = self.inferType(node.a)
|
|
let typeOfB = self.inferType(node.b)
|
|
let funct = self.matchImpl(node.token.lexeme, Type(kind: Function, returnType: Type(kind: Any), args: @[("", typeOfA), ("", typeOfB)]))
|
|
self.callBinaryOp(funct, node)
|
|
|
|
|
|
proc declareName(self: Compiler, node: Declaration, mutable: bool = false) =
|
|
## Statically declares a name into the current scope.
|
|
## "Declaring" a name only means updating our internal
|
|
## list of identifiers so that further calls to resolve()
|
|
## correctly return them. There is no code to actually
|
|
## declare a variable at runtime: the value is already
|
|
## on the stack
|
|
case node.kind:
|
|
of NodeKind.varDecl:
|
|
var node = VarDecl(node)
|
|
# Creates a new Name entry so that self.identifier emits the proper stack offset
|
|
if self.names.high() > 16777215:
|
|
# If someone ever hits this limit in real-world scenarios, I swear I'll
|
|
# slap myself 100 times with a sign saying "I'm dumb". Mark my words
|
|
self.error("cannot declare more than 16777215 variables at a time")
|
|
for name in self.findByName(node.name.token.lexeme):
|
|
if name.depth == self.scopeDepth and not name.isFunctionArgument:
|
|
# Trying to redeclare a variable in the same scope/context is an error, but it's okay
|
|
# if it's a function argument (for example, if you want to copy a number to
|
|
# mutate it)
|
|
self.error(&"attempt to redeclare '{node.name.token.lexeme}', which was previously defined in '{name.owner}' at line {name.line}")
|
|
self.names.add(Name(depth: self.scopeDepth,
|
|
name: node.name,
|
|
isPrivate: node.isPrivate,
|
|
owner: self.currentModule,
|
|
isConst: node.isConst,
|
|
valueType: self.inferType(node.value),
|
|
codePos: self.chunk.code.len(),
|
|
isLet: node.isLet,
|
|
isClosedOver: false,
|
|
line: node.token.line,
|
|
belongsTo: self.currentFunction
|
|
))
|
|
if mutable:
|
|
self.names[^1].valueType.mutable = true
|
|
# We emit a jump of 0 because this may become a
|
|
# StoreHeap instruction. If they variable is
|
|
# not closed over, we'll sadly be wasting a
|
|
# VM cycle. The previous implementation used 4 no-op
|
|
# instructions, which wasted 4 times as many clock
|
|
# cycles.
|
|
# TODO: Optimize this. It's a bit tricky because
|
|
# deleting bytecode would render all of our
|
|
# jump offsets and other absolute indeces in the
|
|
# bytecode wrong
|
|
if self.scopeDepth > 0:
|
|
# Closure variables are only used in local
|
|
# scopes
|
|
self.emitByte(JumpForwards)
|
|
self.emitBytes(0.toTriple())
|
|
of NodeKind.funDecl:
|
|
var node = FunDecl(node)
|
|
# We declare the generics before the function so we
|
|
# can refer to them
|
|
for gen in node.generics:
|
|
self.names.add(Name(depth: self.scopeDepth + 1,
|
|
isPrivate: true,
|
|
isConst: false,
|
|
owner: self.currentModule,
|
|
line: node.token.line,
|
|
valueType: Type(kind: Generic, mutable: false, node: gen.name),
|
|
name: gen.name))
|
|
self.names.add(Name(depth: self.scopeDepth,
|
|
isPrivate: node.isPrivate,
|
|
isConst: false,
|
|
owner: self.currentModule,
|
|
valueType: Type(kind: Function,
|
|
name: node.name.token.lexeme,
|
|
returnType: self.inferType(node.returnType),
|
|
args: @[],
|
|
fun: node),
|
|
codePos: self.chunk.code.len(),
|
|
name: node.name,
|
|
isLet: false,
|
|
isClosedOver: false,
|
|
line: node.token.line,
|
|
isFunDecl: true))
|
|
let fn = self.names[^1]
|
|
var name: Name
|
|
for argument in node.arguments:
|
|
if self.names.high() > 16777215:
|
|
self.error("cannot declare more than 16777215 variables at a time")
|
|
# wait, no LoadVar? Yes! That's because when calling functions,
|
|
# arguments will already be on the stack, so there's no need to
|
|
# load them here
|
|
name = Name(depth: self.scopeDepth + 1,
|
|
isPrivate: true,
|
|
owner: self.currentModule,
|
|
isConst: false,
|
|
name: argument.name,
|
|
valueType: nil,
|
|
codePos: 0,
|
|
isLet: false,
|
|
isClosedOver: false,
|
|
line: argument.name.token.line,
|
|
isFunctionArgument: true,
|
|
belongsTo: fn
|
|
)
|
|
self.names.add(name)
|
|
name.valueType = self.inferType(argument.valueType)
|
|
# If it's still nil, it's an error!
|
|
if name.valueType.isNil():
|
|
self.error(&"cannot determine the type of argument '{argument.name.token.lexeme}'", argument.name)
|
|
fn.valueType.args.add((argument.name.token.lexeme, name.valueType))
|
|
else:
|
|
discard # TODO: Types, enums
|
|
|
|
|
|
proc identifier(self: Compiler, node: IdentExpr) =
|
|
## Compiles access to identifiers
|
|
var s = self.resolve(node)
|
|
if s.isNil():
|
|
self.error(&"reference to undeclared name '{node.token.lexeme}'")
|
|
elif s.isConst:
|
|
# Constants are always emitted as Load* instructions
|
|
# no matter the scope depth
|
|
self.emitConstant(node, self.inferType(node))
|
|
else:
|
|
self.detectClosureVariable(s)
|
|
if s.valueType.kind == Function and s.isFunDecl:
|
|
# Functions have no runtime
|
|
# representation, so we need
|
|
# to create one on the fly
|
|
self.emitByte(LoadFunction)
|
|
self.emitBytes(s.codePos.toTriple())
|
|
elif not s.isClosedOver:
|
|
# Static name resolution, loads value at index in the stack. Very fast. Much wow.
|
|
self.emitByte(LoadVar)
|
|
# No need to check for -1 here: we already did a nil-check above!
|
|
self.emitBytes(self.getStackPos(s).toTriple())
|
|
else:
|
|
# Loads a closure variable. Stored in a separate "closure array" in the VM that does not
|
|
# align its semantics with the call stack. This makes closures work as expected and is
|
|
# not much slower than indexing our stack (since they're both dynamic arrays at runtime anyway)
|
|
self.emitByte(LoadClosure)
|
|
self.emitBytes((self.getClosurePos(s)).toTriple())
|
|
|
|
|
|
proc assignment(self: Compiler, node: ASTNode) =
|
|
## Compiles assignment expressions
|
|
case node.kind:
|
|
of assignExpr:
|
|
let node = AssignExpr(node)
|
|
let name = IdentExpr(node.name)
|
|
var r = self.resolve(name)
|
|
if r.isNil():
|
|
self.error(&"assignment to undeclared name '{name.token.lexeme}'", name)
|
|
elif r.isConst:
|
|
self.error(&"cannot assign to '{name.token.lexeme}' (constant)", name)
|
|
elif r.isLet:
|
|
self.error(&"cannot reassign '{name.token.lexeme}'", name)
|
|
self.expression(node.value)
|
|
self.detectClosureVariable(r)
|
|
if not r.isClosedOver:
|
|
self.emitByte(StoreVar)
|
|
self.emitBytes(self.getStackPos(r).toTriple())
|
|
else:
|
|
# Loads a closure variable. Stored in a separate "closure array" in the VM that does not
|
|
# align its semantics with the call stack. This makes closures work as expected and is
|
|
# not much slower than indexing our stack (since they're both dynamic arrays at runtime anyway)
|
|
self.emitByte(StoreClosure)
|
|
self.emitBytes(self.getClosurePos(r).toTriple())
|
|
of setItemExpr:
|
|
let node = SetItemExpr(node)
|
|
let typ = self.inferType(node)
|
|
if typ.isNil():
|
|
self.error(&"cannot determine the type of '{node.name.token.lexeme}'")
|
|
# TODO
|
|
else:
|
|
self.error(&"invalid AST node of kind {node.kind} at assignment(): {node} (This is an internal error and most likely a bug)")
|
|
|
|
|
|
proc beginScope(self: Compiler) =
|
|
## Begins a new local scope by incrementing the current
|
|
## scope's depth
|
|
inc(self.scopeDepth)
|
|
self.scopes.add(self.currentFunction.valueType)
|
|
|
|
|
|
proc endScope(self: Compiler) =
|
|
## Ends the current local scope
|
|
if self.scopeDepth < 0:
|
|
self.error("cannot call endScope with scopeDepth < 0 (This is an internal error and most likely a bug)")
|
|
dec(self.scopeDepth)
|
|
if self.scopeDepth > 0:
|
|
discard self.scopes.pop()
|
|
var names: seq[Name] = @[]
|
|
var popCount = 0
|
|
for name in self.names:
|
|
if name.depth > self.scopeDepth:
|
|
names.add(name)
|
|
if name.valueType.kind notin {Generic, CustomType} and not name.isFunDecl:
|
|
# We don't increase the pop count for these kinds of objects
|
|
# because they're not stored the same way as regular variables
|
|
inc(popCount)
|
|
if name.isFunDecl and not name.valueType.childFunc.isNil() and name.valueType.childFunc.isClosure:
|
|
var i = 0
|
|
var closure: tuple[name: Name, count: int]
|
|
for y in name.valueType.childFunc.closureBounds.start..name.valueType.childFunc.closureBounds.stop:
|
|
closure = self.closedOver[y + i]
|
|
self.closedOver.delete(y + i)
|
|
for _ in 0..<closure.count:
|
|
self.emitByte(PopClosure)
|
|
self.emitBytes((y + i).toTriple())
|
|
inc(i)
|
|
if popCount > 1:
|
|
# If we're popping less than 65535 variables, then
|
|
# we can emit a PopN instruction. This is true for
|
|
# 99.99999% of the use cases of the language (who the
|
|
# hell is going to use 65 THOUSAND variables?), but
|
|
# if you'll ever use more then Peon will emit a PopN instruction
|
|
# for the first 65 thousand and change local variables and then
|
|
# emit another batch of plain ol' Pop instructions for the rest
|
|
self.emitByte(PopN)
|
|
self.emitBytes(popCount.toDouble())
|
|
if popCount > uint16.high().int():
|
|
for i in countdown(self.names.high(), popCount - uint16.high().int()):
|
|
if self.names[i].depth > self.scopeDepth:
|
|
self.emitByte(PopC)
|
|
elif popCount == 1:
|
|
# We only emit PopN if we're popping more than one value
|
|
self.emitByte(PopC)
|
|
# This seems *really* slow, but
|
|
# what else should I do? Nim doesn't
|
|
# allow the removal of items during
|
|
# seq iteration so ¯\_(ツ)_/¯
|
|
var idx = 0
|
|
while idx < self.names.len():
|
|
for name in names:
|
|
if self.names[idx] == name:
|
|
self.names.delete(idx)
|
|
inc(idx)
|
|
|
|
|
|
proc blockStmt(self: Compiler, node: BlockStmt) =
|
|
## Compiles block statements, which create a new
|
|
## local scope
|
|
self.beginScope()
|
|
for decl in node.code:
|
|
self.declaration(decl)
|
|
self.endScope()
|
|
|
|
|
|
proc ifStmt(self: Compiler, node: IfStmt) =
|
|
## Compiles if/else statements for conditional
|
|
## execution of code
|
|
self.check(node.condition, Type(kind: Bool))
|
|
self.expression(node.condition)
|
|
let jump = self.emitJump(JumpIfFalsePop)
|
|
self.statement(node.thenBranch)
|
|
let jump2 = self.emitJump(JumpForwards)
|
|
self.patchJump(jump)
|
|
if not node.elseBranch.isNil():
|
|
self.statement(node.elseBranch)
|
|
self.patchJump(jump2)
|
|
|
|
|
|
proc emitLoop(self: Compiler, begin: int) =
|
|
## Emits a JumpBackwards instruction with the correct
|
|
## jump offset
|
|
let offset = self.chunk.code.high() - begin + 4
|
|
if offset > 16777215:
|
|
self.error("cannot jump more than 16777215 bytecode instructions")
|
|
self.emitByte(JumpBackwards)
|
|
self.emitBytes(offset.toTriple())
|
|
|
|
|
|
proc whileStmt(self: Compiler, node: WhileStmt) =
|
|
## Compiles C-style while loops and
|
|
## desugared C-style for loops
|
|
self.check(node.condition, Type(kind: Bool))
|
|
let start = self.chunk.code.high()
|
|
self.expression(node.condition)
|
|
let jump = self.emitJump(JumpIfFalsePop)
|
|
self.statement(node.body)
|
|
self.emitLoop(start)
|
|
self.patchJump(jump)
|
|
|
|
|
|
proc checkCallIsPure(self: Compiler, node: ASTnode): bool =
|
|
## Checks if a call has any side effects. Returns
|
|
## true if it doesn't and false otherwise
|
|
return true # TODO
|
|
|
|
|
|
proc callExpr(self: Compiler, node: CallExpr): Name {.discardable.} =
|
|
## Compiles code to call a function
|
|
var args: seq[tuple[name: string, kind: Type]] = @[]
|
|
var argExpr: seq[Expression] = @[]
|
|
var kind: Type
|
|
var onStack = false
|
|
# TODO: Keyword arguments
|
|
for i, argument in node.arguments.positionals:
|
|
kind = self.inferType(argument)
|
|
if kind.isNil():
|
|
if argument.kind == identExpr:
|
|
self.error(&"reference to undeclared name '{IdentExpr(argument).name.lexeme}'")
|
|
self.error(&"cannot infer the type of argument {i + 1} in function call")
|
|
args.add(("", kind))
|
|
argExpr.add(argument)
|
|
for argument in node.arguments.keyword:
|
|
# TODO
|
|
discard
|
|
if args.len() >= 16777216:
|
|
self.error(&"cannot pass more than 16777215 arguments")
|
|
var funct: Name
|
|
case node.callee.kind:
|
|
of identExpr:
|
|
funct = self.matchImpl(IdentExpr(node.callee).name.lexeme, Type(kind: Function, returnType: Type(kind: Any), args: args))
|
|
of NodeKind.callExpr:
|
|
funct = self.callExpr(CallExpr(node.callee))
|
|
funct = Name(valueType: Type(kind: Function, returnType: Type(kind: Any), args: args))
|
|
onStack = true
|
|
# TODO: Calling lambdas on-the-fly (i.e. on the same line)
|
|
else:
|
|
let typ = self.inferType(node)
|
|
if typ.isNil():
|
|
self.error(&"expression has no type")
|
|
else:
|
|
self.error(&"object of type '{self.typeToStr(typ)}' is not callable")
|
|
result = funct
|
|
self.generateCall(funct, argExpr, onStack)
|
|
if not self.checkCallIsPure(node.callee):
|
|
if self.currentFunction.valueType.name != "":
|
|
self.error(&"cannot make sure that calls to '{self.currentFunction.valueType.name}' are side-effect free")
|
|
else:
|
|
self.error(&"cannot make sure that call is side-effect free")
|
|
|
|
|
|
proc expression(self: Compiler, node: Expression) =
|
|
## Compiles all expressions
|
|
case node.kind:
|
|
of NodeKind.callExpr:
|
|
self.callExpr(CallExpr(node)) # TODO
|
|
of getItemExpr:
|
|
discard # TODO: Get rid of this
|
|
of pragmaExpr:
|
|
discard # TODO
|
|
# Note that for setItem and assign we don't convert
|
|
# the node to its true type because that type information
|
|
# would be lost in the call anyway. The differentiation
|
|
# happens in self.assignment()
|
|
of setItemExpr, assignExpr: # TODO: Get rid of this
|
|
self.assignment(node)
|
|
of identExpr:
|
|
self.identifier(IdentExpr(node))
|
|
of unaryExpr:
|
|
# Unary expressions such as ~5 and -3
|
|
self.unary(UnaryExpr(node))
|
|
of groupingExpr:
|
|
# Grouping expressions like (2 + 1)
|
|
self.expression(GroupingExpr(node).expression)
|
|
of binaryExpr:
|
|
# Binary expressions such as 2 ^ 5 and 0.66 * 3.14
|
|
self.binary(BinaryExpr(node))
|
|
of intExpr, hexExpr, binExpr, octExpr, strExpr, falseExpr, trueExpr,
|
|
infExpr, nanExpr, floatExpr, nilExpr:
|
|
# Since all of these AST nodes share the
|
|
# same overall structure and the kind
|
|
# field is enough to tell one from the
|
|
# other, why bother with specialized
|
|
# cases when one is enough?
|
|
self.literal(node)
|
|
else:
|
|
self.error(&"invalid AST node of kind {node.kind} at expression(): {node} (This is an internal error and most likely a bug)")
|
|
|
|
|
|
proc awaitStmt(self: Compiler, node: AwaitStmt) =
|
|
## Compiles await statements. An await statement
|
|
## is like an await expression, but parsed in the
|
|
## context of statements for usage outside expressions,
|
|
## meaning it can be used standalone. It's basically the
|
|
## same as an await expression followed by a semicolon.
|
|
## Await expressions are the only native construct to
|
|
## run coroutines from within an already asynchronous
|
|
## context (which should be orchestrated by an event loop).
|
|
## They block in the caller until the callee returns
|
|
self.expression(node.expression)
|
|
self.emitByte(OpCode.Await)
|
|
|
|
|
|
proc deferStmt(self: Compiler, node: DeferStmt) =
|
|
## Compiles defer statements. A defer statement
|
|
## is executed right before its containing function
|
|
## exits (either because of a return or an exception)
|
|
let current = self.chunk.code.len
|
|
self.expression(node.expression)
|
|
for i in countup(current, self.chunk.code.high()):
|
|
self.deferred.add(self.chunk.code[i])
|
|
self.chunk.code.delete(i) # TODO: Do not change bytecode size
|
|
|
|
|
|
|
|
proc returnStmt(self: Compiler, node: ReturnStmt) =
|
|
## Compiles return statements
|
|
var expected = self.currentFunction.valueType.returnType
|
|
self.check(node.value, expected)
|
|
if not node.value.isNil():
|
|
self.expression(node.value)
|
|
self.emitByte(OpCode.SetResult)
|
|
self.emitByte(OpCode.Return)
|
|
if not node.value.isNil():
|
|
self.emitByte(1)
|
|
else:
|
|
self.emitByte(0)
|
|
|
|
|
|
# TODO: Implement this as a custom operator
|
|
proc yieldStmt(self: Compiler, node: YieldStmt) =
|
|
## Compiles yield statements
|
|
self.expression(node.expression)
|
|
self.emitByte(OpCode.Yield)
|
|
|
|
# TODO: Implement this as a custom operator
|
|
proc raiseStmt(self: Compiler, node: RaiseStmt) =
|
|
## Compiles raise statements
|
|
self.expression(node.exception)
|
|
self.emitByte(OpCode.Raise)
|
|
|
|
|
|
proc continueStmt(self: Compiler, node: ContinueStmt) =
|
|
## Compiles continue statements. A continue statements
|
|
## jumps to the next iteration in a loop
|
|
if self.currentLoop.start > 16777215:
|
|
self.error("too much code to jump over in continue statement")
|
|
self.emitByte(Jump)
|
|
self.emitBytes(self.currentLoop.start.toTriple())
|
|
|
|
|
|
proc breakStmt(self: Compiler, node: BreakStmt) =
|
|
## Compiles break statements. A continue statement
|
|
## jumps to the next iteration in a loop
|
|
self.currentLoop.breakPos.add(self.emitJump(OpCode.JumpForwards))
|
|
if self.currentLoop.depth > self.scopeDepth:
|
|
# Breaking out of a loop closes its scope
|
|
self.endScope()
|
|
|
|
|
|
proc patchBreaks(self: Compiler) =
|
|
## Patches the jumps emitted by
|
|
## breakStmt. This is needed
|
|
## because the size of code
|
|
## to skip is not known before
|
|
## the loop is fully compiled
|
|
for brk in self.currentLoop.breakPos:
|
|
self.patchJump(brk)
|
|
|
|
|
|
proc assertStmt(self: Compiler, node: AssertStmt) =
|
|
## Compiles assert statements (raise
|
|
## AssertionError if the expression is falsey)
|
|
self.expression(node.expression)
|
|
self.emitByte(OpCode.Assert)
|
|
|
|
|
|
proc forEachStmt(self: Compiler, node: ForEachStmt) =
|
|
## Compiles foreach loops
|
|
# TODO
|
|
|
|
|
|
proc importStmt(self: Compiler, node: ImportStmt) =
|
|
## Imports a module at compile time
|
|
# TODO: This is obviously horrible. It's just a test
|
|
let filename = node.moduleName.token.lexeme & ".pn"
|
|
try:
|
|
self.compileModule(filename)
|
|
except IOError:
|
|
self.error(&"""could not import '{filename}': {getCurrentExceptionMsg()}""")
|
|
except OSError:
|
|
self.error(&"""could not import '{filename}': {getCurrentExceptionMsg()} [errno {osLastError()}]""")
|
|
|
|
|
|
proc statement(self: Compiler, node: Statement) =
|
|
## Compiles all statements
|
|
case node.kind:
|
|
of exprStmt:
|
|
let expression = ExprStmt(node).expression
|
|
let kind = self.inferType(expression)
|
|
self.expression(expression)
|
|
if kind.isNil():
|
|
# The expression has no type and produces no value,
|
|
# so we don't have to pop anything
|
|
discard
|
|
elif self.replMode:
|
|
case kind.kind:
|
|
of Int64:
|
|
self.emitByte(PrintInt64)
|
|
of UInt64:
|
|
self.emitByte(PrintUInt64)
|
|
of Int32:
|
|
self.emitByte(PrintInt32)
|
|
of UInt32:
|
|
self.emitByte(PrintInt32)
|
|
of Int16:
|
|
self.emitByte(PrintInt16)
|
|
of UInt16:
|
|
self.emitByte(PrintUInt16)
|
|
of Int8:
|
|
self.emitByte(PrintInt8)
|
|
of UInt8:
|
|
self.emitByte(PrintUInt8)
|
|
of Float64:
|
|
self.emitByte(PrintFloat64)
|
|
of Float32:
|
|
self.emitByte(PrintFloat32)
|
|
of Bool:
|
|
self.emitByte(PrintBool)
|
|
of Nan:
|
|
self.emitByte(PrintNan)
|
|
of Inf:
|
|
self.emitByte(PrintInf)
|
|
of String:
|
|
self.emitByte(PrintString)
|
|
else:
|
|
self.emitByte(PrintHex)
|
|
else:
|
|
self.emitByte(Pop)
|
|
of NodeKind.ifStmt:
|
|
self.ifStmt(IfStmt(node))
|
|
of NodeKind.assertStmt:
|
|
self.assertStmt(AssertStmt(node))
|
|
of NodeKind.raiseStmt:
|
|
self.raiseStmt(RaiseStmt(node))
|
|
of NodeKind.breakStmt:
|
|
self.breakStmt(BreakStmt(node))
|
|
of NodeKind.continueStmt:
|
|
self.continueStmt(ContinueStmt(node))
|
|
of NodeKind.returnStmt:
|
|
self.returnStmt(ReturnStmt(node))
|
|
of NodeKind.importStmt:
|
|
self.importStmt(ImportStmt(node))
|
|
of NodeKind.whileStmt:
|
|
# Note: Our parser already desugars
|
|
# for loops to while loops
|
|
let loop = self.currentLoop
|
|
self.currentLoop = Loop(start: self.chunk.code.len(),
|
|
depth: self.scopeDepth, breakPos: @[])
|
|
self.whileStmt(WhileStmt(node))
|
|
self.patchBreaks()
|
|
self.currentLoop = loop
|
|
of NodeKind.forEachStmt:
|
|
self.forEachStmt(ForEachStmt(node))
|
|
of NodeKind.blockStmt:
|
|
self.blockStmt(BlockStmt(node))
|
|
of NodeKind.yieldStmt:
|
|
self.yieldStmt(YieldStmt(node))
|
|
of NodeKind.awaitStmt:
|
|
self.awaitStmt(AwaitStmt(node))
|
|
of NodeKind.deferStmt:
|
|
self.deferStmt(DeferStmt(node))
|
|
of NodeKind.tryStmt:
|
|
discard
|
|
else:
|
|
self.expression(Expression(node))
|
|
|
|
|
|
proc varDecl(self: Compiler, node: VarDecl) =
|
|
## Compiles variable declarations
|
|
let expected = self.inferType(node.valueType)
|
|
let actual = self.inferType(node.value)
|
|
if expected.isNil() and actual.isNil():
|
|
if node.value.kind == identExpr or node.value.kind == callExpr and CallExpr(node.value).callee.kind == identExpr:
|
|
var name = node.value.token.lexeme
|
|
if node.value.kind == callExpr:
|
|
name = CallExpr(node.value).callee.token.lexeme
|
|
self.error(&"reference to undeclared name '{name}'")
|
|
self.error(&"'{node.name.token.lexeme}' has no type")
|
|
elif not expected.isNil() and expected.mutable: # I mean, variables *are* already mutable (some of them anyway)
|
|
self.error(&"invalid type '{self.typeToStr(expected)}' for var")
|
|
elif not self.compareTypes(expected, actual):
|
|
if not expected.isNil():
|
|
self.error(&"expected value of type '{self.typeToStr(expected)}', but '{node.name.token.lexeme}' is of type '{self.typeToStr(actual)}'")
|
|
self.expression(node.value)
|
|
self.declareName(node, mutable=node.token.kind == TokenType.Var)
|
|
self.emitByte(StoreVar)
|
|
self.emitBytes(self.getStackPos(self.names[^1]).toTriple())
|
|
|
|
|
|
proc typeDecl(self: Compiler, node: TypeDecl) =
|
|
## Compiles type declarations
|
|
# TODO
|
|
|
|
|
|
proc handleMagicPragma(self: Compiler, pragma: Pragma, node: ASTNode) =
|
|
## Handles the "magic" pragma. Assumes the given name is already
|
|
## declared
|
|
if pragma.args.len() != 1:
|
|
self.error("'magic' pragma: wrong number of arguments")
|
|
elif pragma.args[0].kind != strExpr:
|
|
self.error("'magic' pragma: wrong type of argument (constant string expected)")
|
|
elif node.kind != NodeKind.funDecl:
|
|
self.error("'magic' pragma is not valid in this context")
|
|
var node = FunDecl(node)
|
|
var fn = self.resolve(node.name)
|
|
fn.valueType.isBuiltinFunction = true
|
|
fn.valueType.builtinOp = pragma.args[0].token.lexeme[1..^2]
|
|
# The magic pragma ignores the function's body
|
|
node.body = nil
|
|
|
|
|
|
proc handlePurePragma(self: Compiler, pragma: Pragma, node: ASTNode) =
|
|
## Handles the "pure" pragma
|
|
case node.kind:
|
|
of NodeKind.funDecl:
|
|
FunDecl(node).isPure = true
|
|
of lambdaExpr:
|
|
LambdaExpr(node).isPure = true
|
|
else:
|
|
self.error("'pure' pragma is not valid in this context")
|
|
|
|
|
|
proc dispatchPragmas(self: Compiler, node: ASTnode) =
|
|
## Dispatches pragmas bound to objects
|
|
var pragmas: seq[Pragma] = @[]
|
|
case node.kind:
|
|
of NodeKind.funDecl, NodeKind.typeDecl, NodeKind.varDecl:
|
|
pragmas = Declaration(node).pragmas
|
|
of lambdaExpr:
|
|
pragmas = LambdaExpr(node).pragmas
|
|
else:
|
|
discard # Unreachable
|
|
for pragma in pragmas:
|
|
if pragma.name.token.lexeme notin self.compilerProcs:
|
|
self.error(&"unknown pragma '{pragma.name.token.lexeme}'")
|
|
self.compilerProcs[pragma.name.token.lexeme](self, pragma, node)
|
|
|
|
|
|
proc fixGenericFunc(self: Compiler, name: Name, args: seq[Expression]): Name =
|
|
## Specializes generic arguments in functions
|
|
var fn = name.deepCopy()
|
|
result = fn
|
|
var typ: Type
|
|
for i in 0..args.high():
|
|
if fn.valueType.args[i].kind.kind == Generic:
|
|
typ = self.inferType(args[i])
|
|
fn.valueType.args[i].kind = typ
|
|
self.resolve(fn.valueType.args[i].name).valueType = typ
|
|
if fn.valueType.args[i].kind.isNil():
|
|
self.error(&"cannot specialize generic function: argument {i + 1} has no type")
|
|
|
|
|
|
proc funDecl(self: Compiler, node: FunDecl, fn: Name = nil, args: seq[Expression] = @[]) =
|
|
## Compiles function declarations
|
|
#[if not node.isNil():
|
|
if node.generics.len() > 0 and fn.isNil() and args.len() == 0:
|
|
# Generic function! We can't compile it right now
|
|
self.declareName(node)
|
|
self.dispatchPragmas(node)
|
|
return]#
|
|
self.declareName(node)
|
|
self.dispatchPragmas(node)
|
|
var node = node
|
|
var fn = if fn.isNil(): self.names[^(node.arguments.len() + 1)] else: fn
|
|
var names = self.names[^(node.arguments.len())..^1]
|
|
if fn.valueType.isBuiltinFunction:
|
|
# We take the arguments off of our name list
|
|
# because they become temporaries on the stack.
|
|
# Builtin functions (usually) map to a single
|
|
# bytecode instruction to avoid unnecessary
|
|
# overhead from peon's calling convention
|
|
# This also means that peon's fast builtins
|
|
# can only be relatively simple
|
|
self.names = self.names[0..^node.arguments.len() + 1]
|
|
else:
|
|
var function = self.currentFunction
|
|
var jmp: int
|
|
# A function's code is just compiled linearly
|
|
# and then jumped over
|
|
jmp = self.emitJump(JumpForwards)
|
|
# Function's code starts after the jump
|
|
fn.codePos = self.chunk.code.len()
|
|
# We let our debugger know a function is starting
|
|
let start = self.chunk.code.high()
|
|
for name in names:
|
|
self.emitBytes([NoOp, NoOp, NoOp, NoOp])
|
|
name.codePos = self.chunk.code.len() - 4
|
|
# We store the current function
|
|
if not self.currentFunction.isNil():
|
|
self.currentFunction.valueType.childFunc = fn.valueType
|
|
self.currentFunction = fn
|
|
if node.isNil():
|
|
# We got called back with more specific type
|
|
# arguments: time to fix them!
|
|
self.currentFunction = self.fixGenericFunc(fn, args)
|
|
node = self.currentFunction.valueType.fun
|
|
elif not node.body.isNil():
|
|
if BlockStmt(node.body).code.len() == 0:
|
|
self.error("cannot declare function with empty body")
|
|
else:
|
|
discard # TODO: Forward declarations
|
|
let impl = self.findByType(fn.name.token.lexeme, fn.valueType, self.scopeDepth)
|
|
if impl.len() > 1:
|
|
# We found more than one (public) implementation of
|
|
# the same function with the same name: this is an
|
|
# error, as it would raise ambiguity when calling them
|
|
var msg = &"multiple matching implementations of '{fn.name.token.lexeme}' found:\n"
|
|
for f in reversed(impl):
|
|
msg &= &"- in module '{f.owner}' at line {f.line} of type {self.typeToStr(f.valueType)}\n"
|
|
self.error(msg)
|
|
# Since the deferred array is a linear
|
|
# sequence of instructions and we want
|
|
# to keep track to whose function's each
|
|
# set of deferred instruction belongs,
|
|
# we record the length of the deferred
|
|
# array before compiling the function
|
|
# and use this info later to compile
|
|
# the try/finally block with the deferred
|
|
# code
|
|
var deferStart = self.deferred.len()
|
|
self.beginScope()
|
|
for decl in BlockStmt(node.body).code:
|
|
self.declaration(decl)
|
|
let typ = self.currentFunction.valueType.returnType
|
|
var hasVal: bool = false
|
|
case self.currentFunction.valueType.fun.kind:
|
|
of NodeKind.funDecl:
|
|
hasVal = self.currentFunction.valueType.fun.hasExplicitReturn
|
|
of NodeKind.lambdaExpr:
|
|
hasVal = LambdaExpr(Declaration(self.currentFunction.valueType.fun)).hasExplicitReturn
|
|
else:
|
|
discard # Unreachable
|
|
if not hasVal and not typ.isNil():
|
|
# There is no explicit return statement anywhere in the function's
|
|
# body: while this is not a tremendously useful piece of information (since
|
|
# the presence of at least one doesn't mean all control flow cases are
|
|
# covered), it definitely is an error worth reporting
|
|
self.error("function has an explicit return type, but no return statement was found")
|
|
hasVal = hasVal and not typ.isNil()
|
|
self.endScope()
|
|
# Terminates the function's context
|
|
self.emitByte(OpCode.Return)
|
|
if hasVal:
|
|
self.emitByte(1)
|
|
else:
|
|
self.emitByte(0)
|
|
# Some debugging info here
|
|
self.chunk.cfi.add(start.toTriple())
|
|
self.chunk.cfi.add(self.chunk.code.high().toTriple())
|
|
self.chunk.cfi.add(uint8(node.arguments.len()))
|
|
if not node.name.isNil():
|
|
self.chunk.cfi.add(fn.name.token.lexeme.len().toDouble())
|
|
var s = fn.name.token.lexeme
|
|
if s.len() >= uint16.high().int:
|
|
s = node.name.token.lexeme[0..uint16.high()]
|
|
self.chunk.cfi.add(s.toBytes())
|
|
else:
|
|
self.chunk.cfi.add(0.toDouble())
|
|
# Currently defer is not functional, so we
|
|
# just pop the instructions
|
|
for _ in deferStart..self.deferred.high():
|
|
discard self.deferred.pop()
|
|
# Well, we've compiled everything: time to patch
|
|
# the jump offset
|
|
self.patchJump(jmp)
|
|
# Pops a call frame
|
|
# discard self.frames.pop()
|
|
# Restores the enclosing function (if any).
|
|
# Makes nested calls work (including recursion)
|
|
self.currentFunction = function
|
|
|
|
|
|
proc patchReturnAddress(self: Compiler, pos: int) =
|
|
## Patches the return address of a function
|
|
## call
|
|
let address = self.chunk.code.len().toQuad()
|
|
self.chunk.code[pos] = address[0]
|
|
self.chunk.code[pos + 1] = address[1]
|
|
self.chunk.code[pos + 2] = address[2]
|
|
self.chunk.code[pos + 3] = address[3]
|
|
|
|
|
|
proc declaration(self: Compiler, node: Declaration) =
|
|
## Compiles all declarations
|
|
case node.kind:
|
|
of NodeKind.varDecl:
|
|
self.varDecl(VarDecl(node))
|
|
of NodeKind.funDecl:
|
|
self.funDecl(FunDecl(node))
|
|
of NodeKind.typeDecl:
|
|
self.typeDecl(TypeDecl(node))
|
|
else:
|
|
self.statement(Statement(node))
|
|
|
|
|
|
proc terminateProgram(self: Compiler, pos: int, terminateScope: bool = true) =
|
|
## Utility to terminate a peon program
|
|
if terminateScope:
|
|
self.endScope()
|
|
self.patchReturnAddress(pos)
|
|
self.emitByte(OpCode.Return)
|
|
self.emitByte(0) # Entry point has no return value (TODO: Add easter eggs, cuz why not)
|
|
|
|
|
|
proc beginProgram(self: Compiler, incremental: bool = false): int =
|
|
## Utility to begin a peon program
|
|
## compiled.
|
|
## Returns a dummy return address of
|
|
## the implicit main to be patched by
|
|
## terminateProgram
|
|
|
|
# Every peon program has a hidden entry point in
|
|
# which user code is wrapped. Think of it as if
|
|
# peon is implicitly writing the main() function
|
|
# of your program and putting all of your code in
|
|
# there. While we call our entry point just like
|
|
# any regular peon function, we can't use our handy
|
|
# helper generateCall() because we need to keep track
|
|
# of where our program ends (which we don't know yet).
|
|
# To fix this, we emit dummy offsets and patch them
|
|
# later, once we know the boundaries of our hidden main()
|
|
var main: Name
|
|
if incremental:
|
|
main = self.names[0]
|
|
else:
|
|
main = Name(depth: 0,
|
|
isPrivate: true,
|
|
isConst: false,
|
|
isLet: false,
|
|
isClosedOver: false,
|
|
owner: self.currentModule,
|
|
valueType: Type(kind: Function,
|
|
name: "",
|
|
returnType: nil,
|
|
args: @[],
|
|
),
|
|
codePos: 13, # Jump address is hardcoded
|
|
name: newIdentExpr(Token(lexeme: "", kind: Identifier)),
|
|
isFunDecl: true,
|
|
line: -1)
|
|
self.names.add(main)
|
|
self.emitByte(LoadFunction)
|
|
self.emitBytes(main.codePos.toTriple())
|
|
self.emitByte(LoadReturnAddress)
|
|
self.emitBytes(0.toQuad())
|
|
self.emitByte(Call)
|
|
self.emitBytes(0.toTriple())
|
|
result = 5
|
|
|
|
|
|
proc compile*(self: Compiler, ast: seq[Declaration], file: string, lines: seq[tuple[start, stop: int]], source: string, chunk: Chunk = nil,
|
|
terminateScope: bool = true, incremental: bool = false): Chunk =
|
|
## Compiles a sequence of AST nodes into a chunk
|
|
## object
|
|
if chunk.isNil():
|
|
self.chunk = newChunk()
|
|
else:
|
|
self.chunk = chunk
|
|
self.ast = ast
|
|
self.file = file
|
|
var terminateScope = terminateScope
|
|
if incremental:
|
|
terminateScope = false
|
|
self.scopeDepth = 0
|
|
self.currentFunction = nil
|
|
self.currentModule = self.file.extractFilename()
|
|
self.current = 0
|
|
self.frames = @[0]
|
|
self.lines = lines
|
|
self.source = source
|
|
let pos = self.beginProgram(incremental)
|
|
if incremental and self.replMode:
|
|
for i in countup(1, 2):
|
|
discard self.chunk.code.pop()
|
|
while not self.done():
|
|
self.declaration(Declaration(self.step()))
|
|
self.terminateProgram(pos, terminateScope)
|
|
result = self.chunk
|
|
if incremental and not self.replMode:
|
|
for i in countup(1, 2):
|
|
discard self.chunk.code.pop()
|
|
|
|
|
|
proc compileModule(self: Compiler, filename: string) =
|
|
## Compiles an imported module into an existing chunk.
|
|
## A temporary compiler object is initialized internally
|
|
let path = joinPath(splitPath(self.file).head, filename)
|
|
if self.modules.contains(path):
|
|
return
|
|
var lexer = newLexer()
|
|
var parser = newParser()
|
|
var compiler = newCompiler()
|
|
lexer.fillSymbolTable()
|
|
let source = readFile(joinPath(splitPath(self.file).head, filename))
|
|
let tokens = lexer.lex(source, filename)
|
|
let ast = parser.parse(tokens, filename, lexer.getLines(), source)
|
|
compiler.names.add(self.names[0])
|
|
discard compiler.compile(ast, filename, lexer.getLines(), source, chunk=self.chunk, incremental=true)
|
|
for name in compiler.names:
|
|
if name.owner in self.modules:
|
|
continue
|
|
self.names.add(name)
|
|
self.modules.incl(path)
|
|
self.closedOver &= compiler.closedOver
|