erg/doc/zh_CN/compiler/hir.md

148 lines
4.3 KiB
Markdown

# 高级中间表示(HIR, High-level Intermediate Representation)
[![badge](https://img.shields.io/endpoint.svg?url=https%3A%2F%2Fgezf7g7pd5.execute-api.ap-northeast-1.amazonaws.com%2Fdefault%2Fsource_up_to_date%3Fowner%3Derg-lang%26repos%3Derg%26ref%3Dmain%26path%3Ddoc/EN/compiler/hir.md%26commit_hash%3D8673a0ce564fd282d0ca586642fa7f002e8a3c50)](https://gezf7g7pd5.execute-api.ap-northeast-1.amazonaws.com/default/source_up_to_date?owner=erg-lang&repos=erg&ref=main&path=doc/EN/compiler/hir.md&commit_hash=8673a0ce564fd282d0ca586642fa7f002e8a3c50)
HIR 是 Erg 编译器从 AST 生成的结构。
此结构包含源代码中每个表达式的完整类型信息,并且在语法上已脱糖。
AST 与源代码(作为纯文本)具有一一对应的关系,但是 HIR 去除了不必要的代码信息并添加了省略的类型信息,因此将 HIR 转换回源代码是困难的。让我们看下面代码中的 HIR 示例。
```python
v = ![]
for! 0..10, i =>
v.push! i
log v.sum()
```
从此代码生成的 AST 如下所示:
```python
AST(Module[
VarDef{
sig: VarSignature{
pat: VarPattern::Ident(None, VarName("v")),
spec_t: None,
},
op: "=",
body: Block[
UnaryOp{
op: "!",
expr: Array([]),
},
],
},
Call{
obj: Accessor::Local("for!"),
args: [
BinOp{
op: "..",
lhs: Literal(0),
rhs: Literal(10),
},
Lambda{
sig: LambdaSignature{
params: [
ParamSignature{
pat: ParamPattern::Name(VarName("i")),
},
],
spec_ret_t: None,
},
body: Block[
Call{
obj: Accessor::Attr{"v", "push!"},
args: [
Accessor::Local("i"),
],
},
],
},
],
},
Call{
obj: Accessor::Local("log"),
args: [
Call{
obj: Accessor::Attr("v", "sum"),
args: [],
}
],
}
])
```
从 AST 生成的 HIR 如下所示:
```python
HIR(Module[
VarDef{
sig: VarSignature{
pat: VarPattern::Ident(None, Name("v")),
t: [0..10, _]!,
},
op: "=",
body: Block[
expr: UnaryOp{
op: "!",
expr: Array([]),
t: [0..10, 0]!,
},
],
},
Call{
obj: Accessor::Local{
name: "for!",
t: (Range Nat, Nat => NoneType) => NoneType,
},
args: [
BinOp{
op: "..",
lhs: Literal(0),
rhs: Literal(10),
t: 0..10,
},
Lambda{
sig: LambdaSignature{
params: [
ParamSignature{
pat: ParamPattern::Name(Name("i")),
t: 0..10,
},
],
t: 0..10 => NoneType,
},
body: Block[
Call{
obj: Accessor::Attr{
obj: Accessor::Local("v"),
field: "push!",
t: Ref!(Self![T ~> T, N ~> N+1]).(Nat) => NoneType,
},
args: [
Accessor::Local("i"),
],
},
],
},
],
},
Call{
obj: Accessor::Local{
name: "log",
*t: Obj => NoneType,
},
args: [
Call{
obj: Accessor::Attr{
obj: Accessor::Local("v"),
field: "sum",
t: [0..10, !_] -> Nat
},
args: [],
t: Nat
}
],
}
])
```
对象类型推断尽可能地准确。另一方面,子程序会推断存在实现的类型。因此,实际参数的类型和形式参数的类型可能不匹配。