In this chaper I'll show how I would make a parser.
A parser, in addition to our lexer, transforms the input program as text, meaning an unstructured sequence of characters, into a structered representation. Structured meaning the representation tells us about the different constructs such as if statements and expressions.
Both `Stmt` (statement) and `Expr` (expression) are polymorphic types, meaning an expression, for example, can be either an addition operation containing 2 inner expressions or an integer expression containing the integer value, etc. This can also be implemented with classes and sub classes.
For both `Stmt` and `Expr` there's an error-kind. This makes the parser simpler, as we won't need to manage parsing failures differently than successful parslings.
To start, we'll implement a `Parser` class, which for now is simply a consumer of a token iterater, meaning the lexer. In simple terms, whereas the lexer is a transformation from text to tokens, the parser is a transformation from token to an AST, except that the parser is not an iterator.
This implementation should look familiar compared to the lexer. We use the `currentToken` as a 'buffer', and then just use the `.next()` on the `lexer`.
Just as the lexer, we'll have a `.pos()` method, returning the current position.
For convenience, although there are other ways of doing it, we'll implement another public method on `Lexer`, which will return the lexer's current position.
```ts
class Lexer {
// ...
public currentPos(): Pos { return this.pos(); }
// ...
}
```
The reason, is that when the lexer has reached the end of the file, the `.next()` method will return `null` instead of a token with a position, meaning we won't get the position after the last token.
The parser does not need to keep track of `index`, `line` and `col` as those are stored in the tokens. The token's position is prefered to the lexer's.
When testing, we first check that we have not reach the end. Either we have to do that here, or the caller will have to write something like `!this.done() && this.test(...)`, and it's easy to do it here.
Operands are the individual parts of an operation. For example, in the math expression `a + b`, (would be `+ a b` in the input language), `a` and `b` are the *operands*, while `+` is the *operator*. In the expression `a + b * c`, the operands are `a`, `b` and `c`. But in the expression `a * (b + c)`, the operands of the multiply operation are `a` and `(b + c)`. `(b + c)` is an operands, because it is enclosed on both sides. This is how we'll define operands.
We'll make a public method in `Parser` called `parseOperand`.
Identifiers and literals (integers, strings) are single token constructs, meaning the parsing consists of translating a token into an ast-node with the value.
A group expression is an expression enclosed in parenthesis, eg `(1 + 2)`. Because the expression is enclosed, meaning starts with a `(`-token and ends with a `)`-token, we will treat is like an operand.
```ts
type ExprKind =
// ...
| { type: "group", expr: Expr }
// ...
;
```
If we find a `(`-token in `.parseOperand()`, we know that we should parse a group expression. We do this by ignoring the `(`-token, parsing an expression using `.parseExpr()` and checking that we find a `)`-token afterwards.
```ts
class Parser {
// ...
public parseOperand(): Expr {
// ...
if (this.test("(")) {
this.step();
const expr = this.parseExpr();
if (!this.test(")")) {
this.report("expected ')'");
return { kind: { type: "error" }, pos };
}
this.step();
return { kind: { type: "group", expr }, pos };
}
// ...
}
// ...
}
```
If we do not find the closing `)`-token, we report an error and return an error expression.
### 3.3.3 Block, if and loop operands
We want to be able to use blocks, if and loop constructs as expressions.
Example:
```rs
let temperature_feeling = if > temperature 20 { "hot" } else { "cold" };
```
Each construct will have their own `.parse...()`-method, so we'll just look for the first `{`-, `if`-, or `loop`-token and call the relevant method.
```ts
class Parser {
// ...
public parseOperand(): Expr {
// ...
if (this.test("{"))
return this.parseBlock();
if (this.test("if"))
return this.parseIf();
if (this.test("loop"))
return this.parseLoop();
// ...
}
// ...
}
```
## 3.4 Postfix operators
Postfix operations are expressions were the operators come after the subject expression. This includes field expressions (eg. `subject.field`), index expressions (eg. `subject[index]`) and call expressions (eg. `subject(...args)`).
A notable detail, is that postfix operations are chainable, eg. `subject[index].field` is valid, likewise with `subject.method(arg)` and `matrix[y][x]`.
We'll make a method `.parsePostfix()` to parse postfix operators.
```ts
class Parser {
// ...
public parsePostfix(): Expr {
let subject = this.parseOperand();
while (true) {
const pos = this.pos();
// ...
break;
}
return subject;
}
// ...
}
```
We start by parsing an operand. Then we enter a loop, which runs until we no longer find any relevant operator tokens. When we parse a postfix expression, the `subject` will be replaced with the new parsed expression.
Notice we don't define `pos` at the start, but after we've parsed the subject. That's because we want `pos` to the reflect the start of the postfix operator, not the start of the subject.
### 3.4.1 Field expression
A field expression is for accessing fields on an object, and consists of a `.`-token and an identifier, eg. `.field`.
If we find a `.`-token, we step over it, and make sure that we've hit an identifier. We save the identifier value and step over the identifier. Then we replace `subject` with a new field expression containing the previous `subject` value. Then we continue to look for the next postfix operator.
### 3.4.2 Index expression
An index operation consists of the subject and an index. The index is an expression, and it is contained in `[`- and `]`-tokens, eg. `subject[value]`.
If we find a `[`-token, we parse the index part exactly the same way, we parse a group expression.
### 3.4.3 Call expression
A call expression is like an index expression, except that it uses `(` and `)` instead of `[` and `]` and that there can be 0 or more expressions (arguments or args) inside the `(` and `)`. The arguments are seperated by `,`.
Similarly to index epxressions, if we find a `(`-token, we step over it, parse the arguments, check for a `)` and replace `subject` with a call expression containing the previous `subject`.
When parsing the arguments, we start by testing if we've reached a `)` to check if there are any arguments. If not, we parse the first argument.
The consecutive arguments are all preceded by a `,`-token. There we test or `,`, to check if we should keep parsing arguments. After checking for a seperating `,`, we check if we've reached a `)` and break if so. This is to allow for trailing comma.
Contrasting postfix expressions, prefix expression are operations where the operator comes first, then the operands are listed. In some languages, operations such as negation (eg. `-value`) and not-operations (eg. `!value`) are prefix operations. In the language we're making, all binary and unary arithmetic operations are prefix. This includes both expressions with a single operand, such as not (eg. `not value`), but also expressions with 2 operands, such ass addition (eg. `+ a b`) and equation (eg. `== a b`).
This is because infix operators (eg. `a + b`) makes parsing more complicated, as it requires reasoning about operator precedence, eg. why `2 + 3 * 4 != (2 + 3) * 4`.
Operations with 1 operand are called unary expression. Operations with 2 are called binary expressions.
Just as with unary, if we find a `+`-token, we ignore it and parse prefix expression recursively. Then we parse the second operand, by parsing another prefix expressions. And then we return a binary expression with the `left` and `right` operands and the binary type.
An if-expression is an `if`-token, followed by an condition in the form of an expressions, followed by a block-expression, optionally followed by an `else`-token and another block.
When parsing an if-expression, we assume we already have reached an `if`-token.
We skip the `if`-token. Then we parse the condition expression `cond`. Then we check for a `{`-token and parse block. Then we check for an `else`-token. If not present, we return an if-expression with no `falsy`-option. Else we skip the `else`-token, check for and parse the `falsy`-block, and return the if-expression with the `falsy`-option.
## 3.8 Loop expressions
A loop expression is a `loop`-token followed by a block expression.
```ts
type ExprKind =
// ...
| { type: "loop", body: Expr }
// ...
;
```
```ts
class Parser {
// ...
public parseLoop(): Expr {
const pos = this.pos();
this.step();
if (!this.test("{")) {
this.report("expected '}'");
return { kind: { type: "error" }, pos };
}
const body = this.parseExpr();
return { kind: { type: "loop", body }, pos };
}
// ...
}
```
We again assume, we've already hit a `loop`-token, which we step over. Then we check for and parse a block expression constituting the loop `body`. We then return the loop expression.
## 3.9 Break statements
A break statement consists of a `break`-token and an optional expression.
The language will allow us to use loop as an expression. The break statement will then deliver the resulting value, eg.
```rs
let value = loop {
let value = get_value();
if acceptable(value) {
break value;
}
};
```
```ts
type StmtKind =
// ...
| { type: "break", expr?: Expr }
// ...
;
```
```ts
class Parser {
// ...
public parseBreak(): Stmt {
const pos = this.pos();
this.step();
if (!this.test(";")) {
return { kind: { type: "break" }, pos };
}
const expr = this.parseExpr();
return { kind: { type: "break", expr }, pos };
}
// ...
}
```
We assume we've already hit a `break`-token and step over it. Then we check if we've hit a `;`-token. If so, there's no expression, so we return a break statement with no expression. If there's no `;`-token, we parse an expression and return a break statement with it.
## 3.9 Return statements
The return statement is for functions what break is for loops.
We first step over the initial `fn`-token. Then we grap the value of an `ident`-token. Then we check for a `(` and call `.parseFnParams()` to parse the parameters, including the encapsulating `(` and `)`. Then we check for and parse a block. And then we return the statement.
Then we define the `.parseFnParams()` method.
```ts
class Parser {
// ...
public parseFnParams(): Param[] {
this.step();
if (this.test(")")) {
this.step();
return [];
}
let params: Param[] = [];
const paramResult = this.parseParam();
if (!paramResult.ok)
return [];
params.push(paramResult.value);
while (this.test(",")) {
this.step();
if (this.test(")"))
break;
const paramResult = this.parseParam();
if (!paramResult.ok)
return [];
params.push(paramResult.value);
}
if (!this.test(")")) {
this.report("expected ')'");
return params;
}
this.step();
return params;
}
// ...
}
```
We represent the parameter list as an array of params. We start by stepping over the `(`-token. If we immediately find a `)`-token, we step and return an empty array. Else we start collecting parameters. The parsing is similar to function call arguments, wherein we parse the first, then keep parsing until we no longer find a `,`-token or find a `)`-token right after a `,`. When done, we check we've reached a `)`, step and return the parameters. If `.parseParam()` fails, we return an empty array.
We then need to define the `.parseParam()` method.
We look for an identifier, and return a parameter with its value. If we don't find an identifier, we return a failed result.
## 3.10 Let statements
A let statement declares a variable. A let statement consists of a `let`-token, a parameter, a `=`-token, and an expression. The expression is the initial value of the variable.
We step over the first `let`-token. Then we parse a parameter using the `.parseParam()` method. If it fails, we return an error statement. Then we check for and step over a `=`-token. We then parse an expressions. And lastly return a let statement with the `ident` and `value`.
We start by parsing an expression. If we do not reach a `=` afterwards, we return an expression statement. Else, we step, parse another expression and return an assignment statement.
## 3.13 Block expressions
Block expressions are statements and an optional expressions surrounded by `{` and `}`. Statements are terminated by `;`, if they don't end in `}`. If an expression is given as the last item without a terminating `;`, it is the blocks resulting value.
