从零写一个编译器（九）：语义分析之构造抽象语法树(AST)

项目的完整代码在 C2j-Compiler

前言

在上一篇完成了符号表的构建，下一步就是输出抽象语法树(Abstract Syntax Tree，AST)

抽象语法树（abstract syntax tree 或者缩写为 AST），是源代码的抽象语法结构的树状表现形式，这里特指编程语言的源代码。树上的每个节点都表示源代码中的一种结构。

AST对于编译器是至关重要的，现在的编译型语言一般通过AST来生成IR，解释型语言也可以不用虚拟机而直接遍历AST来解释执行，之后要写解释器和编译器都依赖这个AST

这一篇主要文件有：

• AstBuilder.java
• AstNode.java
• AstNodeImpl.java
• NodeKey.java
• NodeFactory.java

主要数据结构

AST节点的表示

public interface AstNode {

    AstNode addChild(AstNode node);

    AstNode getParent();

    ArrayList<AstNode> getChildren();

    void setAttribute(NodeKey key, Object value);

    Object getAttribute(NodeKey key);

    boolean isChildrenReverse();

    void reverseChildren();

    AstNode copy();

}

这是对AstNode接口的实现，并且继承HashMap，这里的NodeKey是

TokenType, VALUE, SYMBOL, PRODUCTION, TEXT

对应的value，

1. TokenType就是非终结符的类型
2. Text用来存储解析对象的文本信息
3. Symbol对应的就是变量的符号对象
4. Value是对应对象解析的值，比如int a = 1，那么value的值就为1

public class AstNodeImpl extends HashMap<NodeKey, Object> implements AstNode {
    private Token type;
    private AstNodeImpl parent;
    private ArrayList<AstNode> children;
    String name;

    private boolean isChildrenReverse = false;

    public AstNodeImpl(Token type) {
        this.type = type;
        this.parent = null;
        this.children = new ArrayList<>();
        setAttribute(NodeKey.TokenType, type);
    }

    @Override
    public AstNode addChild(AstNode node) {
        if (node != null) {
            children.add(node);
            ((AstNodeImpl) node).parent = this;
        }

        return node;
    }

    @Override
    public AstNode getParent() {
        return parent;
    }

    @Override
    public void reverseChildren() {
        if (isChildrenReverse) {
            return;
        }

        Collections.reverse(children);
        isChildrenReverse = true;
    }

    @Override
    public boolean isChildrenReverse() {
        return isChildrenReverse;
    }

    @Override
    public ArrayList<AstNode> getChildren() {
        reverseChildren();

        return children;
    }

    @Override
    public void setAttribute(NodeKey key, Object value) {
        if (key == NodeKey.TEXT) {
            name = (String) value;
        }
        put(key, value);
    }

    @Override
    public Object getAttribute(NodeKey key) {
        return get(key);
    }

    @Override
    public String toString() {
        String info = "";
        if (get(NodeKey.VALUE) != null) {
            info += "Node Value is " + get(NodeKey.VALUE).toString();
        }

        if (get(NodeKey.TEXT) != null) {
            info += "
Node Text is " + get(NodeKey.TEXT).toString();
        }

        if (get(NodeKey.SYMBOL) != null) {
            info += "
Node Symbol is " + get(NodeKey.SYMBOL).toString();
        }

        return info + "
 Node Type is " + type.toString();
    }

    @Override
    public AstNode copy() {
        AstNodeImpl copy = (AstNodeImpl) NodeFactory.createICodeNode(type);
        Set<Entry<NodeKey, Object>> attributes = entrySet();
        Iterator<Map.Entry<NodeKey, Object>> it = attributes.iterator();

        while (it.hasNext()) {
            Map.Entry<NodeKey, Object> attribute = it.next();
            copy.put(attribute.getKey(), attribute.getValue());
        }

        return copy;
    }
}

NodeFactory

NodeFactory就是简单的返回一个节点的实现

public class NodeFactory {
    public static AstNode createICodeNode(Token type) {
        return new AstNodeImpl(type);
    }
}

构造AST

AST的创建也是需要在语法分析过程中根据reduce操作进行操作的。也就是在takeActionForReduce方法中调用AstBuilder的buildSyntaxTree方法

在AstBuilder里面还是需要两个堆栈来辅助操作

private Stack<AstNode> nodeStack = new Stack<>();

private LRStateTableParser parser = null;
private TypeSystem typeSystem = null;
private Stack<Object> valueStack = null;
private String functionName;
private HashMap<String, AstNode> funcMap = new HashMap<>();

private static AstBuilder instance;

构造AST的主要逻辑在buildSyntaxTree方法里，需要注意的是有一些节点在解释执行和代码生成的时候是不一样的，有时代码生成需要的节点解释执行的话并不需要

在这里提一下UNARY这个非终结符，这个非终结符和NAME很像，但是它一般是代表进行运算和一些操作的时候，比如数组，++，--或者函数调用的时候

其实构建AST的过程和符号表的构建过程有点儿类似，都是根据reduce操作来创建信息和组合信息，符号表是组合修饰符说明符等，而AST则是组合节点间的关系变成一棵树

我们只看几个操作

• Specifiers_DeclList_Semi_TO_Def

这个节点需要注意的是，从堆栈的什么地方拿到Symbol，这个需要从reduce次数和推导式中得出

* DEF -> SPECIFIERS  DECL_LIST SEMI
* DECL -> VAR_DECL
* VAR_DECL -> NEW_NAME
*             | VAR_DECL LP RP
*             | VAR_DECL LP VAR_LIST RP
*             | LP VAR_DECL RP
*             | START VAR_DECL 

从推导式可以看出，DEF节点的符号应该在valueStack.size() - 3，但是DECL和VAR_DECL没有做reduce操作，所以符号应该在valueStack.size() - 2。这其实和前面的符号表构建算出之前符号的位置是一样的。

• TO_UNARY

这里则是变量、数字或者字符串的节点，如果是个变量的号，这个节点就需要一个Symbol的value了

case SyntaxProductionInit.Number_TO_Unary:
case SyntaxProductionInit.Name_TO_Unary:
case SyntaxProductionInit.String_TO_Unary:
    node = NodeFactory.createICodeNode(Token.UNARY);
    if (production == SyntaxProductionInit.Name_TO_Unary) {
        assignSymbolToNode(node, text);
    }

    node.setAttribute(NodeKey.TEXT, text);
    break;

其余的节点无非是把一些语句拆分它的逻辑然后组成节点，真正的求值部分像Name_TO_Unary比较少，更多是比如把一个if else块分成if节点、判断节点、else节点，之后再按照这棵树进行解释执行或者代码生成

public AstNode buildSyntaxTree(int production, String text) {
    AstNode node = null;
    Symbol symbol = null;
    AstNode child = null;

    if (Start.STARTTYPE == Start.INTERPRETER) {
        int p1 = SyntaxProductionInit.Specifiers_DeclList_Semi_TO_Def;
        int p2 = SyntaxProductionInit.Def_To_DefList;
        int p3 = SyntaxProductionInit.DefList_Def_TO_DefList;

        boolean isReturn = production == p1 || production == p2 || production == p3;
        if (isReturn) {
            return null;
        }
    }

    switch (production) {
        case SyntaxProductionInit.Specifiers_DeclList_Semi_TO_Def:
            node = NodeFactory.createICodeNode(Token.DEF);
            symbol = (Symbol) valueStack.get(valueStack.size() - 2);
            node.setAttribute(NodeKey.SYMBOL, symbol);
            break;
        case SyntaxProductionInit.Def_To_DefList:
            node = NodeFactory.createICodeNode(Token.DEF_LIST);
            node.addChild(nodeStack.pop());
            break;
        case SyntaxProductionInit.DefList_Def_TO_DefList:
            node = NodeFactory.createICodeNode(Token.DEF_LIST);
            node.addChild(nodeStack.pop());
            node.addChild(nodeStack.pop());
            break;

        case SyntaxProductionInit.Number_TO_Unary:
        case SyntaxProductionInit.Name_TO_Unary:
        case SyntaxProductionInit.String_TO_Unary:
            node = NodeFactory.createICodeNode(Token.UNARY);
            if (production == SyntaxProductionInit.Name_TO_Unary) {
                assignSymbolToNode(node, text);
            }

            node.setAttribute(NodeKey.TEXT, text);
            break;

        case SyntaxProductionInit.Unary_LP_RP_TO_Unary:
            node = NodeFactory.createICodeNode(Token.UNARY);
            node.addChild(nodeStack.pop());
            break;

        case SyntaxProductionInit.Unary_LP_ARGS_RP_TO_Unary:
            node = NodeFactory.createICodeNode(Token.UNARY);
            node.addChild(nodeStack.pop());
            node.addChild(nodeStack.pop());
            break;

        case SyntaxProductionInit.Unary_Incop_TO_Unary:
        case SyntaxProductionInit.Unary_DecOp_TO_Unary:
        case SyntaxProductionInit.LP_Expr_RP_TO_Unary:
        case SyntaxProductionInit.Start_Unary_TO_Unary:
            node = NodeFactory.createICodeNode(Token.UNARY);
            node.addChild(nodeStack.pop());
            break;

        case SyntaxProductionInit.Unary_LB_Expr_RB_TO_Unary:
            node = NodeFactory.createICodeNode(Token.UNARY);
            node.addChild(nodeStack.pop());
            node.addChild(nodeStack.pop());

            break;

        case SyntaxProductionInit.Uanry_TO_Binary:
            node = NodeFactory.createICodeNode(Token.BINARY);
            child = nodeStack.pop();
            node.setAttribute(NodeKey.TEXT, child.getAttribute(NodeKey.TEXT));
            node.addChild(child);
            break;

        case SyntaxProductionInit.Binary_TO_NoCommaExpr:
        case SyntaxProductionInit.NoCommaExpr_Equal_NoCommaExpr_TO_NoCommaExpr:
            node = NodeFactory.createICodeNode(Token.NO_COMMA_EXPR);
            child = nodeStack.pop();
            String t = (String) child.getAttribute(NodeKey.TEXT);
            node.addChild(child);
            if (production == SyntaxProductionInit.NoCommaExpr_Equal_NoCommaExpr_TO_NoCommaExpr) {
                child = nodeStack.pop();
                t = (String) child.getAttribute(NodeKey.TEXT);
                node.addChild(child);
            }
            break;

        case SyntaxProductionInit.Binary_Plus_Binary_TO_Binary:
        case SyntaxProductionInit.Binary_DivOp_Binary_TO_Binary:
        case SyntaxProductionInit.Binary_Minus_Binary_TO_Binary:
        case SyntaxProductionInit.Binary_Start_Binary_TO_Binary:
            node = NodeFactory.createICodeNode(Token.BINARY);
            node.addChild(nodeStack.pop());
            node.addChild(nodeStack.pop());
            break;

        case SyntaxProductionInit.Binary_RelOP_Binary_TO_Binray:
            node = NodeFactory.createICodeNode(Token.BINARY);
            node.addChild(nodeStack.pop());

            AstNode operator = NodeFactory.createICodeNode(Token.RELOP);
            operator.setAttribute(NodeKey.TEXT, parser.getRelOperatorText());
            node.addChild(operator);

            node.addChild(nodeStack.pop());

            break;

        case SyntaxProductionInit.NoCommaExpr_TO_Expr:
            node = NodeFactory.createICodeNode(Token.EXPR);
            node.addChild(nodeStack.pop());
            break;

        case SyntaxProductionInit.Expr_Semi_TO_Statement:
        case SyntaxProductionInit.CompountStmt_TO_Statement:
            node = NodeFactory.createICodeNode(Token.STATEMENT);
            node.addChild(nodeStack.pop());
            break;

        case SyntaxProductionInit.LocalDefs_TO_Statement:
            node = NodeFactory.createICodeNode(Token.STATEMENT);
            if (Start.STARTTYPE == Start.CODEGEN) {
                node.addChild(nodeStack.pop());
            }
            break;

        case SyntaxProductionInit.Statement_TO_StmtList:
            node = NodeFactory.createICodeNode(Token.STMT_LIST);
            if (nodeStack.size() > 0) {
                node.addChild(nodeStack.pop());
            }
            break;

        case SyntaxProductionInit.FOR_OptExpr_Test_EndOptExpr_Statement_TO_Statement:
            node = NodeFactory.createICodeNode(Token.STATEMENT);
            node.addChild(nodeStack.pop());
            node.addChild(nodeStack.pop());
            node.addChild(nodeStack.pop());
            node.addChild(nodeStack.pop());
            break;

        case SyntaxProductionInit.StmtList_Statement_TO_StmtList:
            node = NodeFactory.createICodeNode(Token.STMT_LIST);
            node.addChild(nodeStack.pop());
            node.addChild(nodeStack.pop());
            break;

        case SyntaxProductionInit.Expr_TO_Test:
            node = NodeFactory.createICodeNode(Token.TEST);
            node.addChild(nodeStack.pop());
            break;

        case SyntaxProductionInit.If_Test_Statement_TO_IFStatement:
            node = NodeFactory.createICodeNode(Token.IF_STATEMENT);
            node.addChild(nodeStack.pop());
            node.addChild(nodeStack.pop());
            break;

        case SyntaxProductionInit.IfElseStatemnt_Else_Statemenet_TO_IfElseStatement:
            node = NodeFactory.createICodeNode(Token.IF_ELSE_STATEMENT);
            node.addChild(nodeStack.pop());
            node.addChild(nodeStack.pop());
            break;

        case SyntaxProductionInit.While_LP_Test_Rp_TO_Statement:
        case SyntaxProductionInit.Do_Statement_While_Test_To_Statement:
            node = NodeFactory.createICodeNode(Token.STATEMENT);
            node.addChild(nodeStack.pop());
            node.addChild(nodeStack.pop());
            break;

        case SyntaxProductionInit.Expr_Semi_TO_OptExpr:
        case SyntaxProductionInit.Semi_TO_OptExpr:
            node = NodeFactory.createICodeNode(Token.OPT_EXPR);
            if (production == SyntaxProductionInit.Expr_Semi_TO_OptExpr) {
                node.addChild(nodeStack.pop());
            }
            break;

        case SyntaxProductionInit.Expr_TO_EndOpt:
            node = NodeFactory.createICodeNode(Token.END_OPT_EXPR);
            node.addChild(nodeStack.pop());
            break;

        case SyntaxProductionInit.LocalDefs_StmtList_TO_CompoundStmt:
            node = NodeFactory.createICodeNode(Token.COMPOUND_STMT);
            node.addChild(nodeStack.pop());
            break;

        case SyntaxProductionInit.NewName_LP_RP_TO_FunctDecl:
        case SyntaxProductionInit.NewName_LP_VarList_RP_TO_FunctDecl:
            node = NodeFactory.createICodeNode(Token.FUNCT_DECL);
            node.addChild(nodeStack.pop());
            child = node.getChildren().get(0);
            functionName = (String) child.getAttribute(NodeKey.TEXT);
            symbol = assignSymbolToNode(node, functionName);

            break;

        case SyntaxProductionInit.NewName_TO_VarDecl:
            nodeStack.pop();
            break;

        case SyntaxProductionInit.NAME_TO_NewName:
            node = NodeFactory.createICodeNode(Token.NEW_NAME);
            node.setAttribute(NodeKey.TEXT, text);
            break;

        case SyntaxProductionInit.OptSpecifiers_FunctDecl_CompoundStmt_TO_ExtDef:
            node = NodeFactory.createICodeNode(Token.EXT_DEF);
            node.addChild(nodeStack.pop());
            node.addChild(nodeStack.pop());
            funcMap.put(functionName, node);
            break;

        case SyntaxProductionInit.NoCommaExpr_TO_Args:
            node = NodeFactory.createICodeNode(Token.ARGS);
            node.addChild(nodeStack.pop());
            break;

        case SyntaxProductionInit.NoCommaExpr_Comma_Args_TO_Args:
            node = NodeFactory.createICodeNode(Token.ARGS);
            node.addChild(nodeStack.pop());
            node.addChild(nodeStack.pop());
            break;

        case SyntaxProductionInit.Return_Semi_TO_Statement:
            node = NodeFactory.createICodeNode(Token.STATEMENT);
            break;
        case SyntaxProductionInit.Return_Expr_Semi_TO_Statement:
            node = NodeFactory.createICodeNode(Token.STATEMENT);
            node.addChild(nodeStack.pop());
            break;

        case SyntaxProductionInit.Unary_StructOP_Name_TO_Unary:
            node = NodeFactory.createICodeNode(Token.UNARY);
            node.addChild(nodeStack.pop());
            node.setAttribute(NodeKey.TEXT, text);
            break;

        default:
            break;
    }

    if (node != null) {
        node.setAttribute(NodeKey.PRODUCTION, production);
        nodeStack.push(node);
    }

    return node;
}

小结

其实构造AST和创建符号表上非常相似，都是依据reduce操作的信息来完成。在AST的构建中的主要任务就是对源代码语句里的逻辑进行分块，比如对于一个ifelse语句：

上面的图是我依据这个意思话的，和上面构造出来的AST不完全一致

另外我的github博客：https://dejavudwh.cn/