Erythro/src/Statements.c

/*************/
/*GEMWIRE    */
/*    ERYTHRO*/
/*************/

#include <Defs.h>
#include <Data.h>


/*
 * Turn a token type into its appropriate
 *  primitive type. 
 * 
 * This is where we do redirections like:
 *  short -> s16
 *  long -> s64
 *  int -> s32
 *  char -> u8
 * 
 */

static int TypeSize[9] = { 0,      1,      4,     8,      0,      8,         8,        8,         8}; // in BYTES

int PrimitiveSize(int Type) {
    if(Type < RET_NONE || Type > PTR_VOID)
        DieDecimal("Checking size of bad data type", Type);
    
    return TypeSize[Type];
}

/*
 * Given two types, determine if they are compatible.
 * 
 *  Depending on the value of STRICT, it will try to
 *      fit the right value into the left value.
 *  
 *  This is valid, for ie. a char into an int, as int is larger than char.
 *      This is called widening the char.
 * 
 *  If STRICT is set, it will only allow widening the left to the right.
 *      This means you cannot `char a; int b; b = 15000; a = b;`
 *      As this would shrink the int and lose resolution.
 *  
 *  NOTE: THIS IS NOT THE DEFAULT BEHAVIOUR
 *      By default, you CAN shrink an int into a char, a la shifting down.
 * 
 * 
 */

int TypesCompatible(int* Left, int* Right, int STRICT) {

    int LeftSize, RightSize;

    // Same types are compatible. No shrinking required
    if(*Left == *Right) {
        *Left = *Right = 0;
        return 1;
    }

    LeftSize = PrimitiveSize(*Left);
    RightSize = PrimitiveSize(*Right);


    // Types of size 0 are incompatible
    if((LeftSize == 0) || (RightSize == 0))
        return 0;


    /*  char x;
     *  int y;
     *  y = 15;
     * 
     *  x = y;
     *      x needs to be widened, y copied in, then x shrunk back down
     *      AKA, the left must be widened.
     */
    if(LeftSize < RightSize) {
        *Left = OP_WIDEN;
        *Right = 0;
        return 1;
    }

    /*
     * char x;
     * int y;
     * 
     * x = 15;
     * 
     * y = x;
     *      x must be widened to fit into y.
     *      if STRICT mode, this is not allowed.
     *      By default, this is valid.
     * 
     */

    if(LeftSize > RightSize) {
        if(STRICT)
            return 0; // Not compatible if STRICT
        
        *Left = 0;
        *Right = OP_WIDEN;
        return 1; // Compatible by default
    }

    /*
     * Any other cases left, by default, are compatible.
     * 
     */

    *Left = *Right = 0;
    return 1;

}

/*
 * Handles the declaration of a type of a variable.
 *  int newVar;
 * 
 * It verifies that we have the `int` keyword followed by a
 *  unique, non-keyword identifier.
 * 
 * It then stores this variable into the symbol table.
 * 
 * //TODO: Assemble this into the symbol table.
 * //TODO: int i = 5;
 * 
 */
void BeginVariableDeclaration(int Type) {
    int ID;
    //printf("type: %s\n", Types[Type]);
    
    while(1) {
    //printf("Identifier: %s\n", CurrentIdentifier);
        printf("Adding symbol %s of type %s.\n", CurrentIdentifier, TypeNames[Type]);
        ID = AddSymbol(CurrentIdentifier, Type, ST_VAR);
        AsNewSymb(ID);

        if(CurrentToken.type == LI_SEMIC) {
            printf("\tEnd of variables to declare in this statement\n");
            Tokenise(&CurrentToken);
            return;
        }

        if(CurrentToken.type == LI_COM) {
            printf("\t\tStatement appears to continue. Integrating new variable.\n");
            Tokenise(&CurrentToken);
            VerifyToken(TY_IDENTIFIER, "ident");
            continue;
        }

        DieDecimal("\nExpected ; after identifier, found", CurrentToken.type);
    }
}

struct ASTNode* ParseFunction(int Type) {
    struct ASTNode* Tree;
    struct ASTNode* FinalStatement;
    int SymbolSlot, BreakLabel;

    printf("\nIdentified function %s\n", CurrentIdentifier);

    BreakLabel = NewLabel();

    SymbolSlot = AddFunctionSymbol(CurrentIdentifier, Type, ST_FUNC, BreakLabel);
    CurrentFunction = SymbolSlot;

    VerifyToken(LI_LPARE, "(");
    VerifyToken(LI_RPARE, ")");

    Tree = ParseCompound();

    if(Type != RET_VOID) {
        // Functions with one statement have no composite node, so we have to check
        FinalStatement = (Tree->Operation == OP_COMP) ? Tree->Right : Tree;

        if(FinalStatement == NULL || FinalStatement->Operation != OP_RET) {
            Die("Function with non-void type does not return");
        }

    }

    return ConstructASTBranch(OP_FUNC, Tree->ExprType, Tree, SymbolSlot);
}

/*
 *  Handles the logic for return.
 *  //TODO: No brackets
 *  //TODO: Type inference
 * 
 * 
 */

struct ASTNode* ReturnStatement() {
    struct ASTNode* Tree;
    int ReturnType, FunctionType;


    if(Symbols[CurrentFunction].Type == RET_VOID)
        Die("Attempt to return from void function");
    
    VerifyToken(KW_RETURN, "return");

    VerifyToken(LI_LPARE, "("); // TODO: Make optional! Reject?

    Tree = ParsePrecedenceASTNode(0);

    ReturnType = Tree->ExprType;
    FunctionType = Symbols[CurrentFunction].Type;

    if(!TypesCompatible(&ReturnType, &FunctionType, 0))
        Die("Returning a value of incorrect type for function");
    

    if(ReturnType)
        Tree = ConstructASTBranch(ReturnType, FunctionType, Tree, 0);
    
    Tree = ConstructASTBranch(OP_RET, RET_NONE, Tree, 0);

    printf("\t\tReturning from function %s\n", Symbols[CurrentFunction].Name);

    VerifyToken(LI_RPARE, ")"); // TODO: OPTIONALISE!

    return Tree;
}

/*
 * Handles the assignment of variables.
 * 
 * You can assign variables with an assignment,
 *  a statement, a function or a literal.
 * 
 * This means we need to do some recursive parsing.
 * 
 */

struct ASTNode* ParseIdentifier() {
    struct ASTNode* Left, *Right, *Tree;
    int LeftType, RightType;
    int ID;

    VerifyToken(TY_IDENTIFIER, "ident");

    printf("\t\tAfter parsing, the identifier name is %s, id %d in the symbol table.\n", CurrentIdentifier, FindSymbol(CurrentIdentifier));

    if(CurrentToken.type == LI_LPARE)
        return CallFunction();

    if((ID = FindSymbol(CurrentIdentifier)) == -1) {
        printf("Symbol %s not in table. Table contents: %s, %s\n", CurrentIdentifier, Symbols[0].Name, Symbols[1].Name);
        DieMessage("Undeclared Variable ", CurrentIdentifier);
    }
    Right = ConstructASTLeaf(LV_IDENT, Symbols[ID].Type, ID);

    VerifyToken(LI_EQUAL, "=");

    Left = ParsePrecedenceASTNode(0);

    LeftType = Left->ExprType;
    RightType = Right->ExprType;

    if(!TypesCompatible(&LeftType, &RightType, 1))
        Die("Incompatible variable types");
    
    if(LeftType)
        Left = ConstructASTBranch(LeftType, Right->ExprType, Left, 0);

    Tree = ConstructASTNode(OP_ASSIGN, RET_INT, Left, NULL, Right, 0);

    return Tree;
}

struct ASTNode* IfStatement() {
    struct ASTNode* Condition, *True, *False = NULL;

    VerifyToken(KW_IF, "if");
    VerifyToken(LI_LPARE, "(");

    Condition = ParsePrecedenceASTNode(0);

    // Limit if(x) to =? != < > <= => 
    // No null checking, no arithmetic, no functions.
    // TODO: this
    if(Condition->Operation < OP_EQUAL || Condition->Operation > OP_GREATE)
        Die("Invalid Comparison in if statement");
    
    VerifyToken(LI_RPARE, ")");

    True = ParseCompound();

    if(CurrentToken.type == KW_ELSE) {
        Tokenise(&CurrentToken);
        False = ParseCompound();
    }

    return ConstructASTNode(OP_IF, RET_NONE, Condition, True, False, 0);
}

struct ASTNode* WhileStatement() {
    struct ASTNode* Condition, *Body;

    VerifyToken(KW_WHILE, "while");
    VerifyToken(LI_LPARE, "(");

    Condition = ParsePrecedenceASTNode(0);

    if(Condition->Operation < OP_EQUAL || Condition->Operation > OP_GREATE)
        Die("Bad Comparison inside while()");
    
    VerifyToken(LI_RPARE, ")");

    Body = ParseCompound();

    return ConstructASTNode(OP_LOOP, RET_NONE, Condition, NULL, Body, 0);
}

struct ASTNode* ForStatement() {

    // for (preop; condition; postop) {
    //    body
    //}

    struct ASTNode* Condition, *Body;
    struct ASTNode* Preop, *Postop;

    struct ASTNode* Tree;

    VerifyToken(KW_FOR, "for");
    VerifyToken(LI_LPARE, "(");

    Preop = ParseStatement();
    VerifyToken(LI_SEMIC, ";");

    Condition = ParsePrecedenceASTNode(0);

    if(Condition->Operation < OP_EQUAL || Condition->Operation > OP_GREATE)
        Die("Bad comparison in for");
    VerifyToken(LI_SEMIC, ";");

    Postop = ParseStatement();
    VerifyToken(LI_RPARE, ")");

    Body = ParseCompound();

    // We need to be able to skip over the body and the postop, so we group them together.
    Tree = ConstructASTNode(OP_COMP, RET_NONE, Body, NULL, Postop, 0);
    // We need to be able to jump to the top of the condition and fall through to the body,
    //  so we group it with the last block
    Tree = ConstructASTNode(OP_LOOP, RET_NONE, Condition, NULL, Tree, 0);

    // We need to append the postop to the loop, to form the final for loop
    return ConstructASTNode(OP_COMP, RET_NONE, Preop, NULL, Tree, 0);
}

struct ASTNode* PrintStatement(void) {
    struct ASTNode* Tree;
    int LeftType, RightType;

    VerifyToken(KW_PRINT, "print");

    Tree = ParsePrecedenceASTNode(0);

    LeftType = RET_INT;
    RightType = Tree->ExprType;

    if(!TypesCompatible(&LeftType, &RightType, 0))
        DieDecimal("Attempting to print an invalid type:", RightType);
    
    if(RightType)
        Tree = ConstructASTBranch(RightType, RET_INT, Tree, 0);
    
    Tree = ConstructASTBranch(OP_PRINT, RET_NONE, Tree, 0);

    //ParseAST(Tree);

    return Tree;
   
}

struct ASTNode* PrefixStatement() {
    struct ASTNode* Tree;

    switch (CurrentToken.type) {
        case LI_AMP:
            Tokenise(&CurrentToken);

            // To allow things like:
            // x = &&y;
            // We need to recursively parse prefixes;
            Tree = PrefixStatement();

            if(Tree->Operation != REF_IDENT) 
                Die("& must be followed by another & or an identifier.");

            Tree->Operation = OP_ADDRESS;
            Tree->ExprType = PointerTo(Tree->ExprType);
            break;
        case AR_STAR:
            Tokenise(&CurrentToken);

            Tree = PrefixStatement();

            if(Tree->Operation != REF_IDENT && Tree->Operation != OP_DEREF)
                Die("* must be followed by another * or an identifier.");
            
            Tree = ConstructASTBranch(OP_DEREF, ValueAt(Tree->ExprType), Tree, 0);
            break;

        default:
            Tree = ParsePrimary();

    }

    return Tree;
}