In this part of our compiler writing journey, I want to begin the work to add pointers to our language. In particular, I want to add this:
- Declaration of pointer variables
- Assignment of an address to a pointer
- Dereferencing a pointer to get the value it points at
Given that this is a work in progress, I'm sure I will implement a simplistic version that works for now, but later on I will have to change or extend it for to be more general.
There are no new keywords this time, only two new tokens:
- '&', T_AMPER, and
- '&&', T_LOGAND
We don't need T_LOGAND yet, but I might as well add this
code to scan() now:
case '&':
if ((c = next()) == '&') {
t->token = T_LOGAND;
} else {
putback(c);
t->token = T_AMPER;
}
break;I've added some new primitive types to the language
(in defs.h):
// Primitive types
enum {
P_NONE, P_VOID, P_CHAR, P_INT, P_LONG,
P_VOIDPTR, P_CHARPTR, P_INTPTR, P_LONGPTR
};We will have new unary prefix operators:
- '&' to get the address of an identifier, and
- '*' to dereference a pointer and get the value it points at.
The type of expression that each operator produces
is different to the type that each works on. We need
a couple of functions in types.c to make the type
change:
// Given a primitive type, return
// the type which is a pointer to it
int pointer_to(int type) {
int newtype;
switch (type) {
case P_VOID: newtype = P_VOIDPTR; break;
case P_CHAR: newtype = P_CHARPTR; break;
case P_INT: newtype = P_INTPTR; break;
case P_LONG: newtype = P_LONGPTR; break;
default:
fatald("Unrecognised in pointer_to: type", type);
}
return (newtype);
}
// Given a primitive pointer type, return
// the type which it points to
int value_at(int type) {
int newtype;
switch (type) {
case P_VOIDPTR: newtype = P_VOID; break;
case P_CHARPTR: newtype = P_CHAR; break;
case P_INTPTR: newtype = P_INT; break;
case P_LONGPTR: newtype = P_LONG; break;
default:
fatald("Unrecognised in value_at: type", type);
}
return (newtype);
}Now, where are we going to use these functions?
We want to be able to declare scalar variables and pointer variables, e.g.
char a; char *b;
int d; int *e;We already have a function parse_type() in decl.c
that converts the type keyword to a type. Let's extend
it to scan the following token and change the type if
the next token is a '*':
// Parse the current token and return
// a primitive type enum value. Also
// scan in the next token
int parse_type(void) {
int type;
switch (Token.token) {
case T_VOID: type = P_VOID; break;
case T_CHAR: type = P_CHAR; break;
case T_INT: type = P_INT; break;
case T_LONG: type = P_LONG; break;
default:
fatald("Illegal type, token", Token.token);
}
// Scan in one or more further '*' tokens
// and determine the correct pointer type
while (1) {
scan(&Token);
if (Token.token != T_STAR) break;
type = pointer_to(type);
}
// We leave with the next token already scanned
return (type);
}This will allow the programmer to try to do:
char *****fred;This will fail because pointer_to() can't convert
a P_CHARPTR to a P_CHARPTRPTR (yet). But the code
in parse_type() is ready to do it!
The code in var_declaration() now quite
happily parses pointer variable declarations:
// Parse the declaration of a variable
void var_declaration(void) {
int id, type;
// Get the type of the variable
// which also scans in the identifier
type = parse_type();
ident();
...
}With declarations out of the road, let's now look at parsing expressions where '*' and '&' are operators that come before an expression. The BNF grammar looks like this:
prefix_expression: primary
| '*' prefix_expression
| '&' prefix_expression
;
Technically this allows:
x= ***y;
a= &&&b;
To prevent impossible uses of the two operators, we add in some semantic checking. Here's the code:
// Parse a prefix expression and return
// a sub-tree representing it.
struct ASTnode *prefix(void) {
struct ASTnode *tree;
switch (Token.token) {
case T_AMPER:
// Get the next token and parse it
// recursively as a prefix expression
scan(&Token);
tree = prefix();
// Ensure that it's an identifier
if (tree->op != A_IDENT)
fatal("& operator must be followed by an identifier");
// Now change the operator to A_ADDR and the type to
// a pointer to the original type
tree->op = A_ADDR; tree->type = pointer_to(tree->type);
break;
case T_STAR:
// Get the next token and parse it
// recursively as a prefix expression
scan(&Token); tree = prefix();
// For now, ensure it's either another deref or an
// identifier
if (tree->op != A_IDENT && tree->op != A_DEREF)
fatal("* operator must be followed by an identifier or *");
// Prepend an A_DEREF operation to the tree
tree = mkastunary(A_DEREF, value_at(tree->type), tree, 0);
break;
default:
tree = primary();
}
return (tree);
}We're still doing recursive descent, but we also put error
checks in to prevent input mistakes. Right now, the limitations
in value_at() will prevent more than one '*' operator in a row,
but later on when we change value_at(), we won't have to come
back and change prefix().
Note that prefix() also calls primary() when it doesn't see
a '*' or '&' operator. That allows us to change our existing code
in binexpr():
struct ASTnode *binexpr(int ptp) {
struct ASTnode *left, *right;
int lefttype, righttype;
int tokentype;
// Get the tree on the left.
// Fetch the next token at the same time.
// Used to be a call to primary().
left = prefix();
...
}Up in prefix() I introduced two new AST node types
(declared in defs.h):
- A_DEREF: Dereference the pointer in the child node
- A_ADDR: Get the address of the identifier in this node
Note that the A_ADDR node isn't a parent node. For the
expression &fred, the code in prefix() replaces
the A_IDENT in the "fred" node with the A_ADDR node type.
In our generic code generator, gen.c, there are only
a few new lines to genAST():
case A_ADDR:
return (cgaddress(n->v.id));
case A_DEREF:
return (cgderef(leftreg, n->left->type));The A_ADDR node generates the code to load the
address of the n->v.id identifier into a register.
The A_DEREF node take the pointer address in lefreg,
and its associated type, and returns a register with
the value at this address.
I worked out the following assembly output by reviewing the assembly code generated by other compilers. It might not be correct!
// Generate code to load the address of a global
// identifier into a variable. Return a new register
int cgaddress(int id) {
int r = alloc_register();
fprintf(Outfile, "\tleaq\t%s(%%rip), %s\n", Gsym[id].name, reglist[r]);
return (r);
}
// Dereference a pointer to get the value it
// pointing at into the same register
int cgderef(int r, int type) {
switch (type) {
case P_CHARPTR:
fprintf(Outfile, "\tmovzbq\t(%s), %s\n", reglist[r], reglist[r]);
break;
case P_INTPTR:
case P_LONGPTR:
fprintf(Outfile, "\tmovq\t(%s), %s\n", reglist[r], reglist[r]);
break;
}
return (r);
}The leaq instruction loads the address of the named identifier.
In the section function, the (%r8) syntax loads the value that
register %r8 points to.
Here's our new test file, tests/input15.c and the result when we
compile it:
int main() {
char a; char *b; char c;
int d; int *e; int f;
a= 18; printint(a);
b= &a; c= *b; printint(c);
d= 12; printint(d);
e= &d; f= *e; printint(f);
return(0);
}$ make test15
cc -o comp1 -g -Wall cg.c decl.c expr.c gen.c main.c misc.c
scan.c stmt.c sym.c tree.c types.c
./comp1 tests/input15.c
cc -o out out.s lib/printint.c
./out
18
18
12
12
I decided to change our test files to end with the .c
suffix, now that they are actually C programs. I also
changed the tests/mktests script to generate the
correct results by using a "real" compiler to
compile our test files.
Well, we have the start of pointers implemented. They are not completely correct yet. For example, if I write this code:
int main() {
int x; int y;
int *iptr;
x= 10; y= 20;
iptr= &x + 1;
printint( *iptr);
}it should print 20 because &x + 1 should address
one int past x, i.e. y. This is eight bytes
away from x. However, our compiler simply adds
one to the address of x, which is incorrect. I'll
have to work out how to fix this.
In the next part of our compiler writing journey, we will try to fix this problem. Next step