In the previous part of our compiler writing journey, I started the groundwork to add variable declarations to our language. I've been able to implement this for global scalar and array variables in this part of our compiler writing journey.
At the same time, I realised that I hadn't designed the symbol table structure to properly deal with the size of a variable and the number of elements in an array variable. So half of this part is going to be a rewrite of some of the code that deals with the symbol table.
As a quick recap, below are a set of example global variable assignments that I want to support:
int x= 2;
char y= 'a';
char *str= "Hello world";
int a[10];
char b[]= { 'q', 'w', 'e', 'r', 't', 'y' };
char c[10]= { 'q', 'w', 'e', 'r', 't', 'y' }; // Zero padded
char *d[]= { "apple", "banana", "peach", "pear" };I'm not going to deal with initialisation of global structs or unions.
Also, for now, I'm not going to deal with putting NULL into char *
variables. I'll come back to that later, if we need it.
In the last part of the journey, I'd written this in decl.c:
static struct symtable *symbol_declaration(...) {
...
// The array or scalar variable is being initialised
if (Token.token == T_ASSIGN) {
...
// Array initialisation
if (stype == S_ARRAY)
array_initialisation(sym, type, ctype, class);
else {
fatal("Scalar variable initialisation not done yet");
// Variable initialisation
// if (class== C_LOCAL)
// Local variable, parse the expression
// expr= binexpr(0);
// else write more code!
}
}
...
}i.e. I knew where to put the code but I didn't know what code to write. First up, we need to parse some literal values...
We are going to need to parse integer and string literals, as these are the
only things which we can assign to global variables. We need to ensure
that the type of each literal is compatible with the variable type that we
are assigning. To this end, there's a new function in decl.c:
// Given a type, check that the latest token is a literal
// of that type. If an integer literal, return this value.
// If a string literal, return the label number of the string.
// Do not scan the next token.
int parse_literal(int type) {
// We have a string literal. Store in memory and return the label
if ((type == pointer_to(P_CHAR)) && (Token.token == T_STRLIT))
return(genglobstr(Text));
if (Token.token == T_INTLIT) {
switch(type) {
case P_CHAR: if (Token.intvalue < 0 || Token.intvalue > 255)
fatal("Integer literal value too big for char type");
case P_INT:
case P_LONG: break;
default: fatal("Type mismatch: integer literal vs. variable");
}
} else
fatal("Expecting an integer literal value");
return(Token.intvalue);
}The first IF statement ensures that we can do:
char *str= "Hello world";and it returns the label number of the address where the string is stored.
For integer literals, we check the range when we are assigning to a char
variable. And for any other token type, we have a fatal error.
The above function always returns an integer, regardless of what type of
literal it parses. Now we need a location in each variable's symbol entry
to store this. So, I've added (and/or modified) these fields in the
symbol entry structure in defs.h:
// Symbol table structure
struct symtable {
...
int size; // Total size in bytes of this symbol
int nelems; // Functions: # params. Arrays: # elements
...
int *initlist; // List of initial values
...
};For a scalar with one initial value, or for an array with several initial
values, we store a count of elements in nelems and attach a list of
integer values to initlist. Let's look at assignment to a scalar variable.
The scalar_declaration() function is modified as follows:
static struct symtable *scalar_declaration(...) {
...
// The variable is being initialised
if (Token.token == T_ASSIGN) {
// Only possible for a global or local
if (class != C_GLOBAL && class != C_LOCAL)
fatals("Variable can not be initialised", varname);
scan(&Token);
// Globals must be assigned a literal value
if (class == C_GLOBAL) {
// Create one initial value for the variable and
// parse this value
sym->initlist= (int *)malloc(sizeof(int));
sym->initlist[0]= parse_literal(type);
scan(&Token);
} // No else code yet, soon
}
// Generate any global space
if (class == C_GLOBAL)
genglobsym(sym);
return (sym);
}We ensure that the assignment can only occur in global or local context,
and we skip over the '=' token. We set up an initlist of exactly one
and call parse_literal() with the type of this variable to get the
literal value (or the label number of a string). Then we skip the
literal value to get to the following token (either a ',' or a ';').
Previously, the sym symbol table entry was created with addglob()
and the number of elements was set to one. I'll cover this change soon.
We now move the call to genglobsym() (which previously was in addglob()
to here, and we wait until the initial value is stored in the sym entry.
This ensures that the literal we just parsed will be put into the storage
for the variable in memory.
As a quick example:
int x= 5;
char *y= "Hello";generates:
.globl x
x:
.long 5
L1:
.byte 72
.byte 101
.byte 108
.byte 108
.byte 111
.byte 0
.globl y
y:
.quad L1
Before we get to the parsing of array intialisation, we need to detour over to the changes to the symbol table code. As I highlighted before, my original code didn't properly handle the storage of the size of a variable nor the number of elements in an array. Let's look at the changes I've made to do this.
Firstly, we have a bug fix. In types.c:
// Return true if a type is an int type
// of any size, false otherwise
int inttype(int type) {
return (((type & 0xf) == 0) && (type >= P_CHAR && type <= P_LONG));
}Previously, there was no test against P_CHAR, so a void type was
treated as an integer type. Oops!
In sym.c we now deal with the fact that each variable now has a:
int size; // Total size in bytes of this symbol
int nelems; // Functions: # params. Arrays: # elementsLater, we will use the size field for the sizeof() operator.
We now need to set up both fields when we add a symbol to the global
or local symbol table.
The newsym() function and all of the addXX() functions in sym.c
now take an nelems argument instead of a size argument. For scalar
variables, this is set to one. For arrays, this is set to the number of
elements in the list. For functions, this is set to the number of function
parameters. And for all other symbol tables, the value is unused.
We now calculate the size value in newsym():
// For pointers and integer types, set the size
// of the symbol. structs and union declarations
// manually set this up themselves.
if (ptrtype(type) || inttype(type))
node->size = nelems * typesize(type, ctype);typesize() consults the ctype pointer to get the size of a struct or
union, or calls genprimsize() (which calls cgprimsize()) to get the
size of a pointer or an integer type.
Note the comment about structs and unions. We can't call addstruct()
(which calls newsym()) with the details of a struct's size,
because:
struct foo { // We call addglob() here
int x;
int y; // before we know the size of the structure
int z;
};So the code in composite_declaration() in decl.c now does this:
static struct symtable *composite_declaration(...) {
...
// Build the composite type
if (type == P_STRUCT)
ctype = addstruct(Text);
else
ctype = addunion(Text);
...
// Scan in the list of members
while (1) {
...
}
// Attach to the struct type's node
ctype->member = Membhead;
...
// Set the overall size of the composite type
ctype->size = offset;
return (ctype);
}So, in summary, the size field in a symbol table entry now holds
the size of all of the elements in the variable, and nelems is
the count of elements in the variable: one for arrays, some non-zero
positive number for arrays.
We can finally get to array initialisation. I want to allow three forms:
int a[10]; // Ten zeroed elements
char b[]= { 'q', 'w', 'e', 'r', 't', 'y' }; // Six elements
char c[10]= { 'q', 'w', 'e', 'r', 't', 'y' }; // Ten elements, zero paddedbut prevent an array declared with size N and more than N initialisation
values. Let's look at the changes to array_declaration(). Previously,
I was going to call an array_initialisation() function, but I decided to
move all of the initialisation code into array_declaration() in decl.c.
We will take it in stages.
// Given the type, name and class of an variable, parse
// the size of the array, if any. Then parse any initialisation
// value and allocate storage for it.
// Return the variable's symbol table entry.
static struct symtable *array_declaration(...) {
int nelems= -1; // Assume the number of elements won't be given
...
// Skip past the '['
scan(&Token);
// See we have an array size
if (Token.token == T_INTLIT) {
if (Token.intvalue <= 0)
fatald("Array size is illegal", Token.intvalue);
nelems= Token.intvalue;
scan(&Token);
}
// Ensure we have a following ']'
match(T_RBRACKET, "]");If there's a number between the '[' ']' tokens, parse it and set nelems
to this value. If there is no number, we leave it set to -1 to indicate this.
We also check that the number is positive and non-zero.
// Array initialisation
if (Token.token == T_ASSIGN) {
if (class != C_GLOBAL)
fatals("Variable can not be initialised", varname);
scan(&Token);
// Get the following left curly bracket
match(T_LBRACE, "{");Right now I'm only dealing with global arrays.
#define TABLE_INCREMENT 10
// If the array already has nelems, allocate that many elements
// in the list. Otherwise, start with TABLE_INCREMENT.
if (nelems != -1)
maxelems= nelems;
else
maxelems= TABLE_INCREMENT;
initlist= (int *)malloc(maxelems *sizeof(int));We create an initial list of either 10 integers, or nelems if
the array was given a fixed size. However, for arrays with no fixed size,
we cannot predict how big the initialisation list will be. So we must be
prepared to grow the list.
// Loop getting a new literal value from the list
while (1) {
// Check we can add the next value, then parse and add it
if (nelems != -1 && i == maxelems)
fatal("Too many values in initialisation list");
initlist[i++]= parse_literal(type);
scan(&Token);Get the next literal value and ensure we don't have more initial values that the array size if it was specified.
// Increase the list size if the original size was
// not set and we have hit the end of the current list
if (nelems == -1 && i == maxelems) {
maxelems += TABLE_INCREMENT;
initlist= (int *)realloc(initlist, maxelems *sizeof(int));
}Here is where we increase the initialisation list size as necessary.
// Leave when we hit the right curly bracket
if (Token.token == T_RBRACE) {
scan(&Token);
break;
}
// Next token must be a comma, then
comma();
}Parse the closing right curly bracket or a comma that separates values.
Once out of the loop, we now have an initlist with values in it.
// Zero any unused elements in the initlist.
// Attach the list to the symbol table entry
for (j=i; j < sym->nelems; j++) initlist[j]=0;
if (i > nelems) nelems = i;
sym->initlist= initlist;
}We may not have been given enough initialisation values to meet the specified size of the initialisation list, so zero out all the ones that were not initialised. It is here that we attach the initialisation list to the symbol table entry.
// Set the size of the array and the number of elements
sym->nelems= nelems;
sym->size= sym->nelems * typesize(type, ctype);
// Generate any global space
if (class == C_GLOBAL)
genglobsym(sym);
return (sym);
}We can finally updated nelems and size in the symbol table entry.
Once this is done, we can call genglobsym() to create the memory
storage for the array.
Before we look at the assembly output of an example array initialisation,
we need to see how the changes of nelems and size have affected the
code that generates the assembly for the memory storage.
genglobsym() is the front-end function which simply calls cgglobsym().
Let's look at this function in cg.c:
// Generate a global symbol but not functions
void cgglobsym(struct symtable *node) {
int size, type;
int initvalue;
int i;
if (node == NULL)
return;
if (node->stype == S_FUNCTION)
return;
// Get the size of the variable (or its elements if an array)
// and the type of the variable
if (node->stype == S_ARRAY) {
size= typesize(value_at(node->type), node->ctype);
type= value_at(node->type);
} else {
size = node->size;
type= node->type;
}Right now, arrays have their type set to be a pointer to the underlying
element type. This allows us to do:
char a[45];
char *b;
b= a; // as they are of same typeIn terms of generating storage, we need to know the size of the elements,
so we call value_at() to do this. For scalars, size and type are
stored as-is in the symbol table entry.
// Generate the global identity and the label
cgdataseg();
fprintf(Outfile, "\t.globl\t%s\n", node->name);
fprintf(Outfile, "%s:\n", node->name);As before. But now the code is different:
// Output space for one or more elements
for (i=0; i < node->nelems; i++) {
// Get any initial value
initvalue= 0;
if (node->initlist != NULL)
initvalue= node->initlist[i];
// Generate the space for this type
switch (size) {
case 1:
fprintf(Outfile, "\t.byte\t%d\n", initvalue);
break;
case 4:
fprintf(Outfile, "\t.long\t%d\n", initvalue);
break;
case 8:
// Generate the pointer to a string literal
if (node->initlist != NULL && type== pointer_to(P_CHAR))
fprintf(Outfile, "\t.quad\tL%d\n", initvalue);
else
fprintf(Outfile, "\t.quad\t%d\n", initvalue);
break;
default:
for (int i = 0; i < size; i++)
fprintf(Outfile, "\t.byte\t0\n");
}
}
}For every element, get its intial value from the initlist or use zero
if no initialisation list. Based on the size of each element, output
either a byte, a long or a quad.
For char * elements, we have the label of the string literal's base
in the initialisation list, so output "L%d" (i.e. the label) instead
of the integer literal value.
Here is a small example of an array initialisation:
int x[4]= { 1, 4, 17 };generates:
.globl x
x:
.long 1
.long 4
.long 17
.long 0
I won't go through the test programs, but the programs
tests/input89.c through to tests/input99.c check that the
compiler is generating sensible initialisation code as well as catching
suitable fatal errors.
So that was a lot of work! Three steps forward and one step back, as they say. I'm happy, though, because the changes to the symbol table make much more sense than what I had before.
In the next part of our compiler writing journey, we will try to add local variable initialisation to the compiler.