Readme.md

Part 39: Variable Initialisation, part 1

We can declare variables in the language that our compiler accepts, but we can't initialise them at the same time. So in this part (and the following parts) I will work on fixing this.

It's worth thinking about this now before we do any actual implementation because, hopefully, I can devise a way to share some of the code. So I might do a bit of a "brain dump" below to help me think about the problem.

Right now, we can declare variables in three places:

• Global variables are declared outside of any function
• Function parameters are declared in a parameter list
• Local variables are declared inside of a function

Each declaration includes a description of the variable's type and its name.

In terms of initialisation:

• We can't initialise function parameters, as they will get values copied in from the function caller's arguments.
• Global variables cannot be initialised with an expression, as there is no function in which the expression assembly code can run.
• Local variables can be initialised with an expression.

We also want to have a list of variable names after the type definition. This means that there will some similarities and some differences to deal with. In semi-BNF syntax:

global_declaration: type_definition global_var_list ';' ;

global_var_list: global_var
               | global_var ',' global_var_list  ;

global_var: variable_name
          | variable_name '=' literal_value ;

local_declaration: type_definition local_var_list ';' ;

local_var_list: local_var
              | local_var ',' local_var_list  ;

local_var: variable_name
         | variable_name '=' expression ;

parameter_list: parameter
              | parameter ',' parameter_list ;

parameter: type_definition variable_name ;

Here is a set of examples that I do want to support in our compiler.

Global Declarations

  int   x= 5;
  int   a, b= 7, c, d= 6;
  char *e, f;                           // e is a pointer, f isn't!
  char  g[]= "Hello", *h= "foo";
  int   j[]= { 1, 2, 3, 4, 5 };
  char *k[]= { "fish", "cat", "ball" };
  int   l[70];

The comment I added has deep implications. We will have to parse the type at the front and, for each following variable, parse any prefix '*' or postfix '[ ]' to decide if it's a pointer or an array.

I will only deal with a single dimensional list of initialisation values as shown in the examples above.

Local Declarations

The above examples also apply, but we should also be able to do these local declarations:

  int u= x + 3;
  char *v= k[0];
  char *w= k[b-6];
  int y= 2*b+c, z= l[d] + j[2*x+5];

I was going to offer to parse int list[]= { x+2, a+b, c*d, u+j[3], j[x] + j[a] }; but that looks like an absolute nightmare to deal with, so I think I will stick with either a list of literal values, or not even allow array initialisation in local scope.

Now What?

Right now, after looking at the above examples, I'm kind of terrified! I think I can do the global variable initialisation, but I'll have to rewrite how I parse the types of each individual variable in a list. Then I can parse the '='.

If we are in global scope, I'll call a function to parse the literal values.

If in local scope, I can't use the existing binexpr() function because it parses the variable name on the left and makes an lvalue AST node for it internally. Perhaps I can hand-build this lvalue AST node and pass the pointer to it into binexpr(). Then I can add code to binexpr() that says:

  if we got an lvalue pointer {
    set left to this pointer
  } else {
    left = prefix();
    deal with the operator token
  }
  rest of the existing code

Ok, so I have a sort-of plan. I'll do some refactoring first. And the first task is to work out how to rewrite the parsing of types and variable names so that we can parse lists of them.

A Look at the Refactoring

So I've just done the refactoring of the code and it feels like I've just rearranged code but that's not entirely true. So what I'll do is show you how all the new functions call each other, and then outline what each one does.

I've drawn a call graph of the code in the new decl.c:

At the top, global_declarations() is called to parse anything which is global. It simply loops and calls declaration list(). Alternatively, we are in a function and we've hit a type token (int, char etc.). We call declaration_list() to parse what should be a variable.

declaration_list() is new. It calls parse_type() to get the type (e.g. int, char, a struct, union or typedef etc.). This is the base type of a list, but each thing in the list can modify this type. As an example:

  int a, *b, c[40], *d[100];

So in declaration_list() we loop for each declaration in the list. For each declaration, we call parse_stars() to see how the base type is modified. At this point we can parse the identifier of the individual declaration, and this is done in symbol_declaration(). Based on what token follows, we call:

• function_declaration() for functions,
• array_declaration for arrays, or
• scalar_delaration for scalar variables

In a function declaration, there can be parameters, so parameter_declaration_list() is invoked to do this. Of course, the parameter list is a declaration, so we call declaration_list() to deal with this!

Over on the left we have parse_type(). This gets ordinary types like int and char, but this is where new types such as structs, unions, enums and typedefs are also parsed.

Parsing a typedef in typedef_declaration() should be easy because there is an existing type which we are aliasing. However, we can also write this:

typedef char * charptr;

Because parse_type() doesn't deal with any * tokens, typedef_declaration() has to manually call parse_stars() to see how the base type is modified before creating the alias.

Any enum declaration is handled by enum_declaration. For structs and unions, we call composite_declaration(). And guess what?! The members inside a new struct or union form a list of member declarations, so we call declaration_list() to parse them!

Regression Testing

I'm so glad that I now have about eighty individual tests, because there is no way I could safely refactor decl.c without being able to confirm that the new code still produces the same errors or assembly output as before.

New Functionality

Although this part of the journey is mostly a redesign to get ready for variable initialisation, we now support lists in global and local variable declarations. Therefore, I have new tests:

// tests/input84.c, locals
int main() {
  int x, y;
  x=2; y=3;
  ..
}

//input88.c, globals
struct foo {
  int x;
  int y;
} fred, mary;

Conclusion and What's Next

I feel a bit happier now that I've got the compiler to parse a list of variables following a type, e.g. int a, *b, **c;. I've also put comments into the code where I will have to write the assignment functionality to go with declarations.

In the next part of our compiler writing journey, we will try to add global variable declarations with assignments to our compiler.