MCS-388 Lab 4: Control Flow and Scoping (Spring 2003)

Due: April 23, 2003

Objective

• Several new kinds of statements:
  • while statements
  • if statements, both with and without else parts
  • brace-enclosed compound statements
• Nested variable scopes (arising from declarations in compound statements)
• The six arithmetic comparison operators (<, >, <=, >=, ==, !=)

Of these, while and if statements introduce interesting control flow, for the first time allowing programs that do something other than continue in a straight-line path. Compound statements are important in conjunction with while and if statements, in order to allow a loop body (for example) to perform more than one action. Moreover, introducing compound statements gives us a natural context for variable scoping. The comparison operators are the only addition of no real substance: they are fundamentally similar to the arithmetic operators, and could equally well have been in the compiler the whole time. The only reason to add them now is that they become particularly handy to have now that we have loops and conditional statements.

Some notes on the new language features

x = y > z;

print_int((x > y) + (z > w));

The syntax of while and if statements should be as in C (and C++ and Java). For example, the controlling expression in each case must be parenthesized, and there is no keyword (like then or do) between the parenthesized expression and the following statement. Remember that any statement can be used as the body of a loop, or as an alternative in an if statement: there is no requirement of curly braces. They have nothing to do with the while or if statement, and are only present if a compound statement is used, as part of the compound statement. For example, the following three while loops are all legal:

while(x >= 10)
  x = x / 10;     // body is an assignment statement

while(x >= 10){   // body is a compound statement, which
  x = x / 10;     // happens to only have one statement in it
  }

while(x >= 10){   // body is a more general compound statement
  int d = x % 10;
  x = x / 10;
  print_int(d);
  }

Each brace-enclosed compound statement constitutes a new scope, nested inside the surrounding scope. It is legal to redeclare a variable that was declared in some outer, surrounding scope, but not legal to redeclare a variable in the exact same scope. Each variable reference refers to a declaration that textually precedes the reference and is in a scope that has not yet ended at the point of the reference. If there is no such declaration, the variable reference is in error. If there is more than one such declaration, the one in the innermost scope is used. To determine whether a declaration textually precedes a reference, consider any declaration that has an initializer to take place at the end of the initializer. For example, in the program:

main(){
  int x;          // declaration 1
  x = 3;          // reference (a)
  while(x){       // reference (b)
    x = x - 1;    // references (c) and (d)
    int x = x + 5;// reference (e) and declaration 2
    print_int(x); // reference (f)
    }
  {
    print_int(x); // reference (g)
  }
  {
    int x = 2;    // declaration 3
  }
  print_int(x);   // reference (h)
  }

Quality expectations for generated code

If you do have the new features and the prior optimizations working, it would be desirable to also turn relational operators around when appropriate to be able to take advantage of immediate operands. For example, if the source code contains 3>x, it would be better to turn this around to x<3 so that you can generate an instruction like slt $t0, $s0, 3.

For while loops, it is preferable to generate code like this:

   jump to label1
label2:
   code for the body goes here
label1:
   code for the test goes here
   branch if true to label2

label1:
   code for the test goes here
   branch if false to label2
   code for the body goes here
   jump to label1
label2:

Normally if statements translate into something like

   code for the test goes here
   branch if false to label1
   code for the first alternative goes here
   jump to label2
label1:
   code for the "else" part (2nd alternative) goes here
label2:

Don't worry if there are unusual circumstances under which your code will jump (or branch) to a jump instruction, or jump (or branch) to the next instruction, so long as you've taken care to not generate routinely stupid code. (Routinely stupid means something like jumping around a non-existent else part.)

Code to start from

Code from prior lab

If any of you thinks your difficulties with the prior lab would put you at a disadvantage for doing the new lab, you are welcome to ask me for my code from the prior lab, to use as a starting point.

Sample ScopedIdentifier class

My intention is that you call all the ScopedIdentifier class's method's during the parsing process, not during the AST's later code generation. This includes enterScope, exitScope, get, declare, getDeclaration and (when errors need reporting) toString. Of course, not all of these need to be literally called directly from parser.cup. It is OK if you want to call some of them indirectly, from something called from the parser. (As one example, you may well be calling the get method from the scanner rather than the parser -- but the parser calls the scanner.)

If you are looking for another learning opportunity, this would be a great class to write yourself.

Shift/reduce conflicts

   java java_cup.Main <parser.cup

   java java_cup.Main -expect 1 <parser.cup

A constant-folding subtlety

It is important for your compiler not to die if the program contains divisions by zero, because they might be in a non-executed part of the program. Thus, you'll have to not constant-fold any divisions where the divisor is zero.

Worse yet, it turns out that SPIM (our MIPS simulator) doesn't recognize that some instructions aren't executed, and so it won't load in an assembly language program that has a division instruction with an immediate 0 operand. Thus you'll need to be careful not to generate such instructions. (You can use register $zero instead.)

Postlab