CSCI305.github.io

CSCI 305 Homework 2

Due Date: February 23, 2018 @ Beginning of Class



Name:                                                                                        




Syntax and Semantics

  1. Starting with the following BNF grammar:
    <exp> ::= <exp> + <mulexp> | <mulexp>
    <mulexp> ::= <mulexp> * <rootexp> | <rootexp>  |
    <rootexp> ::= ( <exp> )
              | a | b | c
    

    Construct an EBNF grammar from this grammar with the following modifications. Use the EBNF extensions wherever possible to simply the grammars. Include whatever notes to the reader required to make the associativity of the operators clear:

    a. Add subtraction and division operators (- and /) with the customary precedence and associativity.

    b. Then add a left-associative operator % between + and * in precedence.

    c. Then add a right-associative operator = at lower precedence than any of the other operators.

    Complete Solution with Recursion

    <exp> ::= <plusexp> [= <exp>]
    <plusexp> ::= [<plusexp> (+ | -)] <modexp>
    <modexp> ::= [<modexp> %] <mulexp>
    <mulexp> ::= [<mulexp> (/ | *)] <rootexp>
    <rootexp> ::= '(' <exp> ')' | a | b | c
    

    OR With Both recursive and non-recursive rules

    <exp> ::= <plusexp> [= <exp>]
    <plusexp> ::= <modexp> {(+ | -) <modexp>}
    <modexp> ::= <mulexp> { % <mulexp>}
    <mulexp> ::= <rootexp> {(/ | *) <rootexp>}
    <rootexp> ::= '(' <exp> ')' | a | b | c
    
    Note: <proda> ::= <prodb> {op <prodb>} only used for left-associative ops
    

    OR With Both recursive and non-recursive rules

    <exp> ::= {<plusexp> =} <plusexp>
    <plusexp> ::= [<plusexp> (+ | -)] <modexp>
    <modexp> ::= [<modexp> %] <mulexp>
    <mulexp> ::= [<mulexp> (/ | *)] <rootexp>
    <rootexp> ::= '(' <exp> ')' | a | b | c
    
    Note: <proda> ::= {<prodb> op} <prodb> only used for right-associative ops
    
  2. Show that the following grammar is ambiguous. (Note: To show that a grammar is ambiguous, you must demonstrate that it can generate two parse trees for the same string.)

      <person> ::= <woman> | <man>
      <woman> ::= wilma | betty | <empty>
      <man> ::= fred | barney | <empty>
    

    Complete Solution

    There are five possible strings that this grammar can generate: “wilma”, “betty”, “fred”, “barney”, and “”. Only one of these leads to more one derivation: “”. Thus the grammar is ambiguous. Thus, for the string “” (the empty string) we can generate the following two derivations:

    Derivation 1:

    <person> => <woman>
             => <empty>
             =>
    

    Derivation 2:

    <person> => <man>
             => <empty>
             =>
    

    Which lead to the following two parse trees:

         <person>              <person>
            |                     |
         <woman>                <man>
            |                     |
         <empty>               <empty>
            |                     |
           ""                     ""
    
  3. For the grammar in Question 2, construct an unambiguous grammar for the same language.

    Complete Solution

    We can easily remove the ambiguity by shifting the <empty> terminal to the <person> rule and removing it from both the <woman> and <man> rules. It we look at this refined grammar below, we can then see that it still generates all five possible strings: “wilma”, “betty”, “fred”, “barney”, and “”, but this time there is only one possible parse tree for each. Therefore, it is now unambiguous.

    <person> ::= <woman> | <man> | <empty>
    <woman> ::= wilma | betty
    <man> ::= fred | barney
    

Language Systems

  1. Suppose the target assembly language for a compiler has these five instructions:

      push address
      add
      sub
      mul
      pop address
    

    In these instructions, and address is the name of a static variable (whose actual address will be filled in by the loader). The machine maintains a stack of integers, which can grow to any size. The push instruction pushes the integer from the given memory address to the top of the stack. The add instruction adds the top integer on the stack to the next-from-the-top integer, pops both off, and pushes the result onto the stack. The sub instruction subtracts the top integer on the stack from the next-from-the-top integer, pops both off, and pushes the result onto the stack. The mul instruction multiplies the top integer on the stack by the next-from-the-top integer, pops both off, and pushes the result onto the stack. The pop instruction pops an integer off the stack and stores it at the given memory address. So, for example, the compiler might translate the assignment result := offset + (width * n) into this:

       push offset
       push width
       push n
       mul
       add
       pop result
    

    Using this assembly language, give translations of the following assignment statements using as few instructions as possible:

    a. net := gross - costs

    b. cube := (x * x) * x

    c. final := ((a - abase) * (b - bbase)) * (c - cbase)

    Complete Solution

    net := gross - costs

      push gross
      push costs
      sub
      pop net
    

    cube := (x * x) * x

      push x
      push x
      push x
      mul
      mul
      pop cube
    

    final := ((a - abase) * (b - bbase)) * (c - cbase)

      push a
      push abase
      sub
      push b
      push bbase
      sub
      mul
      push c
      push cbase
      sub
      mul
      pop final
    
  2. Investigate the COMMON keyword in Fortran. Describe how Fortran common blocks work and give an example. What happens if two named common blocks with the same name contain different variables? What is the difference between a blank common and a named common? What does the linker have to do to make this language construct work?