# A compiler as described by Jack Crenshaw in his famous book "Let's # Build a Compiler". At least in the beginning, this code will # closely reflect the Pascal code written by Jack. Over time it may # become more idiomatic, however this is an academic exercise. # # sjs # may 2009 class ParseError < StandardError; end class Compiler attr_reader :data, :bss, :code def initialize(input=STDIN) @look = '' # next lookahead char @input = input # stream to read from @data = '' # data section @bss = '' # bss section @code = '' # code section @vars = {} # seed the lexer get_char skip_whitespace end def parse statement until eof? [@data, @bss, @code] end # Read the next character from the input stream. def get_char @look = if @input.eof? nil else @input.readbyte.chr end end # Report error and halt def abort(msg) raise ParseError, msg end # Report what was expected def expected(what) if eof? raise ParseError, "Premature end of file, expected: #{what}." else raise ParseError, "Expected: #{what}, got: #{@look} (##{@look[0]})." end end # Recognize an alphabetical character. def alpha?(char) ('A'..'Z') === char.upcase end # Recognize a decimal digit. def digit?(char) ('0'..'9') === char end # Recognize an alphanumeric character. def alnum?(char) alpha?(char) || digit?(char) end def whitespace?(char) char == ' ' || char == '\t' end # Match a specific input character. def match(char) expected("'#{char}'") unless @look == char get_char skip_whitespace end # Parse zero or more consecutive characters for which the test is # true. def many(test) token = '' while test.call(@look) token << @look get_char end skip_whitespace token end # Get an identifier. def get_name expected('identifier') unless alpha?(@look) many(method(:alnum?)) end # Get a number. def get_num expected('integer') unless digit?(@look) many(method(:digit?)) end # Skip all leading whitespace. def skip_whitespace get_char while whitespace?(@look) end # Define a constant in the .data section. def equ(name, value) @data << "#{name}\tequ #{value}" end # Define a variable with the given name and size (in dwords). def var(name, dwords=1) unless @vars[name] @bss << "#{name}: resd #{dwords}\n" @vars[name] = name # else # raise ParseError, "identifier #{name} redefined" end end # Emit a line of code wrapped between a tab and a newline. def emit(s) @code << "\t#{s}\n" end # Parse and translate an identifier or function call. def identifier name = get_name if @look == '(' # function call match('(') match(')') x86_call(name) else # variable access x86_mov(:eax, "dword [#{name}]") end end # Parse and translate a single factor. Result is in eax. def factor if @look == '(' match('(') expression match(')') elsif alpha?(@look) identifier elsif digit?(@look) x86_mov(:eax, get_num) else expected("a number, identifier, or an expression wrapped in parens") end end # Parse and translate a single term (factor or mulop). Result is in # eax. def term factor # Result in eax. while mulop? # Stash the 1st factor on the stack. This is expected by # multiply & divide. Because they leave their results in eax # associativity works. Each interim result is pushed on the # stack here. x86_push(:eax) if @look == '*' multiply else divide end x86_add(:esp, 4) # Remove the 1st factor from the stack. end end # Parse and translate a general expression of terms. Result is # in eax. def expression if addop? # Clear eax simulating a zero before unary plus and minus # operations. x86_xor(:eax, :eax) else term # Result is in eax. end while addop? # Stash the 1st term on the stack. This is expected by add & # subtract. Because they leave their results in eax # associativity works. Each interim result is pushed on the # stack here. x86_push(:eax) if @look == '+' add else subtract end x86_add(:esp, 4) # Remove 1st term (a) from the stack. end end # Parse an assignment statement. Value is in eax. def assignment name = get_name match('=') expression var(name) x86_mov("dword [#{name}]", :eax) end # Parse one or more newlines. def newline if @look == "\n" || @look == "\r" get_char while @look == "\n" || @look == "\r" else expected('newline') end end # Parse an assignment expression followed by a newline. def statement assignment newline end # Parse an addition operator and the 2nd term (b). The result is # left in eax. The 1st term (a) is expected on the stack. def add match('+') term # Result is in eax. x86_add(:eax, '[esp]') # Add a to b. end # Parse a subtraction operator and the 2nd term (b). The result is # left in eax. The 1st term (a) is expected on the stack. def subtract match('-') term # Result, b, is in eax. x86_neg(:eax) # Fake the subtraction. a - b == a + -b x86_add(:eax, '[esp]') # Add a and -b. end # Parse an addition operator and the 2nd term (b). The result is # left in eax. The 1st term (a) is expected on the stack. def multiply match('*') factor # Result is in eax. x86_imul('dword [esp]') # Multiply a by b. end # Parse a division operator and the divisor (b). The result is # left in eax. The dividend (a) is expected on the stack. def divide match('/') factor # Result is in eax. x86_xchg(:eax, '[esp]') # Swap the divisor and dividend into # the correct places. # idiv uses edx:eax as the dividend so we need to ensure that edx # is correctly sign-extended w.r.t. eax. emit('cdq') # Sign-extend eax into edx (Convert Double to # Quad). x86_idiv('dword [esp]') # Divide a (eax) by b ([esp]). end ####### private ####### def eof? @input.eof? && @look.nil? end def addop? @look == '+' || @look == '-' end def mulop? @look == '*' || @look == '/' end # Some asm methods for convenience and arity checks. def x86_mov(dest, src) emit("mov #{dest}, #{src}") end def x86_add(dest, src) emit("add #{dest}, #{src}") end def x86_sub(dest, src) emit("sub #{dest}, #{src}") end def x86_imul(op) emit("imul #{op}") end def x86_idiv(op) emit("idiv #{op}") end def x86_push(reg) emit("push #{reg}") end def x86_call(label) emit("call #{label}") end def x86_neg(reg) emit("neg #{reg}") end def x86_xchg(op1, op2) emit("xchg #{op1}, #{op2}") end def x86_xor(op1, op2) emit("xor #{op1}, #{op2}") end end