1.1 --- a/bytecode.py	Tue Nov 09 00:51:39 2004 +0100
     1.2 +++ b/bytecode.py	Tue Nov 09 00:59:16 2004 +0100
     1.3 @@ -17,9 +17,19 @@
     1.4      "A Python bytecode writer."
     1.5  
     1.6      def __init__(self):
     1.7 -        self.loops = []
     1.8 +        # A stack of loop block or exception block start positions.
     1.9 +        self.blocks = []
    1.10 +
    1.11 +        # A stack of exception block handler pointers.
    1.12 +        self.exception_handlers = []
    1.13 +
    1.14 +        # A dictionary mapping labels to jump instructions referencing such labels.
    1.15          self.jumps = {}
    1.16 +
    1.17 +        # The output values, including "lazy" subvalues which will need evaluating.
    1.18          self.output = []
    1.19 +
    1.20 +        # The current Python bytecode instruction position.
    1.21          self.position = 0
    1.22  
    1.23          # Stack depth estimation.
    1.24 @@ -91,6 +101,7 @@
    1.25              # NOTE: Assume a 16-bit value.
    1.26              self.output.append(value.values[0])
    1.27              self.output.append(value.values[1])
    1.28 +            self.position += 2
    1.29          elif value <= 0xffff:
    1.30              self.output.append(value & 0xff)
    1.31              self.output.append((value & 0xff00) >> 8)
    1.32 @@ -100,15 +111,18 @@
    1.33              raise ValueError, value
    1.34  
    1.35      def setup_loop(self):
    1.36 -        self.loops.push(self.position)
    1.37 +        self.blocks.append(self.position)
    1.38          self.output.append(opmap["SETUP_LOOP"])
    1.39          self.position += 1
    1.40          self._write_value(0) # To be filled in later
    1.41  
    1.42      def end_loop(self):
    1.43 -        current_loop_start = self.loops.pop()
    1.44 +        current_loop_start = self.blocks.pop()
    1.45          self.jump_absolute(current_loop_start)
    1.46 -        self.output[current_loop_start + 1] = self.position
    1.47 +        # NOTE: Using 3 as the assumed length of the SETUP_LOOP instruction.
    1.48 +        # NOTE: 8-bit limit.
    1.49 +        self.output[current_loop_start + 1] = self.position - current_loop_start - 3
    1.50 +        self.output[current_loop_start + 2] = 0
    1.51          self.pop_block()
    1.52  
    1.53      def jump_to_label(self, status, name):
    1.54 @@ -128,7 +142,9 @@
    1.55      def start_label(self, name):
    1.56          # Fill in all jump instructions.
    1.57          for jump_instruction, following_instruction in self.jumps[name]:
    1.58 +            # NOTE: 8-bit limit.
    1.59              self.output[jump_instruction + 1] = self.position - following_instruction
    1.60 +            self.output[jump_instruction + 2] = 0
    1.61          del self.jumps[name]
    1.62  
    1.63      def load_const_ret(self, value):
    1.64 @@ -136,19 +152,32 @@
    1.65          self.load_const(value)
    1.66  
    1.67      def ret(self, index):
    1.68 -        # Previously, the constant stored on the stack by jsr/jsr_w was stored
    1.69 -        # in a local variable. In the JVM, extracting the value from the local
    1.70 -        # variable and jumping can be done at runtime. In the Python VM, any
    1.71 -        # jump target must be known in advance and written into the bytecode.
    1.72 -        self.load_fast(index)
    1.73 -        for constant in self.constants_for_exceptions:
    1.74 -            self.dup_top()              # Stack: actual-address, actual-address
    1.75 -            self.load_const(constant)   # Stack: actual-address, actual-address, suggested-address
    1.76 -            self.compare_op("==")       # Stack: actual-address, result
    1.77 -            self.jump_to_label(0, "const")
    1.78 -            self.jump_absolute(constant)
    1.79 -            self.start_label("const")
    1.80 -            self.pop_top()              # Stack: actual-address
    1.81 +        self.end_finally()
    1.82 +
    1.83 +    def setup_except(self, target):
    1.84 +        self.blocks.append(self.position)
    1.85 +        self.exception_handlers.append(target)
    1.86 +        self.output.append(opmap["SETUP_EXCEPT"])
    1.87 +        self.position += 1
    1.88 +        self._write_value(0) # To be filled in later
    1.89 +
    1.90 +    def setup_finally(self, target):
    1.91 +        self.blocks.append(self.position)
    1.92 +        self.exception_handlers.append(target)
    1.93 +        self.output.append(opmap["SETUP_FINALLY"])
    1.94 +        self.position += 1
    1.95 +        self._write_value(0) # To be filled in later
    1.96 +
    1.97 +    def end_exception(self):
    1.98 +        current_exception_start = self.blocks.pop()
    1.99 +        # Convert the "lazy" absolute value.
   1.100 +        current_exception_target = self.exception_handlers.pop()
   1.101 +        target = current_exception_target.get_value()
   1.102 +        # NOTE: Using 3 as the assumed length of the SETUP_* instruction.
   1.103 +        # NOTE: 8-bit limit.
   1.104 +        self.output[current_exception_start + 1] = target - current_exception_start - 3
   1.105 +        self.output[current_exception_start + 2] = 0
   1.106 +        # NOTE: The POP_BLOCK instruction gets slipped in before this method is called.
   1.107  
   1.108      # Complicated methods.
   1.109  
   1.110 @@ -206,7 +235,7 @@
   1.111      # Normal bytecode generators.
   1.112  
   1.113      def for_iter(self):
   1.114 -        self.loops.push(self.position)
   1.115 +        self.blocks.append(self.position)
   1.116          self.output.append(opmap["FOR_ITER"])
   1.117          self.position += 1
   1.118          self._write_value(0) # To be filled in later
   1.119 @@ -347,6 +376,10 @@
   1.120          self.output.append(opmap["POP_BLOCK"])
   1.121          self.position += 1
   1.122  
   1.123 +    def end_finally(self):
   1.124 +        self.output.append(opmap["END_FINALLY"])
   1.125 +        self.position += 1
   1.126 +
   1.127  # Utility classes and functions.
   1.128  
   1.129  class LazyDict(UserDict):
   1.130 @@ -378,8 +411,9 @@
   1.131              raise ValueError, value
   1.132      def get_value(self):
   1.133          value = 0
   1.134 -        for i in range(0, len(self.values)):
   1.135 -            value = (value << 8) + self.values.pop().value
   1.136 +        values = self.values[:]
   1.137 +        for i in range(0, len(values)):
   1.138 +            value = (value << 8) + values.pop().value
   1.139          return value
   1.140  
   1.141  class LazySubValue:
   1.142 @@ -419,13 +453,65 @@
   1.143          self.class_file = class_file
   1.144          self.position_mapping = LazyDict()
   1.145  
   1.146 -    def process(self, code, program):
   1.147 +    def process(self, code, exception_table, program):
   1.148          self.java_position = 0
   1.149 +
   1.150 +        # Produce a structure which permits fast access to exception details.
   1.151 +        exception_block_start = {}
   1.152 +        exception_block_end = {}
   1.153 +        exception_block_handler = {}
   1.154 +        reversed_exception_table = exception_table[:]
   1.155 +        reversed_exception_table.reverse()
   1.156 +
   1.157 +        # Later entries have wider coverage than earlier entries.
   1.158 +        for exception in reversed_exception_table:
   1.159 +            # Index start positions.
   1.160 +            if not exception_block_start.has_key(exception.start_pc):
   1.161 +                exception_block_start[exception.start_pc] = []
   1.162 +            exception_block_start[exception.start_pc].append(exception)
   1.163 +            # Index end positions.
   1.164 +            if not exception_block_end.has_key(exception.end_pc):
   1.165 +                exception_block_end[exception.end_pc] = []
   1.166 +            exception_block_end[exception.end_pc].append(exception)
   1.167 +            # Index handler positions.
   1.168 +            if not exception_block_handler.has_key(exception.handler_pc):
   1.169 +                exception_block_handler[exception.handler_pc] = []
   1.170 +            exception_block_handler[exception.handler_pc].append(exception)
   1.171 +
   1.172 +        # Process each instruction in the code.
   1.173          while self.java_position < len(code):
   1.174              self.position_mapping[self.java_position] = program.position
   1.175 +
   1.176 +            # Insert exception handling constructs.
   1.177 +            for exception in exception_block_start.get(self.java_position, []):
   1.178 +                # Note that the absolute position is used.
   1.179 +                if exception.catch_type == 0:
   1.180 +                    program.setup_finally(self.position_mapping[exception.handler_pc])
   1.181 +                else:
   1.182 +                    program.setup_except(self.position_mapping[exception.handler_pc])
   1.183 +
   1.184 +            # Insert exception handler end details.
   1.185 +            for exception in exception_block_end.get(self.java_position, []):
   1.186 +                program.end_exception()
   1.187 +
   1.188 +            # Where handlers are begun, do not produce equivalent bytecode since
   1.189 +            # the first handler instruction typically involves saving a local
   1.190 +            # variable that is not applicable to the Python VM.
   1.191 +            #if not exception_block_handler.get(self.java_position, []):
   1.192 +
   1.193 +            # Process the bytecode at the current position.
   1.194              bytecode = ord(code[self.java_position])
   1.195              mnemonic, number_of_arguments = self.java_bytecodes[bytecode]
   1.196 -            self.process_bytecode(mnemonic, number_of_arguments, code, program)
   1.197 +            number_of_arguments = self.process_bytecode(mnemonic, number_of_arguments, code, program)
   1.198 +            next_java_position = self.java_position + 1 + number_of_arguments
   1.199 +
   1.200 +            # Insert exception handler end instructions.
   1.201 +            for exception in exception_block_end.get(next_java_position, []):
   1.202 +                program.pop_block()
   1.203 +
   1.204 +            # Only advance the JVM position after sneaking in extra Python
   1.205 +            # instructions.
   1.206 +            self.java_position = next_java_position
   1.207  
   1.208      def process_bytecode(self, mnemonic, number_of_arguments, code, program):
   1.209          if number_of_arguments is not None:
   1.210 @@ -435,11 +521,10 @@
   1.211  
   1.212              # Call the handler.
   1.213              getattr(self, mnemonic)(arguments, program)
   1.214 +            return number_of_arguments
   1.215          else:
   1.216              # Call the handler.
   1.217 -            number_of_arguments = getattr(self, mnemonic)(code[self.java_position+1:], program)
   1.218 -
   1.219 -        self.java_position = self.java_position + 1 + number_of_arguments
   1.220 +            return getattr(self, mnemonic)(code[self.java_position+1:], program)
   1.221  
   1.222      java_bytecodes = {
   1.223          # code : (mnemonic, number of following bytes, change in stack)
   1.224 @@ -664,6 +749,14 @@
   1.225  
   1.226  class BytecodeDisassemblerProgram:
   1.227      position = 0
   1.228 +    def setup_except(self, target):
   1.229 +        print "(setup_except %s)" % target
   1.230 +    def setup_finally(self, target):
   1.231 +        print "(setup_finally %s)" % target
   1.232 +    def end_exception(self):
   1.233 +        print "(end_exception)"
   1.234 +    def pop_block(self):
   1.235 +        print "(pop_block)"
   1.236  
   1.237  class BytecodeTranslator(BytecodeReader):
   1.238  
   1.239 @@ -763,7 +856,8 @@
   1.240      def checkcast(self, arguments, program):
   1.241          index = (arguments[0] << 8) + arguments[1]
   1.242          target_name = self.class_file.constants[index - 1].get_name()
   1.243 -        target_components = target_name.split("/")
   1.244 +        # NOTE: Using the string version of the name which may contain incompatible characters.
   1.245 +        target_components = str(target_name).split("/")
   1.246  
   1.247          program.dup_top()                   # Stack: objectref, objectref
   1.248          program.load_global("isinstance")   # Stack: objectref, objectref, isinstance
   1.249 @@ -1065,7 +1159,8 @@
   1.250      def instanceof(self, arguments, program):
   1.251          index = (arguments[0] << 8) + arguments[1]
   1.252          target_name = self.class_file.constants[index - 1].get_name()
   1.253 -        target_components = target_name.split("/")
   1.254 +        # NOTE: Using the string version of the name which may contain incompatible characters.
   1.255 +        target_components = str(target_name).split("/")
   1.256  
   1.257          program.load_global("isinstance")   # Stack: objectref, isinstance
   1.258          program.rot_two()                   # Stack: isinstance, objectref
   1.259 @@ -1401,15 +1496,15 @@
   1.260          # NOTE: To be implemented.
   1.261          return number_of_arguments
   1.262  
   1.263 -def disassemble(class_file, code):
   1.264 +def disassemble(class_file, code, exception_table):
   1.265      disassembler = BytecodeDisassembler(class_file)
   1.266 -    disassembler.process(code, BytecodeDisassemblerProgram())
   1.267 +    disassembler.process(code, exception_table, BytecodeDisassemblerProgram())
   1.268  
   1.269 -def translate(class_file, code):
   1.270 +def translate(class_file, code, exception_table):
   1.271      translator = BytecodeTranslator(class_file)
   1.272      writer = BytecodeWriter()
   1.273 -    translator.process(code, writer)
   1.274 -    return writer
   1.275 +    translator.process(code, exception_table, writer)
   1.276 +    return translator, writer
   1.277  
   1.278  if __name__ == "__main__":
   1.279      import sys