def _build_iterator_warp(state, last_addr, instructions): iterator = instructions[-2] iterator_addr = last_addr - 1 assert iterator.opcode in (ins.ITERC.opcode, ins.ITERN.opcode) warp = nodes.IteratorWarp() base = iterator.A warp.controls.contents = [ _build_slot(state, iterator_addr, base - 3), # generator _build_slot(state, iterator_addr, base - 2), # state _build_slot(state, iterator_addr, base - 1) # control ] last_slot = base + iterator.B - 2 slot = base while slot <= last_slot: variable = _build_slot(state, iterator_addr - 1, slot) warp.variables.contents.append(variable) slot += 1 jump = instructions[-1] jump_addr = last_addr destination = get_jump_destination(jump_addr, jump) warp.way_out = state._warp_in_block(jump_addr + 1) warp.body = state._warp_in_block(destination) return warp, 2
def _build_iterator_warp(state, last_addr, instructions):
iterator = instructions[-2]
iterator_addr = last_addr - 1
assert iterator.opcode in (ins.ITERC.opcode, ins.ITERN.opcode)
warp = nodes.IteratorWarp()
base = iterator.A
warp.controls.contents = [
_build_slot(state, iterator_addr, base - 3), # generator
_build_slot(state, iterator_addr, base - 2), # state
_build_slot(state, iterator_addr, base - 1) # control
]
last_slot = base + iterator.B - 2
slot = base
while slot <= last_slot:
variable = _build_slot(state, iterator_addr - 1, slot)
warp.variables.contents.append(variable)
slot += 1
jump = instructions[-1]
jump_addr = last_addr
destination = get_jump_destination(jump_addr, jump)
warp.way_out = state._warp_in_block(jump_addr + 1)
warp.body = state._warp_in_block(destination)
return warp, 2
def _build_conditional_warp(state, last_addr, instructions):
condition = instructions[-2]
condition_addr = last_addr - 1
warp = nodes.ConditionalWarp()
if condition.opcode in (ins.ISTC.opcode, ins.ISFC.opcode):
expression = _build_unary_expression(state, condition_addr, condition)
setattr(warp, "_slot", condition.A)
elif condition.opcode >= ins.IST.opcode:
expression = _build_unary_expression(state, condition_addr, condition)
setattr(warp, "_slot", condition.CD)
else:
expression = _build_comparison_expression(state, condition_addr,
condition)
warp.condition = expression
jump = instructions[-1]
jump_addr = last_addr
destination = get_jump_destination(jump_addr, jump)
# A condition is inverted during the preparation phase above
warp.false_target = state._warp_in_block(destination)
warp.true_target = state._warp_in_block(jump_addr + 1)
return warp, 2
def _build_conditional_warp(state, last_addr, instructions): condition = instructions[-2] condition_addr = last_addr - 1 warp = nodes.ConditionalWarp() if condition.opcode in (ins.ISTC.opcode, ins.ISFC.opcode): expression = _build_unary_expression(state, condition_addr, condition) setattr(warp, "_slot", condition.A) elif condition.opcode >= ins.IST.opcode: expression = _build_unary_expression(state, condition_addr, condition) setattr(warp, "_slot", condition.CD) else: expression = _build_comparison_expression(state, condition_addr, condition) warp.condition = expression jump = instructions[-1] jump_addr = last_addr destination = get_jump_destination(jump_addr, jump) # A condition is inverted during the preparation phase above warp.false_target = state._warp_in_block(destination) warp.true_target = state._warp_in_block(jump_addr + 1) return warp, 2
def _blockenize(state, instructions):
# Fix inverted comparison expressions (e.g. 0 < variable):
_fix_inverted_comparison_expressions(state, instructions)
# Fix "repeat until true" encapsulated by another loop
_fix_broken_repeat_until_loops(state, instructions)
# Fix "var_1 = var_1 [comparison] var_2 and (operation) var_1 or var_1" edge case
_fix_broken_unary_expressions(state, instructions)
addr = 1
# Duplicates are possible and ok, but we need to sort them out
last_addresses = set()
while addr < len(instructions):
instruction = instructions[addr]
opcode = instruction.opcode
if opcode not in _WARP_INSTRUCTIONS:
addr += 1
continue
if opcode in _JUMP_WARP_INSTRUCTIONS:
destination = get_jump_destination(addr, instruction)
if opcode != ins.UCLO.opcode or destination != addr + 1:
last_addresses.add(destination - 1)
last_addresses.add(addr)
else:
last_addresses.add(addr)
addr += 1
last_addresses = sorted(list(last_addresses))
last_addresses.append(len(instructions) - 1)
# This could happen if something jumps to the first instruction
# We don't need "zero block" with function header, so simply ignore
# this
if last_addresses[0] == 0:
last_addresses.pop(0)
previous_last_address = 0
index = 0
for last_address in last_addresses:
block = nodes.Block()
block.index = index
block.first_address = previous_last_address + 1
block.last_address = last_address
state.blocks.append(block)
state.block_starts[block.first_address] = block
previous_last_address = last_address
index += 1
def _fix_broken_unary_expressions(state, instructions):
enumerated_instructions = enumerate(instructions)
for i, instruction in enumerated_instructions:
if i > 2 and instruction.opcode == ins.ISTC.opcode \
and ins.ADDVN.opcode <= instructions[i - 1].opcode <= ins.CAT.opcode:
# Search for a jump that precedes the ISTC op
leading_jump_found = False
for j in range(1, i):
if instructions[i - j].opcode == ins.JMP.opcode:
leading_jump_found = True
break
elif instructions[i - j].opcode not in range(
ins.ADDVN.opcode, ins.CAT.opcode):
break
# Make sure the preceding jump matches the destination of the ISTC op
instruction_destination = get_jump_destination(
i + 1, instructions[i + 1])
if instruction_destination == i + 2 and leading_jump_found:
# Additional jump edge case of an edge case when expression is in an else body
if not instruction_destination == get_jump_destination(
i - j, instructions[i - j]):
if instructions[i + 2].opcode == ins.JMP.opcode:
instruction_destination = get_jump_destination(
i + 2, instructions[i + 2])
if instruction_destination == get_jump_destination(
i - j, instructions[i - j]):
instructions[i - 1].A = instruction.A
# Remove the broken condition
_remove_instruction(state, instructions, i + 1)
_remove_instruction(state, instructions, i)
else:
instructions[i - 1].A = instruction.A
# Remove the broken condition
_remove_instruction(state, instructions, i + 1)
_remove_instruction(state, instructions, i)
def _build_conditional_warp(state, last_addr, instructions):
condition = instructions[-2]
condition_addr = last_addr - 1
warp = nodes.ConditionalWarp()
if condition.opcode in (ins.ISTC.opcode, ins.ISFC.opcode):
expression = _build_unary_expression(state, condition_addr, condition)
setattr(warp, "_slot", condition.A)
elif condition.opcode >= ins.IST.opcode:
expression = _build_unary_expression(state, condition_addr, condition)
setattr(warp, "_slot", condition.CD)
else:
expression = _build_comparison_expression(state, condition_addr,
condition)
warp.condition = expression
jump = instructions[-1]
jump_addr = last_addr
destination = get_jump_destination(jump_addr, jump)
# A condition is inverted during the preparation phase above
warp.false_target = state._warp_in_block(destination)
warp.true_target = state._warp_in_block(jump_addr + 1)
shift = 2
if destination == (jump_addr + 1) \
and condition.opcode not in (ins.ISTC.opcode, ins.ISFC.opcode):
# This is an empty 'then' or 'else'. The simplest way to handle it is
# to insert a Block containing just a no-op statement.
block = nodes.Block()
block.first_address = jump_addr + 1
block.last_address = block.first_address
block.index = warp.true_target.index
block.warpins_count = 1
setattr(block, "_last_body_addr", block.last_address - shift)
block.warp = nodes.UnconditionalWarp()
block.warp.type = nodes.UnconditionalWarp.T_FLOW
block.warp.target = warp.true_target
setattr(block.warp, "_addr", block.last_address - shift + 1)
state.blocks.insert(state.blocks.index(warp.true_target), block)
warp.true_target = block
_create_no_op(state, jump_addr, block)
return warp, shift
def _blockenize(state, instructions):
"将指令划分成代码块"
addr = 1
# Duplicates are possible and ok, but we need to sort them out
last_addresses = set()
while addr < len(instructions):
instruction = instructions[addr]
opcode = instruction.opcode
if opcode not in _WARP_INSTRUCTIONS:
addr += 1
continue
if opcode in _JUMP_WARP_INSTRUCTIONS: # 跳转：当前指令地址，及跳转目的地的前一地址为一个代码块的末端
destination = get_jump_destination(addr, instruction)
if opcode != ins.UCLO.opcode or destination != addr + 1:
last_addresses.add(destination - 1)
last_addresses.add(addr)
else:
last_addresses.add(addr)
addr += 1
last_addresses = sorted(list(last_addresses))
last_addresses.append(len(instructions) - 1)
# This could happen if something jumps to the first instruction
# We don't need "zero block" with function header, so simply ignore
# this
if last_addresses[0] == 0:
last_addresses.pop(0)
previous_last_address = 0
index = 0
for last_address in last_addresses:
block = nodes.Block()
block.index = index
block.first_address = previous_last_address + 1
block.last_address = last_address
state.blocks.append(block)
state.block_starts[block.first_address] = block
previous_last_address = last_address
index += 1
def _blockenize(state, instructions): addr = 1 # Duplicates are possible and ok, but we need to sort them out last_addresses = set() while addr < len(instructions): instruction = instructions[addr] opcode = instruction.opcode if opcode not in _WARP_INSTRUCTIONS: addr += 1 continue if opcode in _JUMP_WARP_INSTRUCTIONS: destination = get_jump_destination(addr, instruction) if opcode != ins.UCLO.opcode or destination != addr + 1: last_addresses.add(destination - 1) last_addresses.add(addr) else: last_addresses.add(addr) addr += 1 last_addresses = sorted(list(last_addresses)) last_addresses.append(len(instructions) - 1) # This could happen if something jumps to the first instruction # We don't need "zero block" with function header, so simply ignore # this if last_addresses[0] == 0: last_addresses.pop(0) previous_last_address = 0 index = 0 for last_address in last_addresses: block = nodes.Block() block.index = index block.first_address = previous_last_address + 1 block.last_address = last_address state.blocks.append(block) state.block_starts[block.first_address] = block previous_last_address = last_address index += 1
def _build_unconditional_warp(state, addr, instruction):
warp = nodes.UnconditionalWarp()
warp.type = nodes.UnconditionalWarp.T_JUMP
opcode = instruction.opcode
warp.is_uclo = opcode == ins.UCLO.opcode
if warp.is_uclo and instruction.CD == 0:
# Not a jump
return _build_flow_warp(state, addr, instruction)
else:
destination = get_jump_destination(addr, instruction)
warp.target = state._warp_in_block(destination)
return warp, 1
def _build_unconditional_warp(state, addr, instruction): warp = nodes.UnconditionalWarp() warp.type = nodes.UnconditionalWarp.T_JUMP opcode = instruction.opcode warp.is_uclo = opcode == ins.UCLO.opcode if warp.is_uclo and instruction.CD == 0: # Not a jump return _build_flow_warp(state, addr, instruction) else: destination = get_jump_destination(addr, instruction) warp.target = state._warp_in_block(destination) return warp, 1
def _build_numeric_loop_warp(state, addr, instruction):
warp = nodes.NumericLoopWarp()
base = instruction.A
warp.index = _build_slot(state, addr, base + 3)
warp.controls.contents = [
_build_slot(state, addr, base + 0), # start
_build_slot(state, addr, base + 1), # limit
_build_slot(state, addr, base + 2) # step
]
destination = get_jump_destination(addr, instruction)
warp.body = state._warp_in_block(destination)
warp.way_out = state._warp_in_block(addr + 1)
return warp, 1
def _shift_warp_destinations(state, instructions, shift, modified_index):
for current_index, moved_instruction in enumerate(instructions):
opcode = moved_instruction.opcode
if opcode in _WARP_INSTRUCTIONS:
if current_index < modified_index and moved_instruction.CD >= 0:
destination = get_jump_destination(current_index,
moved_instruction)
if destination > modified_index or (
destination == modified_index and shift > 0):
moved_instruction.CD += shift
elif current_index >= modified_index and moved_instruction.CD < 0:
destination = current_index + moved_instruction.CD - shift + 1
if destination < modified_index or (
destination == modified_index and shift > 0):
moved_instruction.CD -= shift
def _build_numeric_loop_warp(state, addr, instruction): warp = nodes.NumericLoopWarp() base = instruction.A warp.index = _build_slot(state, addr, base + 3) warp.controls.contents = [ _build_slot(state, addr, base + 0), # start _build_slot(state, addr, base + 1), # limit _build_slot(state, addr, base + 2) # step ] destination = get_jump_destination(addr, instruction) warp.body = state._warp_in_block(destination) warp.way_out = state._warp_in_block(addr + 1) return warp, 1
def _establish_warps(state, instructions):
state.blocks[0].warpins_count = 1
enumerated_blocks = enumerate(state.blocks[:-1])
for i, block in enumerated_blocks:
if state.blocks.__contains__(block) is None:
continue
state.block = block
end_addr = block.last_address + 1
start_addr = max(block.last_address - 1, block.first_address)
# Catch certain double unconditional jumps caused by logical primitives in expressions:
if start_addr == (end_addr - 1) \
and end_addr + 1 < len(instructions) \
and instructions[start_addr].opcode == ins.JMP.opcode \
and instructions[end_addr].opcode == ins.JMP.opcode \
and instructions[start_addr].A == instructions[end_addr].A \
and instructions[start_addr].CD == 0:
end_instruction_destination = end_addr + instructions[
end_addr].CD + 1
target_instruction_A = instructions[start_addr].A
exit_instruction_found = False
# When two consecutive jumps are found with the same A operand, lookahead for the end jump.
following_destination = -1
for j in range(end_addr + 1, len(instructions) - 1):
following_instruction = instructions[j]
if following_instruction.opcode == ins.JMP.opcode:
if following_instruction.A == target_instruction_A:
following_destination = get_jump_destination(
j, following_instruction)
exit_instruction_found = True
break
# If we find the exit jump and we're not skipping it (if true then break else),
# form the original two jumps into a fake conditional warp.
if exit_instruction_found \
and end_instruction_destination <= following_destination:
fixed_instruction = ins.ISF()
fixed_instruction.CD = ins.SLOT_FALSE
instructions[start_addr] = fixed_instruction
state.blocks.pop(state.blocks.index(block) + 1)
block.last_address += 1
start_addr = max(block.last_address - 1, block.first_address)
end_addr = block.last_address + 1
warp = instructions[start_addr:end_addr]
block.warp, shift = _build_warp(state, block.last_address, warp)
setattr(block, "_last_body_addr", block.last_address - shift)
setattr(block.warp, "_addr", block.last_address - shift + 1)
last_block = state.blocks[-1]
last_block.warp = nodes.EndWarp()
setattr(last_block, "_last_body_addr", last_block.last_address)
setattr(last_block.warp, "_addr", last_block.last_address)
def _fix_broken_repeat_until_loops(state, instructions):
enumerated_instructions = enumerate(instructions)
for i, instruction in enumerated_instructions:
if instruction.opcode == ins.LOOP.opcode:
# Check for the conditional jump that restarts the loop
loop_exit_addr = get_jump_destination(i, instruction)
loop_condition_addr = loop_exit_addr - 1
loop_condition_instruction = instructions[loop_condition_addr]
if not loop_condition_instruction.opcode == ins.JMP.opcode:
if get_jump_destination(loop_condition_addr,
loop_condition_instruction) <= i:
continue
# It's not there, so this is probably a repeat-until true loop.
# We need a fake conditional warp that is treated as 'true' by the writer
fixed_cond_instruction = ins.ISF()
fixed_cond_instruction.CD = ins.SLOT_TRUE
# Resulting jump to the loop starting point
fixed_jump_instruction = ins.JMP()
fixed_jump_instruction.CD = i - loop_condition_addr - 1
# Add fake conditional instructions
insertion_index = loop_condition_addr + 1
_insert_instruction(state, instructions, insertion_index,
fixed_jump_instruction)
_insert_instruction(state, instructions, insertion_index,
fixed_cond_instruction)
shift = 2
# Fix non-break destinations within the loop
# Breaks in the empty-condition loop point towards the same exit destination
# as non-breaks, so we'll have to search for a pattern of jumps.
leading_jump = False
start_index = i + 1
for j in range(start_index, insertion_index):
checked_instruction = instructions[j]
# Look for following JMP instructions
if checked_instruction.opcode == ins.JMP.opcode:
# Leading jump indicates this is a break?
if not leading_jump:
checked_instruction_destination \
= get_jump_destination(j, checked_instruction)
# If the destination would've been moved
if checked_instruction.CD >= shift \
and checked_instruction_destination == insertion_index + shift:
# Check for an inverted jump pair
next_index = j + 1
following_instruction = instructions[
next_index]
if following_instruction.opcode == ins.JMP.opcode:
following_destination \
= get_jump_destination(next_index, following_instruction)
# e.g. goto 277 followed directly by goto 176
if following_destination < checked_instruction_destination:
leading_jump = True
continue
# e.g. goto 277 followed directly by goto 277
elif following_destination == checked_instruction_destination:
leading_jump = False
continue
# Check for else-break-end following this jump
following_else_break_found = False
prev_jump = False
for k in range(next_index, insertion_index):
following_instruction = instructions[k]
if following_instruction.opcode == ins.JMP.opcode:
if not prev_jump:
prev_jump = True
else:
following_destination \
= get_jump_destination(k, following_instruction)
# Don't adjust the checked jump, it's probably a break
if following_instruction.CD >= shift \
and following_destination \
== checked_instruction_destination:
following_else_break_found = True
break
prev_jump = False
else:
if prev_jump:
last_destination \
= get_jump_destination(k - 1, instructions[k - 1])
# We can adjust, it's probably not a break
if last_destination < checked_instruction_destination:
break
prev_jump = False
if not following_else_break_found:
checked_instruction.CD -= shift
leading_jump = True
else:
leading_jump = False