zrax
Support for JUMP_BACKWARD
According to this article the JUMP_BACKWARD was a replacement for the same opcode JUMP_ABSOLUTE https://bugs.python.org/issue47120
Test Case
def JUMP_BACKWARD():
iterable = [1, 2, 3]
for item in iterable:
pass
while True:
break
pycdas for TC
[Code]
File Name: jmp_back.py
Object Name: <module>
Qualified Name: <module>
Arg Count: 0
Pos Only Arg Count: 0
KW Only Arg Count: 0
Stack Size: 1
Flags: 0x00000000
[Names]
'JUMP_BACKWARD'
[Locals+Names]
[Constants]
[Code]
File Name: jmp_back.py
Object Name: JUMP_BACKWARD
Qualified Name: JUMP_BACKWARD
Arg Count: 0
Pos Only Arg Count: 0
KW Only Arg Count: 0
Stack Size: 2
Flags: 0x00000003 (CO_OPTIMIZED | CO_NEWLOCALS)
[Names]
[Locals+Names]
'iterable'
'item'
[Constants]
None
(
1
2
3
)
[Disassembly]
0 RESUME 0
2 BUILD_LIST 0
4 LOAD_CONST 1: (1, 2, 3)
6 LIST_EXTEND 1
8 STORE_FAST 0: iterable
10 LOAD_FAST 0: iterable
12 GET_ITER
14 FOR_ITER 2 (to 20)
18 STORE_FAST 1: item
20 JUMP_BACKWARD 4 (to 14)
22 END_FOR
24 NOP
26 RETURN_CONST 0: None
None
[Disassembly]
0 RESUME 0
2 LOAD_CONST 0: <CODE> JUMP_BACKWARD
4 MAKE_FUNCTION 0
6 STORE_NAME 0: JUMP_BACKWARD
8 RETURN_CONST 1: None
before
def JUMP_BACKWARD():
Unsupported opcode: JUMP_BACKWARD
iterable = [
1,
2,
3]
# WARNING: Decompyle incomplete
after
def JUMP_BACKWARD():
Unsupported opcode: END_FOR
iterable = [
1,
2,
3]
# WARNING: Decompyle incomplete
@zrax Do you think it makes sense to add support into this PR for the next missed operand?
PS the test was done on Python 3.12
I compiled with pycdc, and it doesn't work if you just add the line case Pyc::JUMP_BACKWARD_A:
thats what I got:
the source:
l = [ch for ch in 'hello']
python 3.11: (didn't work, because JUMP_BACKWARD is used, but implementation isn't exactly like JUMP_ABSOLUTE)
# Source Generated with Decompyle++
# File: test_311.pyc (Python 3.11)
l = 'hello'()
python 3.8: (still didn't work, but thats because of list comprehension and lambda together is not supported yet as you can see)
# Source Generated with Decompyle++
# File: test_38.pyc (Python 3.8)
l = (lambda .0: [ ch for ch in .0 ])('hello')
the bytecode is almost exactly the same, only difference is JUMP_BACKWARD instead of JUMP_ABSOLUTE
I mean that the bytecode of this source code
l = [ch for ch in 'hello']
between version 3.10 and 3.11 is this: 3.10:
1 0 LOAD_CONST 0 (<code object <listcomp> at 0x000001AEABDC8F50, file "test.py", line 1>)
2 LOAD_CONST 1 ('<listcomp>')
4 MAKE_FUNCTION 0
6 LOAD_CONST 2 ('hello')
8 GET_ITER
10 CALL_FUNCTION 1
12 STORE_NAME 0 (l)
14 LOAD_CONST 3 (None)
16 RETURN_VALUE
Disassembly of <code object <listcomp> at 0x000001AEABDC8F50, file "test.py", line 1>:
1 0 BUILD_LIST 0
2 LOAD_FAST 0 (.0)
>> 4 FOR_ITER 4 (to 14)
6 STORE_FAST 1 (ch)
8 LOAD_FAST 1 (ch)
10 LIST_APPEND 2
12 JUMP_ABSOLUTE 2 (to 4)
>> 14 RETURN_VALUE
3.11:
0 0 RESUME 0
1 2 LOAD_CONST 0 (<code object <listcomp> at 0x0000028633FF92E0, file "test.py", line 1>)
4 MAKE_FUNCTION 0
6 LOAD_CONST 1 ('hello')
8 GET_ITER
10 PRECALL 0
14 CALL 0
24 STORE_NAME 0 (l)
26 LOAD_CONST 2 (None)
28 RETURN_VALUE
Disassembly of <code object <listcomp> at 0x0000028633FF92E0, file "test.py", line 1>:
1 0 RESUME 0
2 BUILD_LIST 0
4 LOAD_FAST 0 (.0)
>> 6 FOR_ITER 4 (to 16)
8 STORE_FAST 1 (ch)
10 LOAD_FAST 1 (ch)
12 LIST_APPEND 2
14 JUMP_BACKWARD 5 (to 6)
>> 16 RETURN_VALUE
if you look, there is not very much difference in the bytecode, but the decompiled source code for these pyc's are different, which means that another thing has to be done, because even after adding the line you added, its still not supported.
i would suggest to add, after the lines:
if (mod->verCompare(3, 10) >= 0)
offs *= sizeof(uint16_t); // // BPO-27129
to add this lines, for it to work:
if (opcode == Pyc::JUMP_BACKWARD_A)
offs = pos - offs;
then this way it is exactly like the jump absolute, and it will work, then you can compile the test file "test_for_loop_py3.8.py" for version 3.11 and add it to the compiled folder, i saw there is return None in the end, so maybe also get the whole code inside a function and remove the return statement, after that compile it for 3.10,3.11, tokenize by the token_dump script in the scripts folder and thats all i think.
Proposal:
case Pyc::JUMP_BACKWARD_A:
case Pyc::JUMP_FORWARD_A:
case Pyc::INSTRUMENTED_JUMP_FORWARD_A:
{
int offs = operand;
if (mod->verCompare(3, 10) >= 0)
offs *= sizeof(uint16_t); // // BPO-27129
if (opcode == Pyc::JUMP_BACKWARD_A)
offs *= -1;
