search

zrax / pycdc

C++ python bytecode disassembler and decompiler
GNU General Public License v3.0
3.24k stars 623 forks source link

pycdc error #486

Closed mehedih4x closed 1 month ago

mehedih4x commented 4 months ago

pycdc error(segmentation fault) but pycdas bytecode decompile successfully. why??

greenozon commented 4 months ago

those are different apps please provide more details of your failure case

Levak commented 3 months ago

I was also getting consistent segmentation faults when trying to decompile python 3.10 and 3.11. it's not surprising considering it's not fully if not at all supported yet.

Afair it was segfaulting trying to dereference an empty stack. Regardless, the decompiled code would be false even if the invalid access was protected by a check, since I'm pretty sure op was also trying to decompile python 3.9+

tjemg commented 1 month ago

I also get segmentation fault if I run the tests. However, these issues only occur for the Release build - in particular, turning on any optimization (i.e. at least -O1), results in segmentation fault with the checks:

...

*** class_NODE_BINARY: PASS (1)
/home/gasiba/Downloads/git/pycdc/tests/decompyle_test.sh: line 41: 56934 Segmentation fault      (core dumped) ./pycdc "$pyc" -o "$base.src.py" > "$base.err" 2>&1
*** test_exceptions: XFAIL (4)
/home/gasiba/Downloads/git/pycdc/tests/decompyle_test.sh: line 41: 56941 Segmentation fault      (core dumped) ./pycdc "$pyc" -o "$base.src.py" > "$base.err" 2>&1
*** chain_assignment: FAIL (2 of 2)
    chain_assignment.2.7.pyc

    chain_assignment.3.7.pyc

...

*** unicode_future: PASS (4)
/home/gasiba/Downloads/git/pycdc/tests/decompyle_test.sh: line 41: 57476 Segmentation fault      (core dumped) ./pycdc "$pyc" -o "$base.src.py" > "$base.err" 2>&1

...

*** lambdas_assignment: PASS (1)
/home/gasiba/Downloads/git/pycdc/tests/decompyle_test.sh: line 41: 58108 Segmentation fault      (core dumped) ./pycdc "$pyc" -o "$base.src.py" > "$base.err" 2>&1
greenozon commented 1 month ago

Very interesting! what Linux distro do you run? what python ver

tjemg commented 1 month ago

This problem first occurred when I was porting the software to FreeBSD (compilation with clang/clang++), and I reproduced the issue on Slackware 15 (fresh install from ISO). I can check if the same issue occurs on Devuan...

tjemg commented 1 month ago

OK, just checked it now - problem also happens in Devuan 5.0.1 Steps to reproduce

mkdir build
cd build
cmake -GNinja -DCMAKE_CXX_FLAGS="-O2" -DCMAKE_BUILD_TYPE=Release ..
ninja -v
ninja check.... Boom!

FreeBSD: python version 3.9.19 Slackware: python version 2.7.18 + python 3.9.10 Devuan: python version 3.11.2

Forgot to mention explicitly, if I turn optimization to -O0, and run the checks, all is fine! So, my best guess it that there is some UB in the code, and this does not depend on the python version - these symptoms are a typical effect thereof (i.e. observable behavior differs between optimized and non-optimized code).

greenozon commented 1 month ago

Thanks for details! I agree, that debug builds are successfully passing the test cases run...

image (2)

but release build (even wihtout -O2 flag did crash a couple of times with coredumps!

I'm using FreeBSD 14.1

BTW, the default shell in above OS is /bin/sh and it is not possible to run the tests, also if you install bash it'll be put not into /bin/bash but intlo /usr/local/bin/bash...

@zrax is it OK to make shebang more generic and cover majoriyt of OSes?

!/bin/bash

->

!/usr/bin/env bash

here is a small PR if you agree https://github.com/zrax/pycdc/pull/506