Open tylerjereddy opened 3 years ago
Was built on andes
node with spack
as follows:
cd /ccs/home/treddy/github_projects/spack
git checkout 2421d90
. share/spack/setup-env.sh
spack install --no-checksum memsurfer +osmesa %gcc@8.3.1
This may be related to the poisson reconstruction python binding interface https://github.com/mmolero/pypoisson/issues/13
@tylerjereddy , thanks for opening the issue and @ctlee , thanks for connecting it with the pypoisson issue.
There appears to be a memory bug in the original code for python reconstruction that slips through the cracks in older compilers, but it appears that gcc>8 is not letting it through. I spent quite some time digging down the cause, and I suspect really fixing it requires an overhaul of the original cpp code (see my comment here.)
Until then, I tried the code again with gcc@7.5.0
and I was able to successfully run.
I will leave this issue open but add to ReadMe the need to stick with gcc@7
until someone figures this out.
Probably there should never be
double free
C++ errors, regardless of the input data, if the memory handling is safe throughout the library and its dependency stack.Here is a reproducer using nothing more than NumPy,
pytest
, and thehypothesis[numpy]
property-based testing library:Run the test below with version
1.0
ofmemsurfer
with:pytest reproducer.py -s
It will hard crash: