search

paulromano / endf-python

Python ENDF Parser
https://endf-python.readthedocs.io
MIT License
25 stars 6 forks source link

Fix issue with C++ endf float converter #7

Closed aidancrilly closed 5 months ago

aidancrilly commented 5 months ago

This closes #6 - at least for me, I noticed someone having similar issues in #4.

The crash seems to be a seg fault that occurs when the endf float being converted is negative, i.e. has a leading minus sign.

Positive numbers have leading whitespace so hit the continue statement and j is not incremented. Therefore, the first digit is placed at arr[0]. However, negative numbers had the leading minus sign placed at arr[0] and first digit at arr[1]. This means when the null terminator is placed at the end, for negative numbers this is at arr[12] so the crash occurs.

EDIT: my initial commit was not the correct solution, I think it is simply a case of increasing arr to be length 13 so it has space for both the e and null terminator if the input buffer is length 11. Apologies, some how my initial incorrect fix ran (I have no idea how!) but on restarting it failed.

This would seem to me to be an issue across systems.

paulromano commented 5 months ago

Thanks @aidancrilly! Can you give a specific example of an 11-character string that triggers the problem?

aidancrilly commented 5 months ago

@paulromano Sure, in ENDF/B-VIII.0 incident neutron data for deuterium there is -2.225002+6 at the beginning of the MF = 3, MT = 16 record. This is the QM/QI value.

paulromano commented 5 months ago

@aidancrilly Would you be up for making a similar PR to OpenMC? The cfloat_endf function was originally borrowed from there: https://github.com/openmc-dev/openmc/blob/develop/openmc/data/endf.c

It would be good to also add the example you mentioned (-2.225002+6) to the tests here: https://github.com/openmc-dev/openmc/blob/develop/tests/unit_tests/test_endf.py