Dear Jakob and all, I've just experienced this problem too. It seems to be a compiler bug with recent icc (it's still there in 2018.1.163). Python 3.6.x now includes sha3 which used to be a pypi module - and there's one tiny bit of code in there that creates a huge line in C using a macro. Icc segfaults on that line, presumably because it can't fit it in a buffer somewhere.
Anyway, I've made a patch (attached) which I'll be contributing back along with a Python-3.6.3-iomkl-2018a.eb in the next couple of weeks. If anyone fancies sanity checking the patch then I'd be very grateful! It does make some assumptions about which version of the macro to use (which it just unpacks explicitly into several lines) so someone might have a better idea. Even then, it still needs '-O0' to build it... Thanks, Ed -- Andrew Edmondson Advanced Application Specialist and Research Software Group Leader Research Computing, IT Services, University of Birmingham Web: - BEAR Software: http://www.birmingham.ac.uk/bear-software - BEAR: http://www.birmingham.ac.uk/bear - Support: http://itservicedesk.bham.ac.uk/ Twitter: https://twitter.com/uob_rescomp Email: [email protected] Tel: 58383 or 0121 4158383 -----Original Message----- From: [email protected] [mailto:[email protected]] On Behalf Of [email protected] Sent: 13 October 2017 07:51 To: [email protected] Subject: [easybuild] Building Python with iomkl (intel) tool chain Hi, I am trying to build Python with the Intel compiler, using the iomkl/2017b toolchain since we do not have a licence for Intel MPI, and therefore cannot use the intel toolchain. As the .eb files are identical for foss/2017b and intel/2017b (a part from the toolchain, of couse), I just made a .eb file using the iomkl chain. Unfortunately, Python does not build correctly, it fails during sanity checks, as some of the extensions have not built correctly (the first it finds is paramiko). The day before I built Python using the foss/2017b toolchain without any problems whatsoever. But it would not be the first time something fails with the Intel compiler, because it may be more picky. Has anybody had any success building Python-3.6.2-intel-2017b.eb from the develop branch? Using iomkl should give exactly the same compiler. I looked at the log files, there are a number of exceptions when building the paramiko module that are not seen in the log of the foss/2017b build. They relate to missing symbols in openssl ERROR:root:code for hash sha3_224 was not found. Traceback (most recent call last): File "/home/niflheim/schiotz/easybuild/nehalem/software/Python/3.6.2-iomkl-2017b/lib/python3.6/hashlib.py", line 121, in __get_openssl_constructor f = getattr(_hashlib, 'openssl_' + name) AttributeError: module '_hashlib' has no attribute 'openssl_sha3_224' During handling of the above exception, another exception occurred: Traceback (most recent call last): File "/home/niflheim/schiotz/easybuild/nehalem/software/Python/3.6.2-iomkl-2017b/lib/python3.6/hashlib.py", line 243, in <module> globals()[__func_name] = __get_hash(__func_name) File "/home/niflheim/schiotz/easybuild/nehalem/software/Python/3.6.2-iomkl-2017b/lib/python3.6/hashlib.py", line 128, in __get_openssl_constructor return __get_builtin_constructor(name) File "/home/niflheim/schiotz/easybuild/nehalem/software/Python/3.6.2-iomkl-2017b/lib/python3.6/hashlib.py", line 113, in __get_builtin_constructor raise ValueError('unsupported hash type ' + name) ValueError: unsupported hash type sha3_224 ERROR:root:code for hash sha3_256 was not found. Traceback (most recent call last): File "/home/niflheim/schiotz/easybuild/nehalem/software/Python/3.6.2-iomkl-2017b/lib/python3.6/hashlib.py", line 121, in __get_openssl_constructor f = getattr(_hashlib, 'openssl_' + name) AttributeError: module '_hashlib' has no attribute 'openssl_sha3_256' During handling of the above exception, another exception occurred: [ And then the same is repeated for another SHA3 variant, and another, and … The weird thing is that if I load Python built with foss/2017b, then hashlib._hashlib is missing these SHA3 variants, too, but the modules have built properly. Does anybody know how to handle this? Or has anybody already managed to build Python-3.6.2-intel-2017b.eb? Best regards Jakob -- Jakob Schiøtz, professor, Ph.D. Department of Physics Technical University of Denmark DK-2800 Kongens Lyngby, Denmark http://www.fysik.dtu.dk/~schiotz/
Python-3.6.3-icc_segfault_workaround.patch
Description: Python-3.6.3-icc_segfault_workaround.patch
