glibc 2.25 has a nasty regression in one of its 32-bit SSE2 implementations of memchr that affects at least older 32-bit Atom systems, and presumably any other 32-bit x86 that uses the same memchr variant. (See linked upstream bug report.) The symptoms I've noticed are emerge immediately segfaulting every time in libpython and awk failing with an internal error, but it almost certainly causes other problems too. Tested with glibc-2.25-r3, gcc 6.4.0, gentoo-sources 4.12.0 on Atom N270. The upstream patch https://sourceware.org/git/?p=glibc.git;a=commit;h=ccb4fd7a657b0fbc4890c98f4586d58a135fc583 fixes the issue for me, though fixing an affected system is complicated by the fact that the bug breaks both emerge and the actual glibc build process itself.
Pushed as glibc-2.25-r4 (patchset 8): https://gitweb.gentoo.org/repo/gentoo.git/commit/?id=457d3ddf64abb388ad418793d5b0a9f5dfd80ee8 Thanks for the detailed report!