Summary: | dev-lang/python-3.6.5 emerge fails: gcc 7.3.0 "internal compiler error: Segmentation fault" | ||
---|---|---|---|
Product: | Gentoo Linux | Reporter: | Andrea Zuccherelli <andrea.zuccherelli> |
Component: | Hardened | Assignee: | Gentoo Toolchain Maintainers <toolchain> |
Status: | RESOLVED FIXED | ||
Severity: | normal | CC: | alix.gentoo, python |
Priority: | Normal | Keywords: | PATCH |
Version: | unspecified | ||
Hardware: | All | ||
OS: | Linux | ||
URL: | https://gcc.gnu.org/PR83317 | ||
See Also: | https://gcc.gnu.org/bugzilla/show_bug.cgi?id=83317 | ||
Whiteboard: | |||
Package list: | Runtime testing required: | --- | |
Attachments: | Output of "emerge -1v =dev-lang/python-3.6.5" |
Description
Andrea Zuccherelli
2018-07-26 20:38:04 UTC
Revert back to gcc-6.4.0: no failure python-3.6.5 & gcc-7.3.0, got the same build error. error doesn't seem to be hardened-specific: it's reproducible on my other system with default/linux/x86/17.0 (stable) profile. builds fine on a x86_64 system, so this is specific to x86 arch... the bug I referenced above might be related, but I've never had a gcc pgo flag enabled on any of my x86 systems that are having the problem. so, the problem in my case is in the -mfpmath=sse along with -msse2 and -mssse3 CFLAGS. removing -msse2 and -mssse3 stops gcc 7.3.0 from segfaulting in cmathmodule.c. -march is set to i686. Andrea Zuccherelli, replace march=prescott with march=i686 in your CFLAGS; should help. (why do you have a prescott march set on an ivy bridge cpu anyway?) Hi Alix. Thank you for your analysis: GCC is infact failing in 'cmath_acos' and for that is probably using SSE optimizations. For the "prescott" architecture: I am running Gentoo in a template Virtualbox VM (so the i7 3632QM is the real CPU where the VM is running on). I have two old ITX boards based on Atom N270 (my router and my NAS). From time to time when the template is stable and bug fixed I convert it to run as a live overlay and then transfer the image to the device CF. I used "prescott" because in 2010 Gentoo suggested that CFLAGS for ATOM N270. That wiki is no more available online, so I report the link to Internet Archive: https://web.archive.org/web/20100629050418/http://en.gentoo-wiki.com/wiki/Safe_Cflags/Intel#Atom_230 For now I will revert back to GCC 6.4.0, hoping and waiting for a solution of this bug in the GCC 7 line. looks like it should be fixed in gcc 7.4: https://gcc.gnu.org/bugzilla/show_bug.cgi?id=83317 Please check if gcc-8 works for you. no errors with python-3.6.5 on gcc-8.2.0 (x86 system) Thank you for the test! Closing as RESOLVED. |