spack: gcc or binutils seems to have been broken recently
My daily build two nights ago worked. A fresh clone of github.com/llnl/spack late yesterday afternoon failed to build, as did my build last night.
I’m building gcc@5.4.0 with the CentOS system compiler, gcc@4.8.[35].
The compiler appears to be failing.
Here’s a bit from a config.log file. The interesting bit is n--eh-frame-hdr, it looks as if there might be a stray ‘n’ somewhere that’s getting pasted onto the front of the --eh-frame-hdr switch:
configure:3851: /scratch/hartzelg/working-dir/workspace/daily-build/spack/lib/spack/env/gcc/gcc conftest.c >&5
/blah/apps/spack/v0.0.10/opt/spack/linux-centos7-x86_64/gcc-4.8.5/binutils-2.28-66ufycihw5nglnjuy4xf6hqpwrru6bx4/bin/ld: cannot find n--eh-frame-hdr: No
such file or directory
collect2: error: ld returned 1 exit status
I’ve been fishing for the change, but perhaps it will ring a bell.
About this issue
- Original URL
- State: closed
- Created 10 years ago
- Comments: 26 (26 by maintainers)
LIne 253?
https://github.com/adamjstewart/spack/blob/83f324c7cf9b7d91120446462fd7311200ed4d20/var/spack/repos/builtin/packages/gcc/package.py#L253
@adamjstewart I think the string
'\\\n'translates to one\and one\nmeaning the backslash is probably some line continuation symbol.I don’t think that would make a difference. Like, there is a change, but the result should be exactly the same as before 😉
@hartzell That HAS to be it. I wasn’t sure how to convert
\\\nto its raw equivalent. I must have messed up. Let’s just revert that line. I’ll send out a PR for you to test.there were changes in the code (thus red/green), but the result should be exactly the same as before.
By “there are no changes” I mean “there are no changes that matter”. As in
'foo' == "foo"andprefix + '/lib' == prefix.lib.I’m not saying that there is anything meaningful here, but this is the bit that caught my eye: https://github.com/LLNL/spack/pull/4144/files#diff-38bf8d6101b9bc88270ac094f38565e1R199 (since what I’m seeing feels like something’s screwed up in the way that gcc’s building it’s command line for
ld).i think the only real change is
lib64->lib, otherwise I can’t really see how that PR can break something… But that would be my best guess as well…Is it caused by #4144?