Try to eliminate binutils on Linux
#1
Comments
|
devtoolset-2 should be new enough. It does strange things with the way it links against libgfortran and libstdc++ though, which may cause issues. It's easier than rebuilding a latest gcc from source which is the other alternative for that problem. |
|
Yeah, that is the plan for the near term. Though this issue has gotten a fair bit of discussion. It may be that devtoolset-2 is not enough because there are a couple C++14 use cases. To get a bit more history (and also feel free to weigh into the discussion), take a look at this issue ( conda-forge/conda-forge.github.io#29 ). Sorry, it's a long read. |
|
I'm working on a gcc-6-on-centos-6 docker image at the moment, which I'll put up once I can get it into a shape where it can build all of Julia successfully (which depends on openblas). |
|
As this started to get kind of long and felt more related with the aforementioned issue, I have added my response there. |
|
Attempting to remove the |
|
Worked great. So, we have dropped the |
* ENH use openmp on linux * Update meta.yaml * Update meta.yaml * Update build.sh * Update meta.yaml * Update build.sh * Update build.sh * MNT: Re-rendered with conda-build 3.18.11, conda-smithy 3.6.5, and conda-forge-pinning 2020.01.10 * try this * bugz for dayz * Fix deps (#1) Co-authored-by: Isuru Fernando <[email protected]>
This can't be done at present because a newer version was required to Linux to build at all (see this comment and the one below). According to this user, they were able to build with binutils 2.25.1. We have the binutils 2.26 package at present. As we are going to start using devtools-2, we will have a binutils version in-between the two mentioned (2.23.52). It is unclear if that is new enough to work here, but it would be worth trying. If it does work, we should drop the
binutilsdependency.The text was updated successfully, but these errors were encountered: