Make eBPF CMAKE robust against llvm check failures #1946
Conversation
backends/ebpf/CMakeLists.txt
Outdated
| Ignoring ebpf tests...") | ||
| set (SUPPORTS_KERNEL False) | ||
| endif() | ||
| else() | ||
| message(STATUS "LLVM missing, disabling kernel tests...") | ||
| message(WARNING | ||
| "Problem with LLVM, disabling kernel tests...\n" |
There was a problem hiding this comment.
This message could be more user-friendly.
Perhaps "Could not detect a version of LLVM which is needed to compile the eBPF kernel tests".
Did you test this on both systems that have working and broken llvms?
There was a problem hiding this comment.
Yes, I tested it on an Ubuntu Xenial image with the broken llvm package and on my local version. Both worked fine. The Travis test also passed. This method does not seem to work on macOS, but we do not support that anyway as far as I remember.
There was a problem hiding this comment.
On Mac we don't support eBPF, but the build should still work, of course.
Travis will test on Mac, so I think that if that passes (and it did previously) then we can merge this.
There was a problem hiding this comment.
Okay, it should pass, it will just complain that it cannot find llvm, which also might be just a setup issue in the macOS Travis image.
This pull request is intended to introduce a workaround for #1376.
Instead of using CMAKE's
find_packagewe manually check the LLVM version and compare. This prevents crashes with specific versions of LLVM and Ubuntu.The downside of this method is that it does not catch versions of LLVM that were installed (via
apt install llvm-3.8for example) and disables the eBPF backend despite an LLVM version that exists.However, because the eBPF framework uses
llcanyway instead of, say,llc-3.8it is probably for the better that this discrepancy is caught.Let me know what you think.