Conversation
|
I think I tried 1.6.0 (maybe even 1.6.1) and it had the same problem, namely that the clock in the VM is set to the boot time, rather than current time, of the host. So many Autoconf builds will fail because the sources are in the future. |
|
@edolstra OK. 1.6.0 was the first version with this problem. Is there a simple way to verify that is also happens in 1.6.1? |
|
You can test it as follows: This is with 1.6.1, so the problem still occurs. I'm using Linux 3.4.67 as the host kernel btw, that might matter. |
|
Could we use builtins.currentTime in the runInLinuxVM function somehow? |
|
Why don't we set the timestamp of sources back to year 1970 at the end of unpackPhase? (I actually thought nix already did that.) |
|
@edolstra It actually seems as if this is working correctly now. There must have been something other than qemu that caused the time issue. But now qemu 1.7.0 has been released, so I'll do a PR for that instead. |
qemu 1.5.2 no longer works for me. I get "Device or resource busy" when I try to create a qemu machine with libvirt (on Linux 3.11). qemu 1.6.1 works fine, though.
However, there was some problem last time I tried updating to 1.6.0. A lot of Hydra builds failed. @edolstra How can I verify that 1.6.1 isn't broken in the same way?