Improve gcroots check & fix CI #111
Conversation
Profpatsch
force-pushed
the
print-warning-for-missing-gcroot-dir
branch
from
4e435a2 to
d2b644c
Compare
Profpatsch
changed the title
Profpatsch
force-pushed
the
print-warning-for-missing-gcroot-dir
branch
from
d2b644c to
c448b51
Compare
|
Looking into this I've found that the per-user gcroots have been essentially deprecated; something about a conflict with a non-root nix builder? The upshot does seem to be that the preference is for any manipulation of the nix gcroots be managed by nix tools or the nix daemon, which seems sensible to me. Elsewhere, I was taking a stab at doing that with |
|
I don’t think it matters very much, nix upstream is a mess. I’d rather just improve what is there right now and worry about it when upstream decides to break us. |
Profpatsch
force-pushed
the
print-warning-for-missing-gcroot-dir
branch
6 times, most recently
from
58bd5d6 to
46989e9
Compare
And remove the weird extra 22.05 target, why was this there in the first place? Nobody knows!
Instead of only checking whether we can create the user gcroot when we actually want to create a root in it, we check once in the beginning of setting up lorri and then pass the path through, assuming it exists. This improves the error message when the necessary setup is not yet there (it’s an enironment problem which can be fixed on the first start of lorri).
Lorri needs this to create gcroots for evaluations.
On MacOS, the `std` type of st_mode is `u32`, but the libc S_IWOTH is u16, meaning we can’t really compare the two … So let’s fetch stat from libc as well.
dropping a pointer does nothing
This way it’s clearer which step fails, and it’s faster to scroll to the test output. We could(?) make every test into its own step at one point, that would be good I think. Given that we use nix for both, we could import the tests and parse their attributes from the list to generate our CI run.
Profpatsch
force-pushed
the
print-warning-for-missing-gcroot-dir
branch
from
46989e9 to
b890754
Compare
|
The one failing watcher test is annoying, I’m trying to figure out what’s going wrong there, but it’s gonna have to be a separate PR. So I’m merging this for now. |
release.nix(seerelease.nixfor instructions)