Erik Sipsma

Yeah starting with no TTL is even simpler, so I'm good with starting there too. > I am happy to cut the TTL feature altogether from the initial implementation. I...

> Since we don't yet have consensus on how to handle TTLs, all the more reason to cut it from the first phase of implementation :) I will update the...

> But couldn't this example be implemented by the function author setting the TTL to the underlying IO operation? Be it Container.withExec, http(), git() and so on. That way, the...

I surprisingly ended up needing to do some anticipatory handling of cases that will become possible once we enable this as part of the secret/socket isolation: https://github.com/dagger/dagger/pull/7804#issuecomment-2216493990 * Just worth...

@marcosnils that SGTM, I will hold off on merging this until right before v0.14.0 to limit the churn time of the temporary name

FWIW I am on linux/arm64 and never saw this happen locally (despite running lots of git related tests lately that sound like they made this common for others on x86)....

More details in the docker thread, but I opened an issue w/ go here after finding an isolated repro: https://github.com/golang/go/issues/73141

> Are there important cache implications? This came up a long time ago and at the time it did indeed seem like it created more cache invalidations since the .git...

Couple more recent occurrences: * https://github.com/dagger/dagger/actions/runs/13427524763/job/37513152438#step:5:7108 * https://github.com/dagger/dagger/actions/runs/13447234632/job/37575134036#step:5:7239 This could be 100% illusory, but for whatever reason it feels like this happens much more often on main runs than PRs....

Something interesting just happened locally *possibly* related to this. I just had a compile error in my engine code, but the end of the progress output for building the engine...