Aleksa Sarai

@poettering Sorry for pinging you directly. This fixes the issue we discussed when we met a few weeks ago (you were right that the format is valid, but casync's output...

/cc @keszybz @bluca

umoci uses 0755 and the root UIDs, btw. We also set the ctime and mtime to the Unix epoch (to have at least some hope of producing a reproducible container)...

> however I still have not considered atime and ctime as they are not mentioned elsewhere in the existing spec. Does anyone think we should consider those as well? `atime`...

@sparr At the moment the behaviour is completely unspecified so different image-spec implementations can produce different results. If you feel that your suggested algorithm makes more sense, then you should...

There were similar issues with systemd communication via dbus caused by the way runc was protecting the executable against attacks. #3931 should have fixed this issue -- you can test...

I would expect it to use a different media-type, but the main takeaway I had from the whole artefacts saga is that public registries are so strict about media-types (which...

> Was thinking of using a list of known offsets first, before resorting to linear scan. Can try different distros to get those offsets (I'm pretty sure there are only...

@kolyshkin I went with writing dummy versions of the structure layouts so we can just use `sizeof` and `offsetof`. Would you prefer it just be a list of magic constants...

On musl, it seems they: 1. Use `CLONE_CHILD_CLEARTID` for some TLS magic to indicate when a thread has died, which means that `PR_GET_TID_ADDRESS` will return the wrong address. 2. Don't...