It would be nice to convert the two messages in the output .spk into capnproto canonical form. This would reduce the number of things that might cause a package to not be reproducible, since it means that e.g. if we upgrade to a newer version of the capnproto library, changes in how it allocates values inside of a message will affect the final output. We already sort the names of files in a directory, so value-significant details of a package archive should already only depend on the actual filesystem in the docker image. Besides this, the only other variable is the details of what our xz compression library does.
The canonicalization function strips off the message header, returning just a segment, so we will have to add that back to conform to the spk format, but this is easy enough to do by hand.
It would be nice to convert the two messages in the output .spk into capnproto canonical form. This would reduce the number of things that might cause a package to not be reproducible, since it means that e.g. if we upgrade to a newer version of the capnproto library, changes in how it allocates values inside of a message will affect the final output. We already sort the names of files in a directory, so value-significant details of a package archive should already only depend on the actual filesystem in the docker image. Besides this, the only other variable is the details of what our xz compression library does.
The canonicalization function strips off the message header, returning just a segment, so we will have to add that back to conform to the spk format, but this is easy enough to do by hand.