A new version reads files written by an old one. That is the part of a release that is hardest to test, because the data is not in the repository and the format change is invisible until someone's file is lost. Migration is a rule per version step, written down before the release.
About 1 min of reading
Detect the format.
The header
Before you run it, say out loud what it will print. Then open the answer.
The drop report
Before you run it, say out loud what it will print. Then open the answer.
Read the first example one line at a time. A line that prints shows what appears on screen; the rest set things up quietly.
A saved file needs its own format version. Without it, the loader is guessing from shape, and a shape that fits two formats is a file loaded wrongly with no error.
Keep a file from each past version in the test fixtures. Those files are the only honest test of a loader, and writing a new one instead of keeping an old one tests the wrong thing.
A loader detects the old format and converts it on read. That keeps every existing file working without touching it, which is safer than a one time rewrite pass over files you cannot enumerate.
A loader that handles 1.0 to 2.0 directly is a guess about every past change. Handling each step in turn composes, and it is the only way a format that changed three times stays readable.
A field the new format has no place for needs a decision written down: dropped, merged, or kept in an extra slot. A silent drop is a release that deletes data on first open.
Each one is defined here on this site. Open the entry for the full definition and the other words it points at.