Add versionedjob contrib package with basic versioning framework

[brandur]

Here, add a contrib package called versionedjob that brings in a very
basic job versioning framework. The framework is very lightweight and
users could easily have written it themselves, but it might be a little
non-obvious how to leverage hooks to bring it in.

Contrib should be a really nice home for this feature because it's a
little too trivial to bother with in core, but it's nice for people to
have a semi-official reference that's well-maintained.

[diegobernardes]

🚀

[brandur]

Nice.

@bgentry Want to take a look at this too? Might make a good blog post at least.

[bgentry]

Seems like one good way to do this! Another is to implement a custom UnmarshalJSON on your arg type.

[bgentry]

I assume you meant encoding/json?

Suggested change

The `VersionTransformer` implementation handles version upgrades one by one. Jobs which are multiple versions old can still be upgraded because multiple version changes can be applied in one go. This implementation uses `gjson`/`sjson` so that each change need only know a minimum about the data object in question and that unknown fields are retained. Other approaches are possible though, including using only Go's built-in `gjson` package.

The `VersionTransformer` implementation handles version upgrades one by one. Jobs which are multiple versions old can still be upgraded because multiple version changes can be applied in one go. This implementation uses `gjson`/`sjson` so that each change need only know a minimum about the data object in question and that unknown fields are retained. Other approaches are possible though, including using only Go's built-in `encoding/json` package.

[brandur]

thx.

[brandur]

Seems like one good way to do this! Another is to implement a custom UnmarshalJSON on your arg type.

I think one reason you might want to do this separately is that eventually all non-trivial version transformations tend to involved fetching data from the DB (this is what happened at Stripe anyway), so you probably want to have a struct you can add a DB pool to and access to ctx.

This is definitely something to make clear in an eventual blog post though.