Scala 3 support

[hughsimpson]

Builds on #676 and #668 to fix compilation for scala 3.

Couple of issues (although I'm not sure if I personally consider them blockers..):

• geom support: This makes makeBox/makeBox3d/makeLine slightly less general, although I've never used this and don't know if it's a real issue
• bumps play-json major version reverted. Not sure why I thought I needed to do that
• does not implement Composite support for scala 3 (I had a crack at implementing that with scala 3 inline, but ran into a wall. Probably needs full-fat macros. Personally I don't use this functionality of slick-pg at all, so I believe it's still worth merging without) Now implements composite (Struct) support for scala 3. It's a different mechanism -- requires some implicits, uses scala 3 inline -- but passes the test suite

[nafg]

Is my understanding correct that scala 2 functionality remains the same and this successfully adds scala 3 support for almost everything?

[hughsimpson]

Pretty much, the differences are clearer if you compare to #676 (which I opened to separate the bulk of the noise from the 'real' changes) but there are some differences:

• I had to remove the DateTimeImplicits alias traits, so that'd probably break some code, but is a very quick fix.
• There are some changes the type signature of 3 methods in the geom package, and I'm not sure whether that could impact a real-world use-case
• Some array support has been moved from using type tags to using izumi. I believe this is fine and causes no conflict with the unmigrated Struct support however.
• Stuff that extends Struct isn't magic any more. I didn't even know slick-pg did this. It's so cool. Anyway I think it might be encodable with inline but I haven't done it yet. Have an implementation for this now that seems to work well

It would be possible to denormalise the code further, as it were, and retain completely identical scala 2 support,. however I felt like that'd be making a rod for our own back and it was probably better to do a more sweeping change

[hughsimpson]

Probably also worth noting that some cool fun syntax doesn't work in scala 3, so there are definitely regressions on that front too, but for the most part they're regexable.

Edit: this only applies to scala 3 but it could make upgrading some stuff more painful

[nafg]

@tminglei can you please review, merge and get this released?

[hughsimpson]

What I should do is remove the Struct entirely from scala 3 to push the breaking change up to the type level

[hughsimpson]

There we go, removed a mine hopefully

Edit: actually I'm now certain that the Struct functionality can indeed be implemented with inline and hope to push that up in the next couple of hours

[hughsimpson]

So this is different to scala 2 usage -- rather than 'register'ing a typetag, we need to have an appropriate implicit 'registered converter' in scope. This all boils down to the composite type being constructed at compile time rather than runtime, and I think it's actually an improvement. Sadly it does mean some inconsistency between scala 2 and scala 3 APIs, because we can't use typetags in scala 3 and can't use inline in scala 2, but hey ho them's the breaks

[hughsimpson]

@tminglei are you able to take a look at this please? Happy to make changes, but would love to start the process of migrating to scala 3 and this is the last significant dependency I have that's not yet crosspublished

[hughsimpson]

Don't know why we needed a second geometry subtype here... Anyway scala 3 seemed to struggle with the type inference, and rather than specifying everything I decided to simplify the signature

[tminglei]

@hughsimpson can't remember it exactly.
I think your changes should be ok after double checked the official docs.

[tminglei]

@hughsimpson merged. Thanks very much! 👍

[nafg]

@tminglei thanks! BTW I see you published to maven central, but I didn't realize it at first because there isn't an M5 tag or release on GitHub.

[tminglei]

Published.
@nafg @hughsimpson sorry for forgot to make a release on Github.