Can a standard compiler plugin add a main entry point?

[soronpo]

I'm fiddling with a compiler plugin that adds a main entry point for companion objects of classes annotated with a certain annotation.
Since I'm running into trouble, it's not clear if this is even possible to do with a standard plugin, so I'll be happy to even just get a clarification it's possible. Specifically:

1. If the companion object exists, adding a def main(args: Array[String]): Unit method to its body should be recognized.
2. It's possible to create a companion object (using ModuleDef) with a def main(args: Array[String]): Unit as its body for it to be recognized during runtime.

[soronpo]

Ok, point 1 is working.
Still need to check for point 2.

[soronpo]

I can't get 2 to work. Is creating a companion object allowed in a standard compiler plugin?

[soronpo]

Thank you @sjrd for the pointer! I got it to almost work the way I want it.
There are two minor issues that maybe you can answer again if it's just possible to fix:
A. When I add the companion, the added companion module class now collides with the existing class that I create a companion for.

[warn] 6 |class Example() extends DFDesign [warn] | ^ [warn] | Example and Example produce classes that overwrite one another 

This is of course different behavior than when I wrote the companion manually, although both create another class in the same name. So I'm guessing I'm missing some wiring that explicitly defines the newly created class as a companion. Do you think it's possible to do in a post-typer plugin?
B. I now noticed that when I add the main entry point I can run it with sbt runMain, but sbt run does not detect it. Is this fixable? I'm not familiar between the two mechanisms.

[soronpo]

B. I now noticed that when I add the main entry point I can run it with sbt runMain, but sbt run does not detect it. Is this fixable? I'm not familiar between the two mechanisms.

I managed to fix this by extending from App. It should be good enough.

[soronpo]

OK, extending App is not a good solution if I want argument support. Any clue why sbt run does not identify this main but sbt runMain does?

[sjrd]

sbt run takes its information from the Zinc analysis. So it's possible that the transformation comes too late for it to be registered in the sbt API generation phase.

[soronpo]

It seems if I publish locally and run sbt or scala-cli as usual it works fine. Don't know why, but I'll take the win.
Thank you very much!