Compiler hangs when using LabelledGeneric to convert classes

[vitorsvieira]

Hi all,

Generic and LabelledGeneric are awesome tools to "convert" classes, but unfortunately this is not true when the classes have approx. more than 10 fields. The compiler loses the track of the implicits and it hangs.

One can easily spot such behaviour by adding more fields to the example given in the Shapeless Guide and in the exercises solutions.

I was pointed to scala/scala#5649 when asking for a help on gitter and I understand that it is a work in progress, aiming to be delivered in scala 2.12.3.

For now, how to use the pattern shown in the book and in recent talks with "real world" classes?

https://github.com/underscoreio/shapeless-guide-code/tree/solutions/migrations/src/main/scala

Case Class Migration at Scala World, from Dave Gurnell

[milessabin]

I don't think this should be an issue until you get to ~30 fields ... do you have an example of a 10 field case class where the compilation time is unacceptable?

[vitorsvieira]

Hi Miles,

Here is an example:

https://github.com/vitorsvieira/shapeless-labelledgeneric-big-class/blob/master/src/main/scala/com/vitorsvieira/shapeless/labelledgeneric/bigclass/Main.scala

[milessabin]

Ahh ... understood. I think it's highly unlikely that the compile time cost is attributable to LabelledGeneric ... I think it's most likely the Align which is the culprit.

[Atry]

I guess a macro materializer for Align may help.

[Atry]

A macro version Align will improve not only the compiler time cost but also run time cost

[milessabin]

@Atry do you have time to investigate whether -Yinduction-heuristics in Typelevel Scala help with this one?

[fspillner]

I have issues with compilation speed with Align too, but only if the implicits could not be resolved (it took 15 minutes until I get the error). If I put the right type, then it compiles fast.

Additional info: My ADT contains 12 fields and is aligned with another ADT with 7 fields (with Intersection before)

[joroKr21]

Should be fixed by #682 - feel free to reopen otherwise.