Catamorphism in Scala

Debasish (দেবাশিস্) Ghosh 🇮🇳 @debasishg

In Scala, using case classes and subtyping versus catamorphism - which one do u recommend as best practice and why ? #idiomaticscala

Tony Morris @dibblego

@debasishg Private case class, smart constructor, no subtyping, no variance. Better abstraction, better type inference, equal performance.

Daniel Spiewak @djspiewak

@debasishg as long as you don't need specific subtypes, private case class and override the companion apply to return parent type.

Daniel Spiewak @djspiewak

@debasishg Don't use catamorphic encoding. The performance hit is pretty heavy.

mergeconflict @mergeconflict

@djspiewak @debasishg what do you mean by "catamorphic encoding"? i can has gist plz?

Debasish (দেবাশিস্) Ghosh 🇮🇳 @debasishg

@mergeconflict @djspiewak Have a look at the Iteratee impl in scalaz done w/ a fold instead of case classes as shown in http://t.co/ZoJsDaHO

Josh Suereth @jsuereth

@djspiewak @debasishg Hopefully not in the long run, but right now, yes.

Josh Suereth @jsuereth

@mergeconflict @djspiewak @debasishg fold vs. pattern matching.

mergeconflict @mergeconflict

@debasishg I don't follow. That article looks like it's representing choice as case classes, and the fold is "just a fold." /cc @djspiewak

Debasish (দেবাশিস্) Ghosh 🇮🇳 @debasishg

@mergeconflict @djspiewak The article uses case classes & subtyping for Done, Cont etc. but the impl uses fold .. https://t.co/HmC2fsnx

Debasish (দেবাশিস্) Ghosh 🇮🇳 @debasishg

@mergeconflict @djspiewak IOW the impl doesn't use subtyping on case classes for the ADT encoding

mergeconflict @mergeconflict

@jsuereth I understand that. I'm asking e.g. what would Either look like without case classes? /cc @djspiewak @debasishg

Daniel Spiewak @djspiewak

@mergeconflict @jsuereth @debasishg http://t.co/jNXlvODT

Uncarved Bitmap @crstry

@djspiewak @mergeconflict @jsuereth @debasishg So using a virtual fold method instead of pattern matching, then?

Daniel Spiewak @djspiewak

@crstry @mergeconflict @jsuereth @debasishg Yes. That is the catamorphic encoding of an ADT.

Uncarved Bitmap @crstry

@djspiewak @mergeconflict @jsuereth @debasishg Aha. Thanks!

mergeconflict @mergeconflict

@djspiewak Ohhh, anonymous classes. I was imagining something terrible: a single concrete class with nullable left and right fields :P

Daniel Spiewak @djspiewak

@mergeconflict You can use concrete classes as well. I was using anonymous classes for the sake of brevity.

Debasish (দেবাশিস্) Ghosh 🇮🇳 @debasishg

@djspiewak @crstry @mergeconflict @jsuereth It's inefficiency is due to the overhead of closures on the JVM ? Subtyping will be faster ?

Daniel Spiewak @djspiewak

@debasishg @crstry @mergeconflict @jsuereth That is part of the inefficiency, yes. Remember, you create a closer for even unused cases.

Daniel Spiewak @djspiewak

@debasishg @crstry @mergeconflict @jsuereth .@nuttycom came up with a clever hack to avoid *some* of this overhead, but not all.

Daniel Spiewak @djspiewak

@debasishg @crstry @mergeconflict @jsuereth It's not so much that subtyping is faster. Rather, pattern matching has less overhead.

Josh Suereth @jsuereth

@debasishg @djspiewak @crstry @mergeconflict to be fair, it's an interaction of Scala's encoding of closures and JIT.

Daniel Spiewak @djspiewak

@jsuereth @debasishg @crstry @mergeconflict More than that. The heap pressure from catamorphic encodings in hot paths is obscene.

Debasish (দেবাশিস্) Ghosh 🇮🇳 @debasishg

@djspiewak @crstry @mergeconflict @jsuereth @nuttycom share the trick ? smart constructor as @dibblego mentioned a la none, some in scalaz ?