Redesign Tuples with HList-like structure #2199

Changes are as follows: - New types for tuples in dotty library: Tuple = Unit | TupleCons Head Tail <: Tuple - Desugaring uses this structure instead of the scala.TupleN for types and expressions - New TupleRewrites phase does rewrites the HList structure into scala.TupleN like case classes (for small arities) or a case class wrapping an Array (for large arities)

To stay binary compatibility with scala 2.11 binaries, we hijack the unpickling of types to fake an HList structure. Later on in erasure, HList types are eraised back to scala.TupleN (for N <= 22). In addition because the actual scala.TupleN classes are neither `<: Tuple` now `<: TupleCons`, these types needs to be systematically araised to Object. In addition to all these carefully compensating hacks, this also imposes a new contain: the dotty-library, which defines several `Tuple/TupleCons` methods, can should now *always* be compiled by dotty itself. Indeed, when compiled with scalac, these types will stay as they are instead of being eraised to Object.

I couldn't get rid of the `dotty-library` project completely, it's still used the first time sbt-briged is compiled with dotty. The dependencies are as follows: - (no dep) compile `dotty-compile` with scalac [0] - (no dep) compile `dotty-library` with scalac [1] - ([0], [1]) → compile `dotty-sbt-bridge` with dotty [2] - ([0], [2]) → compile `dotty-library-bootstraped` with dotty + sbt After that, [1] should never be used again.

@unchecked

- dotc/scala-collections.blacklist: Blacklist Unit and TupleN from scala-collections test - tests/neg/i1653.scala: Having an additional superclass on Unit seams to affect the error here (not sure why). - tests/pos/t0301.scala: Since `_2` is now added thought implicit conversion, the pattern used in this case is no longer stable. - tests/pos/tcpoly_bounds1.scala: This test uses the fact that `(A, B) <: Product2[A, B]`, which is no longer the case. - tests/repl/patdef.check: Having the `_i` via implicit conversions remove the @unchecked here - tests/run/tuple-match.scala: Replace Tuple1 in pattern matching with explicit TupleCons pattern. TupleN can still be used in apply position (defined as defs in DottyPredef), but not in unapply position. - tests/run/unapply.scala: Also uses the fact that `(A, B) <: Product2[A, B]`. - tests/run/withIndex.scala: Might be a limitation in typer?: 14 | case _: Array[Tuple2[_, _]] => true | ^^^^^^^^^^^^ |unreducible application of higher-kinded type [A, B] => dotty.TupleCons[A, dotty.TupleCons[B, Unit]] to wildcard arguments

The LogPendingSubTypesThreshold change is requires for tests on size ~25 tuples that would otherwise trigger -Yno-deep-subtypes

Because all tests depend on dotty-library-bootstrapped publishLocal is always needed to run tests.

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

Redesign Tuples with HList-like structure #2199

Redesign Tuples with HList-like structure #2199

Commits on May 1, 2017