Fix #3866: Add support for typetags #3867

odersky · 2018-01-18T17:14:34Z

Based on feedback with the early implementation, we found out that we need type lifting
in the framework. The first commit explains what needs to be done in the meta programming doc.

The other commits are a first implementation. I am still not quite happy with it because we are lacking a way to do quoting and splicing directly on types. So for now we generate a type tree, but this cannot express all possible positions of typetags. For instance, we could not splice
a type tag in the position of a type parameter of a refined method since refined methods are
represented as types, not as type trees.

So maybe we need to take a step back and allow that. But that would upset quite a few core structures of the compiler.

odersky · 2018-01-18T18:14:56Z

test performance please

dottybot · 2018-01-18T18:15:52Z

performance test scheduled: 4 job(s) in queue, 1 running.

dottybot · 2018-01-18T20:26:43Z

Performance test finished successfully:

Visit http://dotty-bench.epfl.ch/3867/ to see the changes.

Benchmarks is based on merging with master (79281da)

liufengyun · 2018-01-19T08:13:58Z

docs/sidebar.yml

-            - title: Symmetric Meta Programming
-              url: docs/reference/symmetric-meta-programming.html
+            - title: Meta Programming
+              url: docs/reference/principled-meta-programming.html


The file name needs to be changed to principled-meta-programming.md as well.

odersky · 2018-01-19T10:09:09Z

I think we found our performance regression culprit causing #3721: isQuote and isSplice did a member lookup on every call because they were implemented in a sub-optimal way in Definitions.
Fixed by e6664ed

nicolasstucki · 2018-01-19T09:52:43Z

docs/docs/reference/symmetric-meta-programming.md

-    }
+    def reflect[T, U](f: Expr[T] => Expr[U])(implicit t: Type[T]): Expr[T => U] =
+      ’{ (x: ~t) => ~f(’(x))
+


Missing } at the end

nicolasstucki · 2018-01-19T10:06:08Z

library/src/scala/quoted/Type.scala

@@ -9,6 +9,8 @@ abstract class Type[T] extends Quoted {
 /** Some basic type tags, currently incomplete */
 object Type {

+  implicit def apply[T]: Type[T] = ???


Are calls to this methods going to be replaced by the tags below when possible and replaced '[T] if the type is statically known?

nicolasstucki · 2018-01-22T09:46:32Z

docs/docs/reference/principled-meta-programming.md

      def unary_~: T   // splice operation
    }
-    class Type[T] {
+    sealed abstract class Type[T] {


@odersky I created #3883 to make these classes sealed, no need to implement it on this PR.

nicolasstucki

Otherwise LGTM

nicolasstucki · 2018-01-23T13:08:18Z

docs/docs/reference/principled-meta-programming.md

+reference to a type `T` to a type-splice, by rewriting `T` to `~implicitly[Type[T]]`.
+For instance, the user-level definition of `reflect`
+
+    def reflect[T: Type, U](f: Expr[T] => Expr[U]) Expr[T => U] =


Missing : before result type.

nicolasstucki · 2018-01-23T13:08:52Z

docs/docs/reference/principled-meta-programming.md

+
+would be rewritten to
+
+    def reflect[T: Type, U](f: Expr[T] => Expr[U]) Expr[T => U] =


Missing : before result type.

nicolasstucki · 2018-01-23T13:38:44Z

compiler/src/dotty/tools/dotc/typer/Implicits.scala

+        }
+
+        val tag = toTree(tpe)
+        if (ok) ref(defn.typeQuoteMethod).appliedToTypeTrees(tag :: Nil)


In the future, we should optimize primitive types to return quoted.Type.XYZTag. But we should do it in general for any '[XYZ] that has a quoted.Type.XYZTag. Probably as part of reification.

nicolasstucki · 2018-01-23T13:46:37Z

compiler/src/dotty/tools/dotc/typer/Implicits.scala

+            RefinedTypeTree(toTree(tpe), refinements, refineCls)
+        }
+
+        def toTree(tpe: Type): Tree = tpe.stripTypeVar match {


As discussed, we should try using the same trick to no need all these trees.

nicolasstucki

LGTM, just missing a couple of details in the documentation.

nicolasstucki · 2018-01-27T11:27:05Z

docs/docs/reference/principled-meta-programming.md

+reference to a type `T` to a type-splice, by rewriting `T` to `~implicitly[Type[T]]`.
+For instance, the user-level definition of `reflect`:
+
+    def reflect[T: Type, U](f: Expr[T] => Expr[U]) Expr[T => U] =


Still missing a :

nicolasstucki · 2018-01-27T11:27:23Z

docs/docs/reference/principled-meta-programming.md

+
+would be rewritten to
+
+    def reflect[T: Type, U](f: Expr[T] => Expr[U]) Expr[T => U] =


Based on feedback with the early implementation, we found out that we need type lifting in the framework. This commit explains what needs to be done in the meta programming doc.

It's not so easy to see what is symmetric about it. "Principled" is bolder, but why not, since we do introduce a new principle?

This is intended as a stub. Calls to this method will be replaced by implicit type tag searches. Also, rework definition of quote and splice methods. This could also have contributed in the slowdown we were seeing in scala#3721.

We need to come back to the earlier scheme that a lifted type needs to refer to an implicit type tag.

Not sure this is the right way. It feels a bit roundabout to take the detour from types to type trees. But I am not sure we can correctly quote types without trees.

Generate a type with embedded quotes instead of a tree.

odersky requested a review from nicolasstucki January 18, 2018 17:14

liufengyun reviewed Jan 19, 2018

View reviewed changes

nicolasstucki reviewed Jan 19, 2018

View reviewed changes

nicolasstucki assigned odersky Jan 20, 2018

odersky force-pushed the fix-typetags branch from e036733 to 17ad246 Compare January 21, 2018 18:19

nicolasstucki reviewed Jan 22, 2018

View reviewed changes

odersky changed the title ~~[WIP] Add type lifting~~ Fix #3866: Add support for typetags Jan 22, 2018

odersky assigned nicolasstucki and unassigned odersky Jan 22, 2018

nicolasstucki reviewed Jan 23, 2018

View reviewed changes

nicolasstucki assigned odersky and unassigned nicolasstucki Jan 23, 2018

nicolasstucki self-requested a review January 23, 2018 13:49

nicolasstucki approved these changes Jan 24, 2018

View reviewed changes

nicolasstucki reviewed Jan 27, 2018

View reviewed changes

odersky added 11 commits January 27, 2018 13:02

Explain type lifting

adf40fd

Based on feedback with the early implementation, we found out that we need type lifting in the framework. This commit explains what needs to be done in the meta programming doc.

Revert [Symmetric -> Principled] meta programming

6a60743

It's not so easy to see what is symmetric about it. "Principled" is bolder, but why not, since we do introduce a new principle?

Add quoted.Type.apply

5125c8e

This is intended as a stub. Calls to this method will be replaced by implicit type tag searches. Also, rework definition of quote and splice methods. This could also have contributed in the slowdown we were seeing in scala#3721.

Lift only implicitly given Types

de2d33c

We need to come back to the earlier scheme that a lifted type needs to refer to an implicit type tag.

Main sharing of TypeBoundsTree bounds through (un)pickling

2f71763

Fix test

17e579f

Add type tag support

3653933

Not sure this is the right way. It feels a bit roundabout to take the detour from types to type trees. But I am not sure we can correctly quote types without trees.

Rename doc file

f7bec70

Avoid SO in typetag search

e8df9a0

Allow top-level splices for types

02714df

Simplify TypeTag generation

80f2c76

Generate a type with embedded quotes instead of a tree.

Fix docs

58677ed

odersky force-pushed the fix-typetags branch from 1d6554a to 58677ed Compare January 27, 2018 12:02

odersky merged commit 31a81fb into scala:master Jan 27, 2018

allanrenucci deleted the fix-typetags branch January 27, 2018 15:41


		would be rewritten to

		def reflect[T: Type, U](f: Expr[T] => Expr[U]) Expr[T => U] =

Fix #3866: Add support for typetags #3867

Fix #3866: Add support for typetags #3867

Uh oh!

Conversation

odersky commented Jan 18, 2018 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

odersky commented Jan 18, 2018

Uh oh!

dottybot commented Jan 18, 2018

Uh oh!

dottybot commented Jan 18, 2018

Uh oh!

Choose a reason for hiding this comment

Uh oh!

odersky commented Jan 19, 2018

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

nicolasstucki left a comment

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

nicolasstucki left a comment

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Uh oh!

odersky commented Jan 18, 2018 •

edited

Loading