Fix #1524: Implement constant folding for +[string]. #2007

tOverney · 2015-11-08T17:04:27Z

/cc @paullepoulpe

scala-jenkins · 2015-11-08T17:04:28Z

Can one of the admins verify this patch?

gzm0 · 2015-11-08T17:25:49Z

What prevents you from folding classOf[T].toString?

gzm0 · 2015-11-08T17:27:23Z

There is already an issue for this. Please prefix the commit message with "Fix #1524: ".

gzm0 · 2015-11-08T17:29:05Z

And as always I forget to write the most important thing first: Thank you!

gzm0 · 2015-11-08T18:54:48Z

Jenkins, ok to test

sjrd · 2015-11-08T19:18:47Z

What prevents you from folding classOf[T].toString?

The result needs to depend on the runtimeClassName function option of the semantics, which is not tracked and difficult to track. Better not to start to depend on that from the optimizer.

sjrd · 2015-11-08T19:53:13Z

test-suite/src/test/scala/org/scalajs/testsuite/compiler/OptimizerTest.scala

@@ -81,6 +81,64 @@ object OptimizerTest extends JasmineTest {

  }

+  describe("String_+ constant folding") {
+    it("must fold two constant string to the correct value") {
+      expect("I am " + "constant").toEqual("I am constant")


Did you check that this is not simply constant-folded by scalac? I think not, but one way to make sure is to extract one of the two sides in an @inline def.

scala-jenkins · 2015-11-08T19:59:05Z

Refer to this link for build results (access rights to CI server needed):

https://scala-webapps.epfl.ch/jenkins/job/scalajs-pr/2395/
https://scala-webapps.epfl.ch/jenkins/job/scalajs-matrix-build/3108/
Test FAILed.

sjrd · 2015-11-08T20:00:01Z

test-suite/src/test/scala/org/scalajs/testsuite/compiler/OptimizerTest.scala

+      def someString(a: Int) = "awesome! " + a + "/10"
+      expect("ScalaJS" + (" is " + someString(10))).toEqual("ScalaJS is awesome! 10/10")
+      expect((someString(10) + " is ") + "ScalaJS").toEqual("awesome! 10/10 is ScalaJS")
+    }


What about "ScalaJS" + (1.4f + someString(10))"?

sjrd · 2015-11-08T20:48:02Z

"and test it" in the commit message should not be there. If it was not tested, it wouldn't get it in the first place ;)

gzm0 · 2015-11-08T21:57:42Z

The result needs to depend on the runtimeClassName function option of the semantics, which is not tracked and difficult to track. Better not to start to depend on that from the optimizer.

What do you mean by tracked? If you are talking about incrementality, changing the semantics requires creating a new optimizer anyways.

sjrd · 2015-11-08T22:13:36Z

Of you're right, I went through that reasoning, now I remember. The real problem is that we also need to know whether the classOf[T] represents a class or an interface. And the optimizer does not have any knowledge query to ask whether a T is a class or an interface.

We could add one, of course, and track it. But it's quite overkill, just to be able to constant fold a classOf[T].toString.

gzm0 · 2015-11-08T22:24:04Z

Ah, nice one. @tOverney please add a comment explaining this. Also, a (doc) comment on literalToString stating that it can't handle class literals would be nice.

scala-jenkins · 2015-11-09T09:18:21Z

Refer to this link for build results (access rights to CI server needed):

https://scala-webapps.epfl.ch/jenkins/job/scalajs-pr/2396/
https://scala-webapps.epfl.ch/jenkins/job/scalajs-matrix-build/3110/
Test FAILed.

sjrd · 2015-11-10T10:05:01Z

tools/shared/src/main/scala/org/scalajs/core/tools/optimizer/OptimizerCore.scala

@@ -2183,6 +2216,43 @@ private[optimizer] abstract class OptimizerCore(
          case _               => default
        }

+      case String_+ =>
+        (lhs, rhs) match {
+          case (l: Literal, r: Literal) => 


OK now I think I see how you can factor that much more. Instead of defining literalToString(Literal): String, define foldToStringForString_+(tree: Tree): Tree which transforms IntLiteral(v) into StringLiteral(v.toString), for example, and returns tree if it cannot fold the conversion of tree to string.

Then, here, you can just do:

val lhs1 = foldToStringForString_+(lhs) val rhs1 = foldToStringForString_+(rhs) (lhs1, rhs1) match { ... }

and now you know that you only have to deal with StringLiterals. Any other stuff that you find is not foldable to a String, and therefore you need not worry about it. Then you will have way less cases here.

(note: from there you can't use default anymore, since at least you need to use lhs1 and rhs1, not lhs and rhs, so defaults should become BinaryOp(String_+, lhs1, rhs1) instead.

tOverney · 2015-11-10T10:05:06Z

@sjrd @gzm0 Thanks for the feedback, Hopefully I've updated everything.
Regarding the real number printing, @paullepoulpe has an implementation ready here to print real number in the JVM to match ECMAScript specifications.

sjrd · 2015-11-19T16:22:14Z

That's all. Good job!

tOverney · 2015-11-19T16:25:41Z

s/ScalaJS/Scala.js

The funny thing is that running that exact regex on the file would break some tests :)
We got the message tho!

scala-jenkins · 2015-11-19T17:08:23Z

Refer to this link for build results (access rights to CI server needed):

https://scala-webapps.epfl.ch/jenkins/job/scalajs-pr/2453/
https://scala-webapps.epfl.ch/jenkins/job/scalajs-matrix-build/3179/
Test FAILed.

scala-jenkins · 2015-11-19T17:09:48Z

Refer to this link for build results (access rights to CI server needed):

https://scala-webapps.epfl.ch/jenkins/job/scalajs-pr/2454/
https://scala-webapps.epfl.ch/jenkins/job/scalajs-matrix-build/3180/
Test FAILed.

sjrd · 2015-11-19T17:14:07Z

Scalastyle says:

[info] scalastyle using config /localhome/jenkins/a/workspace/scalajs-task-worker/scalastyle-config.xml
[error] /localhome/jenkins/a/workspace/scalajs-task-worker/test-suite/js/src/test/scala/org/scalajs/testsuite/compiler/OptimizerTest.scala:144:0: Whitespace at end of line
[error] /localhome/jenkins/a/workspace/scalajs-task-worker/test-suite/js/src/test/scala/org/scalajs/testsuite/compiler/OptimizerTest.scala:163:0: Whitespace at end of line
[error] /localhome/jenkins/a/workspace/scalajs-task-worker/test-suite/js/src/test/scala/org/scalajs/testsuite/compiler/OptimizerTest.scala:291:39: Whitespace at end of line
[error] /localhome/jenkins/a/workspace/scalajs-task-worker/test-suite/js/src/test/scala/org/scalajs/testsuite/compiler/OptimizerTest.scala:294:38: Whitespace at end of line
[error] /localhome/jenkins/a/workspace/scalajs-task-worker/test-suite/js/src/test/scala/org/scalajs/testsuite/compiler/OptimizerTest.scala:86:18: Public method must have explicit type
[error] /localhome/jenkins/a/workspace/scalajs-task-worker/test-suite/js/src/test/scala/org/scalajs/testsuite/compiler/OptimizerTest.scala:91:20: Public method must have explicit type
[error] /localhome/jenkins/a/workspace/scalajs-task-worker/test-suite/js/src/test/scala/org/scalajs/testsuite/compiler/OptimizerTest.scala:102:20: Public method must have explicit type
[error] /localhome/jenkins/a/workspace/scalajs-task-worker/test-suite/js/src/test/scala/org/scalajs/testsuite/compiler/OptimizerTest.scala:108:18: Public method must have explicit type
[error] /localhome/jenkins/a/workspace/scalajs-task-worker/test-suite/js/src/test/scala/org/scalajs/testsuite/compiler/OptimizerTest.scala:109:18: Public method must have explicit type
[error] /localhome/jenkins/a/workspace/scalajs-task-worker/test-suite/js/src/test/scala/org/scalajs/testsuite/compiler/OptimizerTest.scala:110:18: Public method must have explicit type

sjrd · 2015-11-19T17:15:50Z

Otherwise it's good.
Please also git rebase -i master and squash all the commits into a single commit with the commit message

Fix #1524: Implement constant folding for +[string].

scala-jenkins · 2015-11-19T17:49:14Z

Refer to this link for build results (access rights to CI server needed):

https://scala-webapps.epfl.ch/jenkins/job/scalajs-pr/2455/
https://scala-webapps.epfl.ch/jenkins/job/scalajs-matrix-build/3181/
Test FAILed.

scala-jenkins · 2015-11-19T17:49:14Z

Refer to this link for build results (access rights to CI server needed):

https://scala-webapps.epfl.ch/jenkins/job/scalajs-pr/2456/
https://scala-webapps.epfl.ch/jenkins/job/scalajs-matrix-build/3182/
Test FAILed.

sjrd · 2015-11-19T18:35:28Z

LGTM

scala-jenkins · 2015-11-19T18:38:32Z

Refer to this link for build results (access rights to CI server needed):

https://scala-webapps.epfl.ch/jenkins/job/scalajs-pr/2457/
https://scala-webapps.epfl.ch/jenkins/job/scalajs-matrix-build/3183/
Test FAILed.

scala-jenkins · 2015-11-19T18:47:41Z

Refer to this link for build results (access rights to CI server needed):

https://scala-webapps.epfl.ch/jenkins/job/scalajs-pr/2458/
https://scala-webapps.epfl.ch/jenkins/job/scalajs-matrix-build/3184/
Test FAILed.

scala-jenkins · 2015-11-19T19:21:25Z

Refer to this link for build results (access rights to CI server needed):

https://scala-webapps.epfl.ch/jenkins/job/scalajs-pr/2459/
https://scala-webapps.epfl.ch/jenkins/job/scalajs-matrix-build/3185/
Test FAILed.

scala-jenkins · 2015-11-19T19:28:49Z

Refer to this link for build results (access rights to CI server needed):

https://scala-webapps.epfl.ch/jenkins/job/scalajs-pr/2460/
https://scala-webapps.epfl.ch/jenkins/job/scalajs-matrix-build/3186/
Test FAILed.

sjrd · 2015-11-19T22:18:05Z

Guess what! Google Closure Compiler wrongly constant-folds Doubles in string concatenation! It emits the JVM version :-p I'll dig if it's a known bug. (Even Rhino does it correctly.)

In any case, you should therefore add a unless("fullopt-stage"). on the test case that is failing. You can see here an example of the coding style we use for unless/when.

sjrd · 2015-11-19T22:34:02Z

A new entry for the hall of fame: google/closure-compiler#1262

scala-jenkins · 2015-11-20T03:16:05Z

Refer to this link for build results (access rights to CI server needed):

https://scala-webapps.epfl.ch/jenkins/job/scalajs-pr/2461/
https://scala-webapps.epfl.ch/jenkins/job/scalajs-matrix-build/3187/
Test FAILed.

sjrd · 2015-11-20T08:46:00Z

The bootstrap test fails with the long strings, at the time of deserializing the Trees (so even before we reach the optimizer). I suggest you simply remove the test with long strings. We have enough evidence that the JS VMs can cope with the long literals, since all the "normal" tests passed.

sjrd · 2015-11-20T11:17:26Z

LGTM

scala-jenkins · 2015-11-20T13:18:23Z

Refer to this link for build results (access rights to CI server needed):

https://scala-webapps.epfl.ch/jenkins/job/scalajs-pr/2468/
https://scala-webapps.epfl.ch/jenkins/job/scalajs-matrix-build/3195/
Test FAILed.

scala-jenkins · 2015-11-20T16:35:38Z

Refer to this link for build results (access rights to CI server needed):

https://scala-webapps.epfl.ch/jenkins/job/scalajs-pr/2469/
https://scala-webapps.epfl.ch/jenkins/job/scalajs-matrix-build/3196/
Test PASSed.

Fix #1524: Implement constant folding for +[string].

sjrd · 2015-11-20T16:42:42Z

Yay!

sjrd reviewed Nov 8, 2015
View reviewed changes

tOverney changed the title ~~Implement constant folding for String_+ and test it.~~ Fix #1524: Implement constant folding for +[string] Nov 9, 2015

tOverney changed the title ~~Fix #1524: Implement constant folding for +[string]~~ Fix #1524: Implement constant folding for +[string]. Nov 9, 2015

tOverney force-pushed the string-plus-cst-fold branch from ee81a96 to 17b1f3f Compare November 9, 2015 07:41

sjrd reviewed Nov 10, 2015
View reviewed changes

tOverney force-pushed the string-plus-cst-fold branch from 6264a74 to 4360190 Compare November 19, 2015 17:47

tOverney force-pushed the string-plus-cst-fold branch from f63010f to 7367804 Compare November 19, 2015 18:07

paullepoulpe force-pushed the string-plus-cst-fold branch from 7367804 to bfbbf7f Compare November 19, 2015 23:34

tOverney force-pushed the string-plus-cst-fold branch from bfbbf7f to b7baee3 Compare November 20, 2015 10:41

Fix scala-js#1524: Implement constant folding for +[string].

61fa19c

paullepoulpe force-pushed the string-plus-cst-fold branch from b7baee3 to 61fa19c Compare November 20, 2015 12:55

sjrd added a commit that referenced this pull request Nov 20, 2015

Merge pull request #2007 from SebsLittleHelpers/string-plus-cst-fold

4cde878

Fix #1524: Implement constant folding for +[string].

sjrd merged commit 4cde878 into scala-js:master Nov 20, 2015

paullepoulpe deleted the string-plus-cst-fold branch December 28, 2015 14:00

Fix #1524: Implement constant folding for +[string]. #2007

Fix #1524: Implement constant folding for +[string]. #2007

Uh oh!

Conversation

tOverney commented Nov 8, 2015

Uh oh!

scala-jenkins commented Nov 8, 2015

Uh oh!

gzm0 commented Nov 8, 2015

Uh oh!

gzm0 commented Nov 8, 2015

Uh oh!

gzm0 commented Nov 8, 2015

Uh oh!

gzm0 commented Nov 8, 2015

Uh oh!

sjrd commented Nov 8, 2015

Uh oh!

sjrd Nov 8, 2015

Choose a reason for hiding this comment

Uh oh!

scala-jenkins commented Nov 8, 2015

Uh oh!

sjrd Nov 8, 2015

Choose a reason for hiding this comment

Uh oh!

sjrd commented Nov 8, 2015

Uh oh!

gzm0 commented Nov 8, 2015

Uh oh!

sjrd commented Nov 8, 2015

Uh oh!

gzm0 commented Nov 8, 2015

Uh oh!

scala-jenkins commented Nov 9, 2015

Uh oh!

sjrd Nov 10, 2015

Choose a reason for hiding this comment

Uh oh!

tOverney commented Nov 10, 2015

Uh oh!

sjrd commented Nov 19, 2015

Uh oh!

tOverney commented Nov 19, 2015

Uh oh!

scala-jenkins commented Nov 19, 2015

Uh oh!

scala-jenkins commented Nov 19, 2015

Uh oh!

sjrd commented Nov 19, 2015

Uh oh!

sjrd commented Nov 19, 2015

Uh oh!

scala-jenkins commented Nov 19, 2015

Uh oh!

scala-jenkins commented Nov 19, 2015

Uh oh!

sjrd commented Nov 19, 2015

Uh oh!

scala-jenkins commented Nov 19, 2015

Uh oh!

scala-jenkins commented Nov 19, 2015

Uh oh!

scala-jenkins commented Nov 19, 2015

Uh oh!

scala-jenkins commented Nov 19, 2015

Uh oh!

sjrd commented Nov 19, 2015

Uh oh!

sjrd commented Nov 19, 2015

Uh oh!

scala-jenkins commented Nov 20, 2015

Uh oh!

sjrd commented Nov 20, 2015

Uh oh!

sjrd commented Nov 20, 2015

Uh oh!

scala-jenkins commented Nov 20, 2015

Uh oh!

scala-jenkins commented Nov 20, 2015