Tests do not compile with Scala 2.11

[propensive]

Looks like @joroKr21's stack reworking introduced something, which appears to affect only Scala 2.11, and currently stops four tests from compiling... I've seen error messages suggestion something relating to lazy vals.

[joroKr21]

I don't know about lazy vals, but one thing I noticed for sure in #73 is that DirectlyReentrantException always triggers on recursive types, but then we recover. That happens because we first try the regular implicit resolution and we always find our own macro and then we fall back to direct recursive derivation. That may or may not have happened before but apart from causing #73 it must be horribly inefficient so there is even room for speedup! I'm not sure how to fix it yet though.

[propensive]

The DirectlyReentrantException hack was very much part of the original design, and it effectively uses implicit search to check whether there is a non-Magolia implicit in scope for that particular search which should take precedence. I don't know of a better way to do this... It's only an exception in that we want a nested macro invocation to fail so that we can search for other suitable implicits that would otherwise be shadowed...

I'm just investigating some more, though... :)

[propensive]

Here are the errors, by the way;

[info] Compiling 1 Scala source to /home/jpretty/dev/propensive.com/magnolia/tests/target/scala-2.11/classes ...
[error] /home/jpretty/dev/propensive.com/magnolia/tests/src/main/scala/tests.scala:130:17: could not find implicit value for parameter e: magnolia.examples.Show[String,magnolia.tests.FruitBasket]
[error]       implicitly[Show[String, FruitBasket]].show(FruitBasket(Fruit("apple"), Fruit("banana")))
[error]                 ^
[error] /home/jpretty/dev/propensive.com/magnolia/tests/src/main/scala/tests.scala:237:15: not found: value companyTypeclass$macro$255$lzy
[error]       Show.gen[Politician[String]].show(Corrupt("wall", Seq(Company("Alice Inc"))))
[error]               ^
[error] /home/jpretty/dev/propensive.com/magnolia/tests/src/main/scala/tests.scala:333:15: not found: value companyTypeclass$macro$352$lzy
[error]       Show.gen[Portfolio].show(Portfolio(Company("Alice Inc"), Company("Bob & Co")))
[error]               ^
[error] /home/jpretty/dev/propensive.com/magnolia/tests/src/main/scala/tests.scala:381:15: not found: value pathTypeclass$macro$425$lzy
[error]       Show.gen[Path[String]].show(OffRoad(Some(Crossroad(Destination("A"), Destination("B")))))
[error]               ^
[error] four errors found

[joroKr21]

I completely forgot about scala/bug#5466 (note the $lzy suffix)

[joroKr21]

Also, it seems that currently crossScalaVersions in tests is only 2.12

[propensive]

Ah, does that mean we can get it working again on Scala 2.11 with some targetted untypechecking?

[joroKr21]

Rather the opposite - the problems is that we use c.untypecheck on code that contains lazy vals.

[propensive]

Oh, I see. I can see some logic (I think) for the cases where it's happening...

I imagine that in an absolute worst-case scenario we could replace them with some dummy code, and rewrite that code as actual lazy vals when the macro finally produces output. But maybe you know a more straightforward workaround?

[joroKr21]

I don't know of any straightforward workaround. I'm not even sure what you're suggesting will work but you're welcome to try. In general the splicing of typed and untyped trees is an unsolved problem in scala.reflect. You can try to avoid calling untypecheck (it's only one place). Maybe typingTransform can be used? No easy solution I'm afraid.

[adamw]

Fixed by removing 2.11 support :)