2.8 "no-symbol does not have owner" in lambdalift phase

[scabug]

See also #710 and #805. I've attached the full console output. It dies in the lambdalift phase

How can I find out what the compiler was doing when this error arose? Its a bit hard to produce a minimal test case without this.

A decidedly un-minimal test case is here: http://scalaz.googlecode.com/svn/branches/scala-2.8, SVN revision 825.

I'm on version 2.8.0.r18583-b20090827020153

[scabug]

Imported From: https://issues.scala-lang.org/browse/SI-2316?orig=1
Reporter: @retronym
Attachments:

• 2316.patch (created on Sep 13, 2009 10:56:21 AM UTC, 4093 bytes)
• implicit-cache-error.scala (created on Sep 12, 2009 6:58:41 PM UTC, 976 bytes)
• implicit-cache-error2.scala (created on Sep 13, 2009 8:08:51 AM UTC, 1026 bytes)
• no-owner.patch (created on Sep 22, 2009 8:59:35 AM UTC, 241140 bytes)
• no-owner-stack-analysis.log (created on Sep 5, 2009 12:53:15 PM UTC, 13173 bytes)
• no-symbol.txt (created on Sep 1, 2009 8:01:20 PM UTC, 68738 bytes)

[scabug]

@retronym said:
console output of compilation

[scabug]

@paulp said:
It sounds like #2201, which I have a patch for but do not want to commit without it being approved (or more likely, improved) by martin or adriaan. That was the bug that needed to be fixed for scalacheck to compile and it's very believable that scalaz and scalacheck would induce the same bugs.

If you want you can build the repo with the patch and see if it works - or hopefully it'll be in trunk soon. http://github.com/paulp/scala/tree/scalacheck

[scabug]

@retronym said:
I've just tried the patched compiler. It gets past the first instance of this error -- encountered compiling the scalacheck sources, which are svn:external-ed into the Scalaz build -- but there is another case in triggered by the Scalaz sources proper.

I've got it pinned down in the debugger, so I'll narrow it down and post a test case soon.

-jason

[scabug]

@retronym said:
Analysis of the failure during lambda lift phase

[scabug]

@retronym said:
I took a look at this in the debugger, and attached an analysis. Still can't pare it down to a small example, unfortunately :(

      Transforming unit Pointed.scala. Inspection of the AST here shows that 's' in PromisePure(s) already has the wrong symbol. I expect it to
      be the parameter 's' from PromisePointed, instead it is bound to the parameter 's' from another file.
      
        // Pointed.scala
        trait Pointed[P[_]] extends Functor[P] with Pure[P]

        object Pointed {
           import concurrent._
           implicit def PromisePointed(implicit s: Strategy[Unit]) = pointed[Promise]
   // = Pointed.this.pointed(
   //         scalaz.this.Functor.PromiseFunctor(),
   //   scalaz.this.Pure.PromisePure(s).$$asInstanceOf[scalaz.Pure]())
   //                                |__ 's' bound to out-of-scope implicit
   //                                     param of Applicative.PromiseApplicative()!
        }
      
        //Applicative.scala:
        trait Applicative[Z[_]] extends Pointed[Z] with Apply[Z] {
          override def fmap[A, B](fa: Z[A], f: A => B): Z[B] = apply(pure(f), fa)
        }

        object Applicative {
           implicit def PromiseApplicative(implicit s: Strategy[Unit]) = applicative[Promise]           
        }               

[scabug]

@cunei said:
Adriaan, could you please review Paul's patch? Thanks!

[scabug]

@adriaanm said:
Patch committed as fix for #2201 in r18692

[scabug]

@retronym said:
small test case

[scabug]

@retronym said:

usr$$ ~/usr/scala-2.8.0.r18697/bin/scala ~/Desktop/implicit-cache-error.scala 
java.lang.Error: no-symbol does not have owner
	at scala.tools.nsc.symtab.Symbols$$NoSymbol$$.owner(Symbols.scala:1869)
        ...

Oddly, scalac doesn't produce the error, but you can see the reference to an out-of-scope symbol from test.A.B2 with -uniqid -print:

def thisSI-11202(): object test$$A$$B1SI-11193 = {
      test$$A$$B1SI-11193.super.thisSI-3699();
      test$$A$$B1SI-11193.this.t2_b1SI-11204 = {
        new test$$A$$B1$$$$anon$$2SI-11206()
      };
      test$$ASI-7966.this.requireT1SI-11191(test$$T1SI-7962.T1FromT2SI-11170(test$$A$$B1SI-11193.this.t2_b1SI-11203().$$asInstanceOfSI-6888[test$$T2SI-7964]()).$$asInstanceOfSI-6888[test$$T1SI-7961]());
      ()
    }
    ...
    
    def thisSI-11564(): object test$$A$$B2SI-11195 = {
      test$$A$$B2SI-11195.super.thisSI-3699();
      test$$A$$B2SI-11195.this.t2_b2SI-11566 = {
        new test$$A$$B2$$$$anon$$3SI-11568()
      };
      test$$ASI-7966.this.requireT1SI-11191(test$$T1SI-7962.T1FromT2SI-11170(test$$A$$B1SI-11193.this.t2_b1SI-11203().$$asInstanceOfSI-6888[test$$T2SI-7964]()).$$asInstanceOfSI-6888[test$$T1SI-7961]());
      ()
    }

Implicits are being cached too aggressively. The patch recently applied for #2201 should be reviewed in light of this test case.

Implicits.cacheResult returns T1.T1FromT2(t2_b1) here, which is bogus. Even though T1.T1FromT2 was found outside of the scope of A.B1, this implicit expression should not be cached, as it includes the bound variable t2_b1 from this scope.

SearchResult could be augmented with an attribute cacheable. In Implicits.bestImplicits(), if the first search in the local context succeeds, return SearchResult(tree, subst, cacheable = false). This would need to be propagated for SearchResults that themselves depend on implicits.

Implicits.cacheResult would then ensure that only cacheable SearchResult-s are cached.

[scabug]

@retronym said:
Another example of seemingly innocent implicit caching going wrong.

[scabug]

@retronym said:
Added another test case that invalidates my suggested algorithm for determining 'cacheability' of implicits. This is something of a can of worms. May I suggest -Xdisable-implicit-caching in the meantime?

[scabug]

@retronym said:
Patch to only cache non-local implicits if safe to do so.

[scabug]

@retronym said:
I've implemented a simpler solution, whereby only the results of implicitsOfExpectedType() are cached.

http://github.com/retronym/scala/commits/implicit-caching

[scabug]

@retronym said:
Patch from http://github.com/retronym/scala/tree/implicit-cache