gmane.comp.lang.scala.internals

Re: ToolBoxes and running multiple trees

Chris Hodapp — 2012-05-16T16:03:36

That is totally understandable.

I care about this because I want to reuse the ToolBox in DynamicProxy in 
order to improve performance (unscientifically, it makes method calls on 
the proxy take about half as long to resolve and execute). I can work 
around the issue by just catching the exception and resetting the FrontEnd. 
However, it really doesn't seem like ToolBoxes should have this limitation 
when there  is a performance cost to making a new one. You make the call. 
If you don't have time, I'll just catch the exception. 

On Wednesday, May 16, 2012 10:47:10 AM UTC-5, Eugene Burmako wrote:
>

Re: ToolBoxes and running multiple trees

Eugene Burmako — 2012-05-16T15:47:10

Well... Martin and I are totally swamped with this reflection refactoring.

Is this bug a blocker for you? If yes, I could take a look tomorrow.

On 16 May 2012 18:37, Chris Hodapp <clhodapp1-Re5JQEeQqe8AvxtiuMwx3w< at >public.gmane.org> wrote:

Re: ToolBoxes and running multiple trees

Chris Hodapp — 2012-05-16T15:37:25

Any progress, Eugene?

Re: virtpatmat and error messages

Adriaan Moors — 2012-05-16T15:28:08

yes

Re: Re: I've updated GenASM as of today's trunk, for easier merging.

iulian dragos — 2012-05-16T14:57:43

On Wed, May 16, 2012 at 4:18 PM, Johannes Rudolph
<johannes.rudolph-gM/Ye1E23mwN+BqQ9rBEUg< at >public.gmane.org> wrote:

This sounds like the best solution.

iulian

Re: Re: I've updated GenASM as of today's trunk, for easier merging.

iulian dragos — 2012-05-16T14:57:00

Yes, I read it, but somehow missed this. Sorry.

Since this is such a standard practice, I'm a bit surprised we can't
use jarjar.

virtpatmat and error messages

Eugene Burmako — 2012-05-16T14:49:56

Are there any plans to improve the situation?

I'm talking about stuff like:

* Could not typecheck extractor call blah blah
* Wrong number of arguments for <none>: (stats:
List[Extractors.this.global.Tree], expr:
Extractors.this.global.Tree)Extractors.this.global.Block
* "Cannot find value foo, cannot find value bar, cannot find value
quux" when you just made a typo in a name of an extractor that binds
foo, bar and quux

Each of these errors are more or less okay, but when all of them kick
in, then a few typos can turn an error log into an unintelligible
mess.

I'm not attaching bug reports and whatnot, because I assume that these
are all facets of a single known issue, right?

Re: GCD issue in both old and new patmat?

√iktor Ҡlang — 2012-05-16T14:26:43


Heck, I'd even settle for:

case (e: Exception) < at > (_: ClassCastException | _: CancelledKeyException |
_: IOException | _: ActorInitializationException) ⇒

But I think it's intuitive to assume that the type of the e will be the LUB
of the match.

Cheers,
√

Re: Re: I've updated GenASM as of today's trunk, for easier merging.

Johannes Rudolph — 2012-05-16T14:18:20

I think there are two issues:

 1) how to get a version of ASM which contains the patch
 2) how to package ASM with the compiler

Getting the patch into upstream ASM would only solve 1), right?

For issue 2) several proposals were discussed / tried:
 * use JarJar to alias ASM to another package. This doesn't work
currently because JarJar is run after the fact, meaning after the
complete compiler was once compiled with the original asm package
names. The problem as I understand it, is that JarJar doesn't know
about Scala's class-file metadata and so after aliasing Scala and Java
metadata conflicts.

 * include a fork of ASM with aliased package names directly in the
sources and use that

There are maybe two other options:
 * use JarJar on the asm.jar alone and then use the aliased package
names to refer to ASM from the scala compiler source code

 * don't alias at all but declare ASM as a maven/ivy dependency. This
way users of the compiler also using ASM could use the conflict
resolver of the dependency man

Re: Re: I've updated GenASM as of today's trunk, for easier merging.

Miguel Garcia — 2012-05-16T14:06:57

I've just asked at the ASM mailing list whether making public those members 
won't hurt: 
  http://mail.ow2.org/wws/arc/asm/2012-05/msg00026.html

Miguel
http://lampwww.epfl.ch/~magarcia/ScalaCompilerCornerReloaded/


On Wednesday, May 16, 2012 3:58:12 PM UTC+2, Adriaan Moors wrote:
>

Re: Re: I've updated GenASM as of today's trunk, for easier merging.

Adriaan Moors — 2012-05-16T13:58:12

... but I think we all agree the optimal solution is to get this patch into
ASM, right?
as far as I can tell, only 4 members are affected

On Wed, May 16, 2012 at 3:48 PM, Josh Suereth <joshua.suereth-Re5JQEeQqe8AvxtiuMwx3w< at >public.gmane.org>wrote:

Re: Re: I've updated GenASM as of today's trunk, for easier merging.

Josh Suereth — 2012-05-16T13:48:56

ASM is one of the  most commonly used dependencies in Java libraries.  If
we don't alias it, most likely we will have binary incompatible conflcits
with other libraries that use ASM.  Convention now is to alias ASM
internally using JarJar or other such tools.   Not necessarily from source.

Did you see the thread where we discussed this before?

On Wed, May 16, 2012 at 9:42 AM, iulian dragos <jaguarul-Re5JQEeQqe8AvxtiuMwx3w< at >public.gmane.org> wrote:

Re: GCD issue in both old and new patmat?

Miles Sabin — 2012-05-16T13:43:22

Fair enough, but it seems a little unfortunate that,

(new IllegalStateException).asInstanceOf[Throwable] match {
  case e : IllegalStateException => val sanity: Exception = e
  case e : ClassCastException => val sanity : Exception = e
}

which you might naively think is equivalent to Viktor's eg. is
perfectly OK, especially since one of the main uses of pattern
alternatives is to eliminate duplicate case clause RHS's in precisely
this kind of situation.

Cheers,


Miles

Re: Re: I've updated GenASM as of today's trunk, for easier merging.

iulian dragos — 2012-05-16T13:42:13

Why is it unacceptable to use the original package names?

Re: Re: I've updated GenASM as of today's trunk, for easier merging.

Josh Suereth — 2012-05-16T13:38:51

+1 on moving to src/asm.

As far as using it directly as a dependency, there's precedence to alias
ASM.   IF Jarjar were actively maintained, we could use that.   As it
stands, this is the best "compromise" we have, unless you're offering a
better solution.

- Josh

On Wed, May 16, 2012 at 9:32 AM, iulian dragos <jaguarul-Re5JQEeQqe8AvxtiuMwx3w< at >public.gmane.org> wrote:

Re: Re: I've updated GenASM as of today's trunk, for easier merging.

iulian dragos — 2012-05-16T13:32:24

I am strongly against pulling in more sources in the compiler project,
especially for a stable and maintained project like ASM. If I
understood correctly, the changes we needed were additions to work
around visibility? Could those be pushed upstream? I'm pretty sure the
ASM guys wouldn't mind supporting such a high-profile project like
Scala. If not, why not have a separate project for ASM, and
publish/maintain a fork *outside* the compiler tree?

And if all else fails, can we at least put those sources somewhere
else than src/compiler? src/asm?

iulian

On Sun, May 6, 2012 at 10:32 PM, Josh Suereth <joshua.suereth-Re5JQEeQqe8AvxtiuMwx3w< at >public.gmane.org> wrote:

Re: GCD issue in both old and new patmat?

Adriaan Moors — 2012-05-16T13:31:58

Miles, your example is different from Viktor's:

contrast
  val sanity: Exception = (e: Throwable)
to
  val sanity : A = (a: A)

On Wed, May 16, 2012 at 3:27 PM, Miles Sabin <miles-XKJT71GPLR04Q++5jOxPmw< at >public.gmane.org> wrote:

Re: GCD issue in both old and new patmat?

martin odersky — 2012-05-16T13:30:28

the root (selector type) to the leaves (variables in patterns),




Cheers

 - Martin

Re: GCD issue in both old and new patmat?

Miles Sabin — 2012-05-16T13:27:46

But surely the LUB here should be Exception not Throwable?

I'm not sure I see how Viktor's example is different from,

scala> class A ; class B extends A ; class C extends A
defined class A
defined class B
defined class C

scala> ((new B) : A) match { case a < at > (_ : B | _ : C) => val sanity : A = a }

scala>

Which is just fine.

Cheers,


Miles

Re: GCD issue in both old and new patmat?

Adriaan Moors — 2012-05-16T13:20:54

I think we're talking about different things here.

recall Viktor's example:
(new IllegalStateException).asInstanceOf[Throwable] match { case e < at > (_:
IllegalStateException | _: ClassCastException) => val sanity: Exception = e}

`e` has type `Throwable` since that's how alternatives are type checked,
thus you can't assign it to sanity

it's trivial to fix this in the type checker for this example, but we'd
also have to update the spec
don't know if now's a good time for that (well, actually, I would say it's
not)

        case Alternative(alts) =>
          val alts1 = alts mapConserve (alt => typed(alt, mode | ALTmode,
pt))
          treeCopy.Alternative(tree, alts1) setType lub(alts1 map (_.tpe))
//pt


On Wed, May 16, 2012 at 3:17 PM, Eugene Burmako <eugene.burmako-p8DiymsW2f8< at >public.gmane.org>wrote:

Re: GCD issue in both old and new patmat?

Eugene Burmako — 2012-05-16T13:17:27

Does this mean that I cannot match something of static type Tree against,
say, Apply?

On May 16, 2012 4:14 PM, "Adriaan Moors" <adriaan.moors-p8DiymsW2f8< at >public.gmane.org> wrote: