yes, what's under the hood? A story of ICode, Pickler, and ... UnPickler.

To all those following this thread,

There's a very clear introduction about Pickler in chapter 2 of
http://lamp.epfl.ch/teaching/projects/archive/coppel_report.pdf

Miguel

On Sun, Nov 15, 2009 at 08:03:15PM +0100, Miguel Garcia wrote:
> To all those following this thread,
>
> There's a very clear introduction about Pickler in chapter 2 of
> http://lamp.epfl.ch/teaching/projects/archive/coppel_report.pdf

Thank you! I have done my share of spelunking through lamp PDFs but I'd
never gotten around to reading that one.

On Sunday November 15 2009, Miguel Garcia wrote:
> ...
>
> There's a very clear introduction about Pickler in chapter 2 of
> http://lamp.epfl.ch/teaching/projects/archive/coppel_report.pdf

That document is now nearly two years old. To what extent does it
describe Scala 2.8 and to what extent is it obsolete?

> Miguel

Randall Schulz

From: "Randall R Schulz"
> That document is now nearly two years old. To what extent does it
> describe Scala 2.8 and to what extent is it obsolete?

Even if it were from last week, one would still need to cross-check with
trunk :-) Some aspects may be obsolete, but all in all saves time for
newcomers. And the discussion on reflection was new to me. It's a cool pdf.

BTW, I forgot to link the accompanying sources:
http://lamp.epfl.ch/teaching/projects/archive/coppel_sources.zip

Miguel

On Sun, Nov 15, 2009 at 11:25:35AM -0800, Randall R Schulz wrote:
> That document is now nearly two years old. To what extent does it
> describe Scala 2.8 and to what extent is it obsolete?

Here's my reliable source on such questions:

git whatchanged --since="Jan 12 2008" -- src/compiler/scala/tools/nsc/symtab/classfile/*Pickler.scala

I'm glad to have any high level documentation; being somewhat dated
doesn't bother me because I can't trust the details anyway.

On Sunday November 15 2009, Miguel Garcia wrote:
> From: "Randall R Schulz"
>
> > That document is now nearly two years old. To what extent does it
> > describe Scala 2.8 and to what extent is it obsolete?
>
> Even if it were from last week, one would still need to cross-check
> with trunk :-) Some aspects may be obsolete, but all in all saves
> time for newcomers. And the discussion on reflection was new to me.
> It's a cool pdf.

Indeed. Please don't take my question as looking a gift horse in the
mouth.

> ...
>
>
> Miguel

Randall Schulz

On Sun, Nov 15, 2009 at 10:49 AM, Miguel Garcia <miguel.garcia@tuhh.de> wrote:

Hi,

I've been further exploring some phases of nsc, and wondered finally what makes icode different from Java bytecode. Are there any notes summarizing that?

No, unfortunately. They will be, though.. in a few months time (my thesis).

So far I've found this: "Unlike Java bytecode, icode uses typed local variables"  [1]

ICode instructions are very close to java bytecode. Besides typed local variables, there are instructions for dealing with objects (in the Scala sense), and boxing/unboxing operations have their own opcode.

More interesting differences appear in structure. Unlike java bytecode which is linear, ICode uses a control-flow graph representation, where nodes are basic blocks. Exception handlers are 'covering' basic blocks instead of ranges, and there's a def-use link between the 'NEW' operation and its corresponding constructor call, to help analysis.

This list is probably incomplete, but should give you a starting point. A good way to see exactly what's different between bytecode and icode is to check the icode reader. That piece of code is reading java bytecodes from disk and presenting them to the analyzer in icode form, so that inlining and all the other goodies can be applied on library code.

Given that I haven't yet made a lot of time-traval across compilation phases, it hasn't striken me with the full force of enlightment what interplay there is between icode (on the one hand) and
scala.tools.nsc.symtab.classfile.Pickler and UnPlickler on the other.

None. Pickling is only about type signatures, and happens immediately after type checking.

cheers,
iulian
 

Pointers to information on the above would be welcome!


Miguel
http://www.sts.tu-harburg.de/people/mi.garcia


[1] Optimizing Higher-Order Functions in Scala, http://infoscience.epfl.ch/record/128135/files/paper.pdf

--
« Je déteste la montagne, ça cache le paysage »
Alphonse Allais