Distributed Meta-Programming

Rui Shi, Chiyan Chen and Hongwei Xi

Boston University

Outline

MP vs. DMP Typeful Code Representation The Language dist

The Language +dist

Discussion and Conclusion

Meta-Programming

(define (run code) (eval code nil))(define (power n x)(if (= n 0) 1 `(* ,x ,(power (- n 1) x))))

;;; (power 2 ‘x) yields (* x (* x 1))(define square(run `(lambda (x) ,(power 2 ‘x))))

;;; (power 3 ‘y) yields (* y (* y (* y 1)))(define cube(run `(lambda (y) ,(power 3 ‘y))))

Distributed Meta-Programming

Code

Code

Run-time code generation

Distributed execution

Outline

MP vs. DMP Typeful Code Representation The Language dist

The Language +dist

Discussion and Conclusion

`(“abc” + 10) (ill-typed) `(( y.x) 10) (open code)

Un-typed Code Representation

No guarantee on the safety of generated code. Code can be syntactically ill-formed or ill-typed,

which lead to run-time errors when executed at remote sites.

Issues of locality and heterogeneity of resources Different sites may have different resources and thus likely to

provide different services.

f : (Int ! Int)@L

L `(f 100)

L’

Typeful Code Representation (T.C.R) The types of object programs can be reflected in the types of meta

programs E.g. code of type h , T i represents an object program of type T under

the type environment .

The type of the code also reflects where it should be sent for execution. E.g. code of type h L, , T i can be sent to location L for execution.

The compiler can statically guarantee that only well-typed code can be constructed at run-time and sent to proper locations for execution.

Outline

MP vs. DMP Typeful Code Representation The Language dist

The Language +dist

Discussion and Conclusion

The Language dist

First-Order Code Constructors

For example, the term x. y. y (x) can be represented as:

which can be of type (with static variables properly instantiated)

Lam (Lam (App (One) (Shi (One))))

8 . h , , int ! (int ! int) ! int i

Some Primitive Functions

Constant Messages

Constant messages refer to the messages which can be interpreted at all sites.

Example: perform addition of two integers at remote site

An Example in dist

f : (Int ! Int)@L

L `(f 100)

rgetint (L, rexec (L, App (Lift (n2m f)), Lift 100) ))

of type h L, , int i

Outline

MP vs. DMP Typeful Code Representation The Language dist

The Language +dist

Discussion and Conclusion

The Language +dist

We extend dist to +dist with some language constructs

adopted from meta-programming (supported in Scheme and MetaML):

`(e) for the code representation of e ^(e) for spliceing the code e into some context %(¢) is a shorthand for ^(Lift(¢))

An Example in Concrete Syntax of +

dist

typedef CodeType (L: loc) = <L, nil, (int -> int) -> int>

fun rZeroFind2{L:loc} (L:loc(L)) (n: (int -> int) @ L) : int = let val zeroFindCode : CodeType (L) = `(lam f => (fix aux i => if f (i) = 0 then i else aux (i+1)) 0) in rexecInt (L, `(^zeroFindCode %(n2m n))) end

Outline

MP vs. DMP Typeful Code Representation The Language dist

The Language +dist

Discussion and Conclusion

Discussion

We have a transformation scheme from +dist to

dist to justify the soundness of +dist .

We have a process to elaborate programs (partial type inference) in the concrete syntax to the formal syntax of +

dist

DMP is implemented as a part of ATS (Applied Type System) framework.

Conclusion

A simple and general approach to support typed mobile computing.

Combines meta-programming with distributed programming in a coherent manner.

In future, we are planning to study issues such as exception handling and distributed garbage collection in this context.

Thank You!