ISSUE-26: ROUNDTRIP ISSUE: Replication of original rules after roundtripping to RIF

ISSUE-26: ROUNDTRIP ISSUE: Replication of original rules after roundtripping to RIF

http://www.w3.org/2005/rules/wg/track/issues/26

Raised by: Christian de Sainte Marie
On product: Design Constraints (Goals, Requirements)

(This issue was originally recorded in Sandro's email of 11/4/06, reporting on 
the naming breakout at F2F4; his text is incorporated below.)

ROUNDTRIP ISSUE: When you round-trip a ruleset from a language L through 
a RIF dialect in which L is covered, you are guaranteed to get back a ruleset 
with identical semantics. But what about the aspects which are not part of
the semantics of the ruleset, such as variable names, whether a variable
is considered "anonymous", rule names, non-normal forms, and ordering in 
languages where ordering affects performance but not semantics?  What 
should be the extent of replication of the original ruleset after 
roundtripping?

Question: What non-semantic features of the original ruleset in a language L, 
which has been translated to RIF, should be replicated when that ruleset is 
translated back to the language L?

Possible features include:
1. original variable names
2. whether variable is "anonymous"
3. names of rules in L
4. non-normal forms used in L
5. original ordering of clauses in rules
There may be others.  (Please add.)

What are some examples in which *not* replicating (1) - (4) when roundtripping 
a ruleset would have an impact?  Possible kinds of impact, e.g., 
(a) degrades efficiency of reasoning with the rule in some reasoners commonly 
used with L;
(b) degrades accuracy of reasoning with the rule in some reasoners commonly 
used with L;
(c) makes the rule harder for humans to read or recognize.
(Please add to this list.)

Impact on RIF: For features that should be replicated upon roundtripping:
 (i) information about the features would have to be tracked in some way to 
make it available to the translator;
(ii) in cases where there is a one-to-one mapping of a rule feature of L to 
RIF, this would not be an issue; but otherwise, RIF would need some way to 
represent the original non-semantic feature (as meta-information about the 
rule), in order to preserve it for roundtripping.

Received on Tuesday, 12 December 2006 20:06:24 UTC