L2/03-088

Re: Comments on BIDI in L2/03-079
From: Mark Davis

The following are some comments relayed from Doug Felt on L2/03-079 for the
UTC meeting.

============

I scanned through L2/03-079 (Unicode: Getting Through the Next Ten Years --
Kaplan, Wissink, O'Donnell) and ran across the following:

.. Steve Atkin of IBM found language and technical flaws in the Bidirectional
Algorithm based on thorough review and a test implementation that found
issues in Unicode's reference implementations (some of which had been there
for years).

This is a bullet point in support of a claim that standards are issued and
finalized before there has been adequate implementation.

I just talked with Steve and looked over his paper (at
http://www.cs.fit.edu/~satkin/docs/bidi.pdf).  In the paper he says:

"The Unicode Java/C reference, Java 1.2, and ICU pass all test cases." (pg
8)
"We have identified potentially difficult areas in Unicode's reference
description." (pg 11)

Reading his paper I think it's fair to say that he finds there are still
some confusing aspects of the text description (though I don't find them
particularly confusing).  His emphasis is on the value of using a functional
language to represent algorithms such as this.  While the functional
language (Haskell) is useful in a research setting, perhaps, I think the
value of using common languages (Java, C/C++) that people are already
familiar with is obvious.

Anyway, I wanted to point out that Steve found no discrepancy between the
java, ICU, and reference implementations of the Bidi algorithm and his own
functional implementation of the standard.  The text description of the
algorithm is ok, though he finds ambiguity in one of the reordering rules,
he does not claim that it in any way contradicts the reference
implementations.  I think the assertion in L2/03-079 mischaracterizes
Steve's results.

I should emphasize that my comments are only in respect to this small
mention of the Bidi algorithm, and I am not making any statements about the
overall thrust of the L2/03-079 paper.

Doug