Mike Beckerle created DAFFODIL-2218:
---------------------------------------
Summary: ICU behavior incompatible - textNumberCheckPolicy lax is
lax about "+" signs. Was not before.
Key: DAFFODIL-2218
URL: https://issues.apache.org/jira/browse/DAFFODIL-2218
Project: Daffodil
Issue Type: Bug
Reporter: Mike Beckerle
ICU libraries changed behavior and now strict behavior is being lax about +
signs.
Daffodil should revert back to the latest ICU version that doesn't have this
problem.
This from a DFDL Workgroup email thread on this subject:
{code:java}
Re: [DFDL-WG] Action 313: Plus '+' sign and lax
textNumberCheckPolicyInboxxSteve Hanson <s...@uk.ibm.com> Fri, Aug 30, 10:56
AMto me, slawrence, DFDL-WG, Liam ICU changing behaviour in an incompatible
way is not good.
IBM DFDL is way behind, and is still
on ICU 51.2. We are limited in what we can do as we try to keep the
same level as IBM Integration Bus & WTX as we have had C namespacing
issues in the past.
Looking at the links, there are other
changes that have crept in when lenient.
- The string must
contain a complete prefix and suffix.
For example, if the pattern is "{#};(#)", then
"{123}" or "(123)" would match, but "{123",
"123}", and "123" would all fail.
(The latter strings would be accepted in lenient mode.)
-
Minus and plus signs can only appear if specified in the pattern.
In lenient mode, a plus or minus sign can always precede
a number.
In typical ICU fashion, even this is
not complete. It says nothing about what happens if the pattern has a sign
and the data doesn't.
I suggest you test all the combos with
Daffodil and establish the truth.
Then we need to decide what to do. If
there is no way of controlling this (eg, parameter or env var) then the
safest option is to backoff Daffodil to the latest ICU release that matches
the DFDL 1.0 spec, and change the spec so that the link to ICU is specific
rather than the generic link which is in the spec today
(http://www.icu-project.org/apiref/icu4c/classDecimalFormat.html#_details)
and which floats to the latest release. We can't have a moving target.
Regards
Steve Hanson
IBM Hybrid Integration, Hursley, UK
Architect, IBM
DFDL
Co-Chair, OGF
DFDL Working Group
s...@uk.ibm.com
tel:+44-1962-815848
mob:+44-7717-378890
Note: I work Tuesday to Friday
From:
Mike Beckerle <mbeckerle.d...@gmail.com>
To:
DFDL-WG <dfdl...@ogf.org>
Date:
29/08/2019 19:49
Subject:
[DFDL-WG] Action
313: Plus '+' sign and lax textNumberCheckPolicy
Sent by:
"dfdl-wg"
<dfdl-wg-boun...@ogf.org>
Looks like ICU changed behavior....
From: Steve Lawrence <slawre...@apache.org>
Sent: Thursday, August 29, 2019 1:30 PM
To: us...@daffodil.apache.org
Subject: Re: Plus '+' sign and lax textNumberCheckPolicy - was: Re: How
to model a fixed-length integer that may be padded with space on the left?
I think this is a difference in ICU version?
A little grepping through ICU source, I found a change [1] to their
number parsing logic in Dec 2017:
+ if (!isStrict) {
+ parser.addMatcher(WhitespaceMatcher.getInstance());
+ parser.addMatcher(new
PlusSignMatcher());
+ }
That looks to me like a change to make it so plus signs are always
matched in lax/lenient mode regardless of the pattern (Daffodils current
behavior). A couple minor changes have been made to that section, but
nothing that allows you to turn if off if lenient is on.
It's hard to tell in the git history what release that was in, but it
looks like around version 61, which is relatively new (Daffodil is on
version 62).
Also, the latest version of DecimalFormatProperties.java (looks to be an
internal implementation, so no online javadocs), has javadocs that
states that plus signs are always allowed in lenient/lax mode [2].
I think this is a change in ICU behavior in newer versions.
- Steve
[1]
https://github.com/unicode-org/icu/commit/68340c8464bd988477d6c88f46f9dfe4562a6d02#diff-565b07c255337881b4e06f766691667cR119-R122
[2]
https://github.com/unicode-org/icu/blob/master/icu4j/main/classes/core/src/com/ibm/icu/impl/number/DecimalFormatProperties.java#L53-L54
--
dfdl-wg mailing list
dfdl...@ogf.org
https://www.ogf.org/mailman/listinfo/dfdl-wg
Unless stated otherwise above:
IBM United Kingdom Limited - Registered in England and Wales with number
741598.
Registered office: PO Box 41, North Harbour, Portsmouth, Hampshire PO6
3AU
{code}
--
This message was sent by Atlassian Jira
(v8.3.4#803005)
