[
https://issues.apache.org/jira/browse/DAFFODIL-1975?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=16831380#comment-16831380
]
Michael Beckerle commented on DAFFODIL-1975:
--------------------------------------------
Research on behavior of IBM DFDL (by IBM) shows that zero-length sequence
groups can have their separators suppressed. So not all model groups are viewed
as "required" with respect to separators.
A sequence that exists purely for carrying an assertion will always be
zero-length, and so need not generate a separator when unparsing if the
dfdl:separatorSuppressionMode is anyEmpty, for example.
When parsing, this will still require some adaptation to avoid requiring a
separator.
> Incorrect separator behavior - missing separator for empty sequence
> (non-backward compatible)
> ---------------------------------------------------------------------------------------------
>
> Key: DAFFODIL-1975
> URL: https://issues.apache.org/jira/browse/DAFFODIL-1975
> Project: Daffodil
> Issue Type: Bug
> Components: Back End, Compatibility
> Affects Versions: 2.1.0
> Reporter: Michael Beckerle
> Assignee: Michael Beckerle
> Priority: Minor
>
> Consider the following
> {{ <xs:element name="NS_13a">
> <xs:complexType>
> <xs:sequence dfdl:separator="," dfdl:separatorPosition="infix"><!--
> OUTER -->
> <xs:sequence> <!-- INNER -->
> <xs:annotation>
> <xs:appinfo source="http://www.ogf.org/dfdl/";>
> <dfdl:assert>{fn:true()}</dfdl:assert>
> </xs:appinfo>
> </xs:annotation>
> </xs:sequence>
> <xs:element name="e2" type="xs:int" minOccurs="1"
> maxOccurs="unbounded" dfdl:textNumberPattern="#####"
> dfdl:occursCountKind="implicit" />
> </xs:sequence>
> </xs:complexType>
> </xs:element>}}
> The sequence labeled INNER has no syntax, only an assertion. However, the
> OUTER sequence specifies infix separator.
> Per clarifying discussion on the DFDL workgroup mailing list, and tests with
> IBM DFDL, daffodil behavior is wrong if it suppresses the separator after the
> INNER sequence. Separator suppression applies only to "optional" children,
> not required children. All model group instances are considered required,
> even if empty.
> Hence, valid data to match the above schema looks like
> ,1,2,3
> The leading comma there is required.
> Daffodil must be doing something clever to detect that this initial sequence
> is empty and so is suppressing the separator for it. But this is incorrect.
> Test test_NumSeq_13a illustrates the bug. (Test added as part of sequence and
> separators refactoring 2018-08-01)
> Fixing this may cause many tests to break. This is a major
> non-backward-compatible change, and so should be fixed carefully. Schemas
> outside of daffodil for delimited formats may be impacted also.
--
This message was sent by Atlassian JIRA
(v7.6.3#76005)
