[
https://issues.apache.org/jira/browse/UIMA-6128?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=17010504#comment-17010504
]
Rune Stilling commented on UIMA-6128:
-------------------------------------
We have found the root of the problem.
When serializing a CAS containing some characters, that in UTF-16 are encoded
via surrogates, the result is invalid XML-characters in the serialized UTF-8
encoded document making it unparsable.
The problem is coming from the Xalan serialization libraries that may be used
in UIMA via the edu.stanford.nlp:stanford-corenlp:3.9.1 and 3.9.2 dependencies
(dependent on xalan:xalan:2.7.0).
The bug is described here (and has never been fixed in an official release):
https://issues.apache.org/jira/browse/XALANJ-2617
We found the solution to be quite straight forward. We simply excluded the
Xalan (and Xerces dependencies) so that the code uses the default Java
implementation instead (org.xml.sax.ContentHandler::startElement())
We have attached two files, that may be used to reproduce the issue. If Xalan
is included, the test code will throw an exception when loading the generated
CASA XMI.
> Allow XMI to be optionally serialized with XML 1.1 instead of only 1.0
> ----------------------------------------------------------------------
>
> Key: UIMA-6128
> URL: https://issues.apache.org/jira/browse/UIMA-6128
> Project: UIMA
> Issue Type: New Feature
> Components: UIMA
> Reporter: Mario Juric
> Priority: Major
> Attachments: OddFeatureText.java, SimpleTypeSystem_TS.xml
>
>
> Some unicode characters are not handled by XML 1.0 and it can require some
> normalization or cleanup to be able to serialize the CAS to XMI, but
> requirements may not necessarily allow all such characters to be fully
> removed from the CAS. It can also be impossible to do such
> normalization/cleanup without full reprocess when converting data already
> stored as compressed binaries to XMI. Being able to optionally select XML 1.1
> instead of the default XML 1.0 would be an easier way for some to bypass many
> of those unicode issues.
> See also discussion on the UIMA mailing list:
> https://lists.apache.org/thread.html/7f8124b7be9ea20ab21dc616243e5661a0b7668a856532031fda71e3@%3Cuser.uima.apache.org%3E
> This feature request suggests that an additional SerialFormat is introduced,
> e.g. XMI_1_1, which can be selected as format parameter in the
> CasIOUtils.save methods.
>
>
--
This message was sent by Atlassian Jira
(v8.3.4#803005)
