Our JIRA users have somehow entered 0x1c-characters (file separator) to description-field on a couple of issues. When the database is exported to XML and imported back, the import fails due to the 0x1c character in a CDATA block. It can't even be fixed by changing the description using JIRA, as the invalid character is also in the change history. We had to remove these by hand from the database.
Marked this issue critical as it also breaks upgrading from 3.0.1 to 3.0.3 which relies on export/import.
We've been working on a fix for this, which is present but disabled in 3.0.3. If you edit WEB-INF/classes/jira-application.properties and set the jira.exportimport.cleanxml flag to true, then any invalid characters in the XML file will be ignored, and the import should succeed.
We've made the same XML cleaning utility available as a standalone tool, downloadable from:
http://repository.atlassian.com/atlassian-xml-cleaner/jars/
You can clean your data export by running 'java -jar atlassian-xml-cleaner-0.1.jar yourdata.zip|yourdata.xml > yourdata-clean.xml' and then importing yourdata-clean.xml
Let us know if you have any problems with this.
Cheers,
Jeff