[
https://issues.apache.org/jira/browse/CXF-2720?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel
]
Michael Berkowitz updated CXF-2720:
-----------------------------------
Description:
Given an object of type {{Map<TKey, TValue>}}, where, say, {{TValue}} is a base
class extended by {{TValuePrime}}, when a serialized instance contains a value
of type {{TValuePrime}}, *MapType* will try to deserialize it as a {{TValue}},
ignoring the fact that it's really an instance of {{TValuePrime}}. In many
cases this will throw an exception when trying to _set_ a member that exists
only in {{TValuePrime}}. (We have observed this for the value class, but it
should be true of the key class as well).
We have fixed this - as far as we can tell - by swiping code from *ArrayType*,
which handles these things correctly. We're attaching our new version of
*MapType.readObject()*.
Lacking a deep understand of the Aegis code, we did this by intuition and
trial-and-error, so it may be inefficient or even technically incorrect, but so
far it works for us. Please get this fix, or its functional equivalent, into
the codebase.
Thanks in advance.
was:
Given an object of type {{Map<TKey, TValue>}}, where, say, {{TValue}} is a base
class extended by {{TValuePrime}}, when a serialized instance contains a value
of type {{TValuePrime}}, *MapType* will try to deserialize it as a {{TValue}},
ignoring the fact that it's really an instance of {{TValuePrime}}. In many
cases this will throw an exception when trying to _set_ a member that exists
only in {{TValuePrime}}. (We have observed this for the value class, but it
should be true of the key class as well).
We have fixed this - as far as we can tell - by swiping code from *ArrayType*,
which handles these things correctly. Here is our new version of
*MapType.readObject()*:
{{ public Object readObject(MessageReader reader, Context context) throws
DatabindingException {
Map<Object, Object> map = instantiateMap();
try {
while (reader.hasMoreElementReaders()) {
MessageReader entryReader = reader.getNextElementReader();
if (entryReader.getName().equals(getEntryName())) {
Object key = null;
Object value = null;
while (entryReader.hasMoreElementReaders()) {
MessageReader evReader =
entryReader.getNextElementReader();
if (evReader.getName().equals(getKeyName())) {
Type kType =
TypeUtil.getReadType(evReader.getXMLStreamReader(), context.getGlobalContext(),
getKeyType());
key = kType.readObject(evReader, context);
} else if (evReader.getName().equals(getValueName())) {
Type vType =
TypeUtil.getReadType(evReader.getXMLStreamReader(), context.getGlobalContext(),
getValueType());
value = vType.readObject(evReader, context);
} else {
readToEnd(evReader);
}
}
map.put(key, value);
} else {
readToEnd(entryReader);
}
}
return map;
} catch (IllegalArgumentException e) {
throw new DatabindingException("Illegal argument.", e);
}
}}}
Lacking a deep understand of the Aegis code, we did this by intuition and
trial-and-error, so it may be inefficient or even technically incorrect, but so
far it works for us. Please get this fix, or its functional equivalent, into
the codebase.
Thanks in advance.
> MapType doesn't deserialize subclass instances
> ----------------------------------------------
>
> Key: CXF-2720
> URL: https://issues.apache.org/jira/browse/CXF-2720
> Project: CXF
> Issue Type: Bug
> Components: Aegis Databinding
> Affects Versions: 2.2.6
> Environment: All
> Reporter: Michael Berkowitz
> Attachments: MapType.java
>
> Original Estimate: 24h
> Remaining Estimate: 24h
>
> Given an object of type {{Map<TKey, TValue>}}, where, say, {{TValue}} is a
> base class extended by {{TValuePrime}}, when a serialized instance contains a
> value of type {{TValuePrime}}, *MapType* will try to deserialize it as a
> {{TValue}}, ignoring the fact that it's really an instance of
> {{TValuePrime}}. In many cases this will throw an exception when trying to
> _set_ a member that exists only in {{TValuePrime}}. (We have observed this
> for the value class, but it should be true of the key class as well).
> We have fixed this - as far as we can tell - by swiping code from
> *ArrayType*, which handles these things correctly. We're attaching our new
> version of *MapType.readObject()*.
> Lacking a deep understand of the Aegis code, we did this by intuition and
> trial-and-error, so it may be inefficient or even technically incorrect, but
> so far it works for us. Please get this fix, or its functional equivalent,
> into the codebase.
> Thanks in advance.
--
This message is automatically generated by JIRA.
-
You can reply to this email to add a comment to the issue online.
