protected Object readResolve() {
if (gridBuilder == null) {
if (selectedJob != null) {
gridBuilder = new DownstreamProjectGridBuilder(selectedJob);
}
}
return this;
}
Persistence is XML-based,
and when the data is read back from disk and the XML does not contain data for a particular field,
the existing field value is left as-is.
"As-is" normally means the JVM default value (0, false
, null
, etc.),
because the persistence logic does not invoke your constructor to create your object as Java serialization would.
There are a few exceptions to this, where objects first create themselves and then load XML onto themselves.
These include Jenkins
and Item
,
but these exceptions are limited to the root object of persistence.
In these exceptional cases, the values set in the constructor will survive.
If you want to fill in your field with a non-trivial default value,
you can write a readResolve
method, which gets invoked right after your object is resurrected from persistence.
readResolve
is called by XStream, but it is not part of the Jenkins class hierarchy, so there is no @Override
annotation.
Just put it right in your class alongside your fields. For example:
protected Object readResolve() {
if (gridBuilder == null) {
if (selectedJob != null) {
gridBuilder = new DownstreamProjectGridBuilder(selectedJob);
}
}
return this;
}
If you need to force the Manage Old Data screen to list jobs, builds, etc. using your data in the old format so that it can be saved in the new format in bulk,
you cannot use readResolve
, since it will not notify this system of the problem.
Instead you must create a static nested class called ConverterImpl
extending XStream2.PassthruConverter
,
which should clean up the storage of your instance and finally call OldDataMonitor#report
to record the conversion.
Removing a field requires that you leave the field with the transient
keyword.
When Jenkins reads the old XML, XStream will set the value for this field,
but it will no longer be written back to XML when data is saved.
Renaming a field is a combination of the above:
mark your old field as transient
, declare your new field, and then migrate the data from the old format to the new in your readResolve
method:
protected transient Long myObjectId;
protected List<Long> myObjectIds;
protected Object readResolve() {
if (myObjectId != null) {
myObjectIds = Arrays.asList(myObjectId)
}
return this;
}
You decide to extend a class and create new choosable classes; for example, adding additional browsers to an SCM plugin.
The old data structure looked like this when you had only one class SCMBrowser
:
<browser>
<url>http://example.com/</url>
</browser>
Now you decide to add a new NewSCMBrowser
. All your SCMBrowsers
extend SCMBrowserBase
and your XML suddenly looks like this:
<browser class="org.jenkinsci.plugins.foo.NewSCMBrowser">
<url>http://example.com/</url>
</browser>
or
<browser class="org.jenkinsci.plugins.foo.SCMBrowser">
<url>http://example.com/</url>
</browser>
With new jobs, no problem. Old jobs however will probably break.
In your SCMBrowserBase
class add a readResolve
method (see the XStream FAQ):
// compatibility with earlier plugins
public Object readResolve() {
if (this.getClass() != SCMBrowserBase.class) {
return this;
}
// make sure to return the default SCMBrowser only if we no class is given in config.
try {
return new SCMBrowser(url.toExternalForm());
} catch (MalformedURLException e) {
throw new RuntimeException(e);
}
}
Sometimes, you need to rename packages or class names. If your serialization data includes a fully qualified class name (which happens, for example, if you have a collection of them), then measures must be taken to maintain backward compatibility.
To do this, use XSTREAM2#addCompatibilityAlias(String, Class)
to register aliases.
You need to do this against the right XStream instance,
as a few different instances are used to persist different parts of data.
Items#XSTREAM2
is used for serializing project configuration,
and Run#XSTREAM2
is used for serializing builds and their associated Action
s.
For example, to alias Foo
in the "old" package to the "updated" one,
you can use this method call:
Items.XSTREAM2.addCompatibilityAlias("org.acme.old.Foo", org.acme.updated.Foo.class);
To ensure your alias is registered early in the Jenkins boot sequence,
you can use the Initializer
annotation on a static method, e.g. in your DescriptorImpl
:
@Initializer(before = InitMilestone.PLUGINS_STARTED)
public static void addAliases() {
Items.XSTREAM2.addCompatibilityAlias("org.acme.old.Foo", Foo.class);
}
Since 1.507 Descriptor#getConfigFile
is overridable and XmlFile
can be instantiated with any XStream instance.