[ovirt-devel] Some ideas on oVirt Java SDK

[Juan Hernández]

On 12/01/2014 04:13 PM, Vojtech Szocs wrote:
> 
> 
> ----- Original Message -----
>> From: "Juan Hernández" <jhernand at redhat.com>
>> To: "Michael Pasternak" <mishka8520 at yahoo.com>, "Vojtech Szocs" <vszocs at redhat.com>, devel at ovirt.org
>> Sent: Monday, December 1, 2014 9:54:51 AM
>> Subject: Re: [ovirt-devel] Some ideas on oVirt Java SDK
>>
>> On 11/30/2014 12:26 PM, Michael Pasternak wrote:
>>> Hey Vojtech,
>>>
>>> How are you?, please see my reply inline.
>>>
>>>
>>>
>>>     On Friday, November 28, 2014 5:26 PM, Vojtech Szocs
>>>     <vszocs at redhat.com <mailto:vszocs at redhat.com>> wrote:
>>>
>>>
>>>     Hi guys,
>>>
>>>     since the initial (small, working & well-tested) version of oVirtJS
>>>     JavaScript SDK is finished [*], I've started working on GWT wrapper
>>>     for oVirtJS.
>>>
>>>     While analyzing/reverse-engineering oVirt Java SDK, some thoughts
>>>     came to my mind, and I wanted to share them with you.
>>>
>>>     [*] TODO(vszocs) upload new patchset with all recent changes
>>>
>>>     First, the way XJC (JAXB binding compiler that generates Java beans
>>>     out of REST XSD schema) is invoked looks a bit weird to me, as Java
>>>     SDK's XsdCodegen does this:
>>>
>>>       Runtime.getRuntime().exec(command)
>>>
>>>     Why not simply use existing Maven plugins to invoke XJC?
>>>     - either: https://github.com/highsource/maven-jaxb2-plugin
>>>     <https://github.com/highsource/maven-jaxb2-plugin>
>>>
>>>
>>> [MP] sdk was using jaxb to begin with, it was replaced with XJC just
>>> recently,
>>> btw Juan, what was the motivation behind this?
>>
>> This didn't change, the use of "xjc" is there since commit 95a25a4, Nov
>> 12 2012.
>>
>> Note that using Maven for this isn't as simple as it may look. The
>> development model of the SDK is that the maven build does *not* generate
>> any code, it just builds what has been manually generated previously.
> 
> To clarify, my question was meant for "ovirt-engine-sdk-java-codegen"
> project and its org.ovirt.engine.sdk.codegen.Main class that produces
> Java classes out of XSD as part of XsdCodegen.generate() method.
> 
> But if XsdCodegen invokes XJC programatically, what is the purpose of:
> 
>   org.jvnet.jaxb2.maven2:maven-jaxb22-plugin:generate
> 
> in "ovirt-engine-sdk-java-codegen" project's pom.xml?
> 
> Is it related to what XsdCodegen is doing?
> 

The code generator invokes "xjc" directly in order to generate from the
XML schema the code that will eventually be part of the generated SDK.
In order to do its work it needs to parse the RSDL metadata, and for
that it uses JAXB and classes generated from the XML schema. Those
classes are generated as part of the build process of the code
generator. So the XML schema is converted into Java classes twice: once
for the internal use of the generator (during build time of the
generator), and another time for the generated SDK (during run time of
the generator). This is convenient in order to avoid dependencies
between the generator and the SDK.

>>
>>> (REST api uses jaxb as well so we used to have 1x1 mappings)
>>>  
>>>
>>>     - or: http://mojo.codehaus.org/jaxb2-maven-plugin/
>>>     <http://mojo.codehaus.org/jaxb2-maven-plugin/>
>>>
>>>
>>> [MP] same.
>>>
>>>
>>>     Second, and most importantly, what's the point of having "group"
>>>     entities? I'll give an example - api.xsd contains this:
>>>
>>>       <xs:complexType name="DataCenters">
>>>         <xs:complexContent>
>>>           <xs:extension base="BaseResources">
>>>             <xs:sequence>
>>>               <xs:annotation>
>>>                 <xs:appinfo>
>>>                     <jaxb:property name="DataCenters"/>
>>>                 </xs:appinfo>
>>>               </xs:annotation>
>>>               <xs:element ref="data_center" minOccurs="0"
>>>     maxOccurs="unbounded"/>
>>>             </xs:sequence>
>>>           </xs:extension>
>>>         </xs:complexContent>
>>>       </xs:complexType>
>>>
>>>     (Same as above for Hosts, Clusters, VMs, etc.)
>>>
>>>     This results in following (IMHO rather meaningless) Java class
>>>     being generated by XJC:
>>>
>>>     public class DataCenters extends BaseResources {
>>>
>>>         @XmlElement(name = "data_center")
>>>         protected List<DataCenter> dataCenters;
>>>
>>>         public List<DataCenter> getDataCenters() {
>>>             if (dataCenters == null) {
>>>                 dataCenters = new ArrayList<DataCenter>();
>>>             }
>>>             return this.dataCenters;
>>>         }
>>>
>>>         public boolean isSetDataCenters() {
>>>             return ((this.dataCenters!=
>>>     null)&&(!this.dataCenters.isEmpty()));
>>>         }
>>>
>>>         public void unsetDataCenters() {
>>>             this.dataCenters = null;
>>>         }
>>>
>>>     }
>>>
>>>     Instead, we could use @XmlElementWrapper as described in [1]
>>>     to avoid generating "group" entities altogether.
>>>
>>>     [1] https://github.com/dmak/jaxb-xew-plugin
>>>     <https://github.com/dmak/jaxb-xew-plugin>
>>>
>>>     The fact that Java SDK provides decorator for each specific
>>>     resource collection (like DataCenters), instead of having ONE
>>>     resource collection type, greatly complicates overall design
>>>     and code-gen aspect.
>>>
>>>
>>> [MP] Well, i guess now is speaking JS constraints ghost, am i right?,
>>> in any case, the reasons for having decorator per collection are:
>>>
>>> 1. compliance with REST API (all SDKs and REST api are sharing same well
>>> know architecture)
>>> 2. "decorator" is a well known and commonly used java design pattern
>>> 3. having one resource type serving all collections would create a
>>> bottleneck
>>> (well it might depend on how you implementing it, but still in my view
>>> it's less convenient/readable
>>> than dedicated collection with own context, verbs and behavior),
>>>
>>> after all the purpose of sdk is being java client serving application in
>>> "Java" way
>>> (i.e type-safe + well bounded interface), while JS use-cases & paradigms
>>> are totally
>>> different, just consider:
>>>
>>> [1] java-sdk stile
>>>
>>> Disk snapshotDisk =
>>> api.getVms().get('my-vm').getSnapshots().get('my-snapshot').getDisks().get('my-disk')
>>>
>>> [2] JS style you propose
>>>
>>> Disk snapshotDisk = getCollections().get(new Params[] { Disk.class,
>>> 'my-vm', 'my-snapshot', 'my-disk'})
>>>
>>> notice:
>>> =====
>>>
>>> in [2] you have a bunch of parameters disconnected form any context
>>> where order
>>> is *important* (other way you heuristic guesses what user meaning by
>>> these params won't work),
>>> obviously it's fragile and error prone,
>>>
>>> while [1] is readable, well bounded, defending it's consumers from
>>> potentials errors
>>> (exactly what SDK should look like),
>>>
>>> hope it helps.
>>>
>>>
>>>
>>>
>>>
>>>
>>> On Friday, November 28, 2014 5:26 PM, Vojtech Szocs <vszocs at redhat.com>
>>> wrote:
>>>
>>>
>>> Hi guys,
>>>
>>> since the initial (small, working & well-tested) version of oVirtJS
>>> JavaScript SDK is finished [*], I've started working on GWT wrapper
>>> for oVirtJS.
>>>
>>> While analyzing/reverse-engineering oVirt Java SDK, some thoughts
>>> came to my mind, and I wanted to share them with you.
>>>
>>> [*] TODO(vszocs) upload new patchset with all recent changes
>>>
>>> First, the way XJC (JAXB binding compiler that generates Java beans
>>> out of REST XSD schema) is invoked looks a bit weird to me, as Java
>>> SDK's XsdCodegen does this:
>>>
>>>   Runtime.getRuntime().exec(command)
>>>
>>> Why not simply use existing Maven plugins to invoke XJC?
>>> - either: https://github.com/highsource/maven-jaxb2-plugin
>>> - or: http://mojo.codehaus.org/jaxb2-maven-plugin/
>>>
>>> Second, and most importantly, what's the point of having "group"
>>> entities? I'll give an example - api.xsd contains this:
>>>
>>>   <xs:complexType name="DataCenters">
>>>     <xs:complexContent>
>>>       <xs:extension base="BaseResources">
>>>         <xs:sequence>
>>>           <xs:annotation>
>>>             <xs:appinfo>
>>>                 <jaxb:property name="DataCenters"/>
>>>             </xs:appinfo>
>>>           </xs:annotation>
>>>           <xs:element ref="data_center" minOccurs="0"
>>> maxOccurs="unbounded"/>
>>>         </xs:sequence>
>>>       </xs:extension>
>>>     </xs:complexContent>
>>>   </xs:complexType>
>>>
>>> (Same as above for Hosts, Clusters, VMs, etc.)
>>>
>>> This results in following (IMHO rather meaningless) Java class
>>> being generated by XJC:
>>>
>>> public class DataCenters extends BaseResources {
>>>
>>>     @XmlElement(name = "data_center")
>>>     protected List<DataCenter> dataCenters;
>>>
>>>     public List<DataCenter> getDataCenters() {
>>>         if (dataCenters == null) {
>>>             dataCenters = new ArrayList<DataCenter>();
>>>         }
>>>         return this.dataCenters;
>>>     }
>>>
>>>     public boolean isSetDataCenters() {
>>>         return ((this.dataCenters!= null)&&(!this.dataCenters.isEmpty()));
>>>     }
>>>
>>>     public void unsetDataCenters() {
>>>         this.dataCenters = null;
>>>     }
>>>
>>> }
>>>
>>> Instead, we could use @XmlElementWrapper as described in [1]
>>> to avoid generating "group" entities altogether.
>>>
>>> [1] https://github.com/dmak/jaxb-xew-plugin
>>>
>>> The fact that Java SDK provides decorator for each specific
>>> resource collection (like DataCenters), instead of having ONE
>>> resource collection type, greatly complicates overall design
>>> and code-gen aspect.
>>>
>>> In oVirtJS GWT wrapper, we'll avoid above complication through
>>> single resource collection type (having common methods like
>>> get(id), list() etc) for all resources.
>>>
>>> Regards,
>>> Vojtech

-- 
Dirección Comercial: C/Jose Bardasano Baos, 9, Edif. Gorbea 3, planta
3ºD, 28016 Madrid, Spain
Inscrita en el Reg. Mercantil de Madrid – C.I.F. B82657941 - Red Hat S.L.