[
https://issues.apache.org/jira/browse/PIG-70?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=12563183#action_12563183
]
Benjamin Reed commented on PIG-70:
----------------------------------
Benjamin, you are correct that PigContext is kind of a hodgepodge of necessary
information. I think most of it will be going away:
HOD is changing (AGAIN!). I haven't seen the latest direction, but it would
probably be good to move to the Hadoop specific package.
Also I the jobconf will move to the PlatformConfiguration Map. Once that
happens PigContext will look more like a singleton than a factory.
In a nutshell, I don't think we can finish this until PIG-66 and PIG-32 get in.
> Improve PigContext code by using Factory Pattern
> ------------------------------------------------
>
> Key: PIG-70
> URL: https://issues.apache.org/jira/browse/PIG-70
> Project: Pig
> Issue Type: Improvement
> Components: impl
> Affects Versions: 0.1.0
> Reporter: Benjamin Francisoud
> Attachments: PIG-70-v01.patch
>
>
> Even if the PigContext code is still quite small at the moment, for an
> outsider (me) it's already hard to understand :(
> If I understand correctly the PigContext purpose (on a Object Oriented point
> of view) is to hold various configuration objects like JobConf, JobClient,
> JobSubmissionProtocol...
> The initialization code mainly use the ExecType parameter but can also use
> quite complex code like: doHod(), initProperties(), connect()...
> (btw, the connect() method is actually doing 2 things: initializing some var
> and trying to connect, it initialization code should be move somewhere else)
> It is the perfect case to apply the [Factory
> Pattern|http://en.wikipedia.org/wiki/Factory_method_pattern] , you can also
> see [Replace Constructor with Factory
> Method|http://www.refactoring.com/catalog/replaceConstructorWithFactoryMethod.html]
> for more details.
> My proposal is to create a new PigContextFactory class, to old the
> initialization code make to PigContext and PigContextFactory 200 lines
> classes instead of one big 500 lines of code class.
> PigContext would hold some getter and setter and methods related to
> instantiate/run "functions"
> The new API would be:
> h4. PigContextFactory .java
> {code:java}
> public class PigContextFactory {
> public static PigContext getInstance(ExecType execType) {...}
> }
> {code}
> h4. PigContext.java
> {code:java}
> public class PigContext implements Serializable, FunctionInstantiator {
> public String getJobName(){...}
> public JobSubmissionProtocol getJobTracker() {...}
> public JobConf getConf() {...}
> public static Object instantiateFuncFromSpec(String funcSpec) throws
> IOException{...}
> public Object instantiateFuncFromAlias(String alias) throws IOException
> {...}
> public void registerFunction(String function, String functionSpec) {...}
> }
> {code}
> h4. Client code
> {code:java}
> PigContext context = PigContextFactory.getInstance(ExecType.MAPREDUCE);
> {code}
--
This message is automatically generated by JIRA.
-
You can reply to this email to add a comment to the issue online.
