Github user paul-rogers commented on a diff in the pull request:

    https://github.com/apache/drill/pull/574#discussion_r77697563
  
    --- Diff: 
exec/java-exec/src/main/java/org/apache/drill/exec/expr/fn/DrillFunctionRegistry.java
 ---
    @@ -218,4 +302,141 @@ private void 
registerOperatorsWithoutInference(DrillOperatorTable operatorTable)
           }
         }
       }
    +
    +  /**
    +   * Function registry holder. Stores function implementations by jar 
name, function name.
    +   */
    +  private class GenericRegistryHolder<T, U> {
    +    private final ReadWriteLock readWriteLock = new 
ReentrantReadWriteLock();
    +    private final AutoCloseableLock readLock = new 
AutoCloseableLock(readWriteLock.readLock());
    +    private final AutoCloseableLock writeLock = new 
AutoCloseableLock(readWriteLock.writeLock());
    +
    +    // jar name, Map<function name, List<signature>
    +    private final Map<T, Map<T, List<T>>> jars;
    +
    +    // function name, Map<signature, function holder>
    +    private final Map<T, Map<T, U>> functions;
    +
    +    public GenericRegistryHolder() {
    +      this.functions = Maps.newHashMap();
    +      this.jars = Maps.newHashMap();
    +    }
    +
    +    public void addJar(T jName, Map<T, Pair<T, U>> sNameMap) {
    +      try (AutoCloseableLock lock = writeLock.open()) {
    +        Map<T, List<T>> map = jars.get(jName);
    +        if (map != null) {
    +          removeAllByJar(jName);
    +        }
    +        map = Maps.newHashMap();
    +        jars.put(jName, map);
    +
    +        for (Entry<T, Pair<T, U>> entry : sNameMap.entrySet()) {
    +          T sName = entry.getKey();
    +          Pair<T, U> pair = entry.getValue();
    +          addFunction(jName, pair.getKey(), sName, pair.getValue());
    --- End diff --
    
    Rather expensive. We have the function map right in our hands. Yet, on 
every add function, we look it up again.
    
    Perhaps the code would be simpler if we declared intermediate classes such 
as one that is the method registry for a function, rather than using generic 
classes. That way, your method registry class could do the work to register 
methods concisely rather than trying to do it generically as here.
    
    Also, having smaller classes makes unit testing easier. (This class is unit 
tested, right? Actually, it can't be: the class is private so not reachable 
from a unit test...) This raises a question: how will we test the detailed 
logic here if we don't have unit tests? Using system level tests that send SQL 
to Drill?


---
If your project is set up for it, you can reply to this email and have your
reply appear on GitHub as well. If your project does not have this feature
enabled and wishes so, or if the feature is enabled but not working, please
contact infrastructure at [email protected] or file a JIRA ticket
with INFRA.
---

Reply via email to