Document status | DRAFT |
---|---|
Document owner |
Introduction
OpenNLP uses System.out()
and System.err()
by default for logging. This is fine for instances in which OpenNLP's CLI tools are being used but may not be ideal when OpenNLP is used as a library. Previous work in this area includes
-
OPENNLP-675Getting issue details...
STATUS
. This proposal presents a method for allowing the user to customize the default logging behavior when OpenNLP is used as a library.
Current State
As discussed above, OpenNLP currently (as of 1.7.2) defaults to using System.out()
and System.err()
for logging messages and errors, respectively.
The Problem
When using OpenNLP as a library the output of logs to standard out and standard error is not ideal as these logs often need to be captured by the application for external storage and reporting (and also just to keep from cluttering up standard out).
Proposed Solution
The proposed solution is to create an OpenNLP Logger
interface that developers can implement to customize the logging. The user can provide their own implementation of this interface to control OpenNLP's logging. This interface will have to exist in a new project (perhaps opennlp-model
?) in order to avoid circular dependencies. (The opennlp-tools
project will have a dependency on this project. The user's project can either have an explicit dependency on opennlp-model
or a transitive dependency based on the project's requirements.)
The naming of objects in this proposal is mainly for illustrative purposes. I'm not proposing any specific naming and am open to naming suggestions.
An example of the logging interface is:
/** * Provides logging capabilities. **/ public interface Logger { void log(String message); void error(String message); void error(String message, Exception ex); }
The default implementation used when the user has not provided an implementation of Logger
will be:
/** * Default implementation of {@link Logger} that logs * all messages to standard out and all errors * to standard error. **/ public class DefaultLogger implements Logger { @Override public void log(String message) { System.out.println(message); } @Override public void error(String message) { System.err.println(message); } @Override public void error(String message, Exception ex) { System.err.println(message); System.err.println(ex.toString()); } }
The DefaultLogger
class mimics the current behavior of OpenNLP of using System.out()
.
A new project contains the Logger
interface to avoid circular dependencies.
A new class in OpenNLP would be created that stores a static reference to the Logger
implementation. By default this static variable would reference the DefaultLogger
. The user can set their own Logger
implementation at any time through the setLogger()
function.
/** * Provides access to the {@link Logger}. **/ public class LoggerConfiguration { private static Logger logger = new DefaultLogger(); public static void setLogger(Logger l) { logger = l; } public static Logger getLogger() { return logger; } }
Existing logging statements in OpenNLP would be modified to perform the logging via the LoggingConfiguration
class. For example, from NameSampleCountersStream:
Current | Proposed |
---|---|
public void printSummary() { System.out.println("Training data summary:"); System.out.println("#Sentences: " + getSentenceCount()); System.out.println("#Tokens: " + getTokenCount()); int totalNames = 0; for (Map.Entry<String, Integer> counter : getNameCounters().entrySet()) { LoggerConfiguration.getLogger().log("#" + counter.getKey() + " entities: " + counter.getValue()); totalNames += counter.getValue(); } } | public void printSummary() { LoggerConfiguration.getLogger().log("Training data summary:"); LoggerConfiguration.getLogger().log("#Sentences: " + getSentenceCount()); LoggerConfiguration.getLogger().log("#Tokens: " + getTokenCount()); int totalNames = 0; for (Map.Entry<String, Integer> counter : getNameCounters().entrySet()) { LoggerConfiguration.getLogger().log("#" + counter.getKey() + " entities: " + counter.getValue()); totalNames += counter.getValue(); } } |
Summary
This proposal:
- Presents a way to let users of OpenNLP as a library to control logging.
- Requires:
- A new project that contains a Logger interface.
- Modifying current
System.out()
andSystem.err()
calls to reference the newLoggerConfiguration
class.
Future Work
A similar approach could be taken for functions that expect a PrintStream
object.