Issue (1) is something I've encountered too, in the
SuffixMaxentModelResourceImpl class. There is a call to DataResource.getUrl()
which doesn't work if the resource is located in a jar file. Replacing this
with the following code (starting on line 55) fixed the problem:
//File modelFile = new File(dr.getUri());
InputStream is = dr.getInputStream();
DataReader dataReader = new PlainTextFileDataReader(is);
GISModelReader modelReader = new GISModelReader(dataReader);
iv_maxentModel = modelReader.getModel();
is.close();
-----Original Message-----
From: Chen, Pei [mailto:[email protected]]
Sent: Tuesday, June 11, 2013 12:50 PM
To: [email protected]
Subject: InputSteam instead of java.io.File
While working on the test cases in cTAKES, I've encountered couple of issues
and suggestions:
1) File or Url.getRawPath() became problematic if they are read in from
the jars from the classpath and which couldn't resolve to a physical File.
a. Suggestion: Wherever possible, replace loading of resouces via
java.io.File with InputStream instead. . We can add a new method in the
FileLocator util and deprecate the old File method.
2) Sentence Dectector is still using the OpenNLP 1.4 mechanism of loading
it's model file.
a. Suggestion: Let's update it to use the new 1.5 way similar to
POSTagger. (Remove non longer required classes: SuffixMaxentModelResourceImpl,
MaxentModelResource, SuffixSensitiveGISModelReader, classes etc.)
Background:
Certain unit tests fail because they can't be resolved via jars from the
classpath because the code is explicitly looking for File on disk instead of
input stream. But in order to solve it appropriately, it had a cascading
effect and required a lot more changes, but it's probably a good time to update
those projects anyhow.
--Pei
