OK. They should not be used, but we have no way of determining if a
IndexInput is actually closed, right? At least Lucene does not track
it. I run into this issue with Lazy Loading. I can still access the
value of a lazy field after the fieldsStream IndexInput is closed _if_ I
have not made my thread local clone yet, but do so after closing the
FieldsReader.
Code looks like (for example):
//This is my new, private Field implementation, which is an inner class
of FieldsReader
private class LazyField implements Fieldable
....
public String stringValue() {
if (fieldsData == null)
{
IndexInput localFieldsStream = (IndexInput) fieldsStreamTL.get();
if (localFieldsStream == null) {
localFieldsStream = (IndexInput) fieldsStream.clone();
fieldsStreamTL.set(localFieldsStream);
}
try {
localFieldsStream.seek(pointer);
//read in chars b/c we already know the length we need to read
if (chars == null || toRead > chars.length)
chars = new char[toRead];
localFieldsStream.readChars(chars, 0, toRead);
fieldsData = new String(chars, 0,
toRead);//fieldsStream.readString();
} catch (IOException e) {
throw new FieldReaderException(e);
}
}
return fieldsData instanceof String ? (String)fieldsData : null;
}
fieldsStreamTL is a private static ThreadLocal belonging to
FieldsReader. In FieldsReader.close() I do close the associated
ThreadsLocal clone if it exists, as in:
final void close() throws IOException {
fieldsStream.close();
indexStream.close();
IndexInput localFieldsStream = (IndexInput) fieldsStreamTL.get();
if (localFieldsStream != null) {
localFieldsStream.close();
fieldsStreamTL.set(null);
}
}
However if the application does something like:
public void testLazyClosedReader() throws Exception {
assertTrue(dir != null);
assertTrue(fieldInfos != null);
FieldsReader reader = new FieldsReader(dir, "test", fieldInfos);
assertTrue(reader != null);
assertTrue(reader.size() == 1);
Set lazyFieldNames = new HashSet();
//new String[]{DocHelper.LARGE_LAZY_FIELD_KEY,
DocHelper.LAZY_FIELD_KEY, DocHelper.LAZY_FIELD_BINARY_KEY};
lazyFieldNames.add(DocHelper.LARGE_LAZY_FIELD_KEY);
lazyFieldNames.add(DocHelper.LAZY_FIELD_KEY);
lazyFieldNames.add(DocHelper.LAZY_FIELD_BINARY_KEY);
Document doc = reader.doc(0, lazyFieldNames);
assertTrue("doc is null and it shouldn't be", doc != null);
Fieldable field = doc.getField(DocHelper.LAZY_FIELD_KEY);
assertTrue("field is null and it shouldn't be", field != null);
reader.close();
String value;
try {
value = field.stringValue();
assertTrue("Should not be able to read value: " + value + " since
the reader is closed", false);
} catch (Exception e) {
}
field = doc.getField(DocHelper.TEXT_FIELD_1_KEY);
assertTrue("field is null and it shouldn't be", field != null);
value = field.stringValue();
assertTrue("value is null and it shouldn't be", value != null);
assertTrue(value + " is not equal to " + DocHelper.FIELD_1_TEXT,
value.equals(DocHelper.FIELD_1_TEXT) == true);
}
This test fails, and I don't think it should. The assert in the try
block is activated. What am I missing?
What do you think of adding an boolean isClosed() onto IndexInput so
that we can test for closure. I suppose this could also be added to
anything that has close() as a method.
Doug Cutting wrote:
Grant Ingersoll wrote:
Should it be the case that you can clone a closed IndexInput and get
a valid object that is capable of reading? B/c this is what I am
seeing in my Lazy implementation (note, it seems to work fine...) I
am just not sure if it should work or if it is a bug.
Cloned IndexInputs are assumed to share resources with the original.
Lucene closes all original IndexInputs when the IndexReader is closed.
Cloned IndexInputs should not be used after the IndexReader is closed.
I don't think we bother to explicitly close clones (as they are
frequently cached in ThreadLocals, etc.).
Doug
---------------------------------------------------------------------
To unsubscribe, e-mail: [EMAIL PROTECTED]
For additional commands, e-mail: [EMAIL PROTECTED]
--
Grant Ingersoll
Sr. Software Engineer
Center for Natural Language Processing
Syracuse University
School of Information Studies
335 Hinds Hall
Syracuse, NY 13244
http://www.cnlp.org
Voice: 315-443-5484
Fax: 315-443-6886
---------------------------------------------------------------------
To unsubscribe, e-mail: [EMAIL PROTECTED]
For additional commands, e-mail: [EMAIL PROTECTED]
