Add numDeletedDocs to IndexReader

• Key: LUCENE-1131
• Type: New Feature
• Status: Closed
• Resolution: Fixed
• Priority: Minor
• Assignee: Otis Gospodnetic
• Reporter: Shai Erera
• Fix Version/s: 2.4

Description

Add numDeletedDocs to IndexReader. Basically, the implementation is as simple as doing:
public int numDeletedDocs() {
return deletedDocs == null ? 0 : deletedDocs.count();
}
in SegmentReader.
Patch to follow to include in all IndexReader extensions.

Comments

[Yonik Seeley]

How about just using maxDoc() - numDocs()?

[Shai Erera]

This is an option, however it will result in two calls to maxDoc() (once by maxDoc() and another by numDocs()).
Like I wrote, it's more of a convenience method and having a complete and clear API. This way, users of Lucene won't need to ask themselves how to obtain this number - they'll have an explicit API for that.

[Shai Erera]

A very simple patch that implements numDeletedDocs in all the necessary readers.

[Otis Gospodnetic]

I think maxDoc() is a cheap call, so calling it twice won't be a performance killer, esp. since this is not something you'd call frequently, I imagine.

However, I do agree about numDeletedDocs() being nice for hiding implementation details.

[Michael McCandless]

Otis is this one ready to go in?

[Otis Gospodnetic]

I think so - applies and compiles.

[Michael McCandless]

Hmm – this breaks back compat (adds new abstract method to IndexReader).

Why don't we fallback to default impl, in IndexReader, of maxDoc() - numDocs()? Patch is much less invasive, and, we don't break back compat? maxDoc() is indeed cheap.

[Shai Erera]

What if we implement numDeletedDocs() in IndexReader, instead of defining it abstract?
Those that extend IndexReader (outside the scope of the attached patch) can then choose to override the implementation or not.

The purpose of the patch is to add an explicit method which developers can use, rather than understand the logic on maxDoc() - numDocs(). Not all extended classes implement it this way BTW. SegmentReader just calls deletedDocs.count(), rather then calling the two separate methods.

[Michael McCandless]

What if we implement numDeletedDocs() in IndexReader, instead of defining it abstract?

Right, that's exactly what I'm thinking, with this body:

public int numDeletedDocs() {
  return maxDoc() - numDocs();
}

Then I think no classes need to override it (perf cost of calling 2 methods is tiny)?

[Shai Erera]

I agree with the body, that's what I had in mind.

As for extending classes, I agree that calling two methods has little performance overhead, but it just looks cleaner (for SegmentReader for example). Anyway, I don't have a strong opinion on whether we should override or not. I'll be fine with either.

[Otis Gospodnetic]

Sending CHANGES.txt
Sending src/java/org/apache/lucene/index/IndexReader.java
Transmitting file data ..
Committed revision 695510.