lucene-dev mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From "Robert Muir (JIRA)" <j...@apache.org>
Subject [jira] [Updated] (LUCENE-6246) Fix DocsEnum -> PostingsEnum transition
Date Tue, 17 Feb 2015 17:26:13 GMT

     [ https://issues.apache.org/jira/browse/LUCENE-6246?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel
]

Robert Muir updated LUCENE-6246:
--------------------------------
    Attachment: LUCENE-6246_backwards.patch

Here is the backwards patch as described. its only for branch_5x... kinda large due to the
test.

I didn't yet renumber any FLAGS or anything, since my tests are passing. We can look into
that separately next, at least so its easier to reason about?

> Fix DocsEnum -> PostingsEnum transition
> ---------------------------------------
>
>                 Key: LUCENE-6246
>                 URL: https://issues.apache.org/jira/browse/LUCENE-6246
>             Project: Lucene - Core
>          Issue Type: Bug
>            Reporter: Robert Muir
>         Attachments: LUCENE-6246-trunk.patch, LUCENE-6246_backwards.patch
>
>
> The current back compat introduced in LUCENE-4524, does not really help the users calling
e.g. LeafReader.termDocsEnum() or LeafReader.termPositionsEnum(), because the former's return
value changes to PostingsEnum, its superclass, and the latter got removed.
> It also does not help users using TermsEnum.docs() or TermsEnum.docsAndPositions() which
got removed and just replaced with postings().
> DocsEnum is different, but not deprecated, instead only used by some codecs as a convenience
class. DocsAndPositionsEnum is removed.
> I think we can do this a little better. First, we need to fix trunk to work the way we
want it to look. I think we should have LeafReader.postings() and TermsEnum.postings(), and
everything should use PostingsEnum. This is simplest.
> But in 5.x, I think we should have DocsEnum and DocsAndPositionsEnum which are deprecated,
to help guide the user.
> The "sugar" methods on LeafReader that exist in 5.0 (termDocsEnum(), termPositionsEnum())
should be deprecated (with message to use postings()) and final, and can just wrap PostingsEnum.
There is no reuse and flags here so this is very simple.
> On TermsEnum its more complicated, but i dont think impossible. We should add back deprecated
and final termDocsEnum() and termPositionsEnum() (with message to use postings()) and these
deprecated ones can have an instanceof check, unwrapping back to PostingsEnum before they
invoke postings behind the scenes. 
> For the 2 remaining ones on TermsEnum that take flags, thats the most tricky. I actually
think we shouldn't change the existing constant values when we dont have to. And I don't think
the names FLAG_FREQS are special, i'd rather these just be constants like FREQS. I looked
thru JDK constants (http://docs.oracle.com/javase/7/docs/api/constant-values.html) and only
one class uses this FLAG_xxx prefix. 
> So I think we should have PostingsEnum.FREQS etc with new values, not conflicting with
the old FLAG_FREQS etc values (which we can add back, deprecated, to DocsEnum and DocsAndPositionsEnum).
We can even add a check to the deprecated methods that only valid values are passed.
> This just means we have contained back compat, only for deprecated and final sugar methods
in LeafReader and TermsEnum, and the 2 deprecated classes. I think we can live with that and
it would save users pain.



--
This message was sent by Atlassian JIRA
(v6.3.4#6332)

---------------------------------------------------------------------
To unsubscribe, e-mail: dev-unsubscribe@lucene.apache.org
For additional commands, e-mail: dev-help@lucene.apache.org


Mime
View raw message