[jsword-devel] regex searching

[DM Smith]

JSword search indexes support the full search syntax of Lucene. This has 
some support for regular expressions. Specifically, it allows for * to 
mean zero or more characters. I think we've got the Lucene flag to allow 
prefix wild cards. Note that this is more like what a shell uses: it is 
not a modifier for the previous character.

It was correctly noted that wash.*word will not search the verse as a 
whole. Lucene search is based on words. The correct pattern in Lucene 
would be wash*word, as '.' does not mean any character but rather it 
means a punctuation mark. So "wash.*word" would be split into "wash 
*word" and would find all verses with the word "wash" and any word 
ending with "word", such as "sword".

The default connector for JSword is OR. It would be good to add the 
ability for a user to change it to AND. With that, "wash* word" would 
produce the expected results as it would be interpreted as "wash* AND 
word" instead of "wash* OR word". (Choice would be some thing like 
"Search for ALL words instead of ANY words.") We made OR the default 
because it more closely matched various familiar search engines, such as 
Google, and because it is the default of Lucene.

Lucene does have a regular expression capability, but it is not part of 
JSword. It would be a good addition. Still, it would be based on words 
and not on the text of a verse.

Adding the ability to search an arbitrary regular expression would be a 
good addition. I don't think it would be too hard to add it. Jsword 
already has the interface for any search implementation. Java's Regex is 
a variant of Perl's but has a bit more power.  There are some issues: 
We'll be adding highlighting to Lucene's search. Adding that to Regex 
would be a separate effort. The regex search would be mutually exclusive 
from Lucene search, so that would need to be made obvious. (In The SWORD 
Project for Windows, they have both it is a bit confusing as it is not 
clear.)

And yes, it will need to go throw the "plain" text of each verse. It 
will be about as slow as creating an index. Basically, it will need to 
take the raw verse and strip the markup. (This is part of JSword already 
and Lucene indexing uses it.) And perhaps strip out the punctuation. 
Finally doing the search.

If you want, add an issue or two to Jira (www.crosswire.org/bugs) under 
JSword. That way it won't be forgotten.

In Christ,
     DM


On 02/23/2010 08:33 PM, Trenton D. Adams wrote:
> But then again, I wonder if it's even needed, who knows.
>
> ----- "Trenton D. Adams"<trent.jsword at trentonadams.ca>  wrote:
>
>    
>> From: "Trenton D. Adams"<trent.jsword at trentonadams.ca>
>> To: "J-Sword Developers Mailing"<jsword-devel at crosswire.org>
>> Sent: Tuesday, February 23, 2010 7:29:08 PM GMT -06:00 US/Canada Central
>> Subject: [jsword-devel] regex searching
>>
>> Hello,
>>
>> I'm getting the impression that regular expression searching is not at
>> all possible without implementing something that loads the books
>> itself, and goes through each verse of the bible.  Is this true?
>>
>> It seems like the Lucene Index just doesn't support regular expression
>> searches, eh?  And it also seems like SearcherFactory is currently not
>> finished being implemented as a SearcherFactory, correct?  Or, perhaps
>> other methods need to be added, like createSearcher(Book, Class
>> searcherClass)???  Then you could provide another searcher type.
>>
>> What would be appropriate for this? adding the new
>> SearcherFactory.createSearcher(), adding another "find" method to
>> AbstractBook/Book that is something like "find(SearchRequest,
>> Searcher)"???
>>
>> Are the books and everything abstract enough that I can search them by
>> loading the text?  Or would I have to restrict it to the book types I
>> know how to parse?
>>
>> Anything else?
>>
>> Thanks.
>>
>> _______________________________________________
>> jsword-devel mailing list
>> jsword-devel at crosswire.org
>> http://www.crosswire.org/mailman/listinfo/jsword-devel
>>      
> _______________________________________________
> jsword-devel mailing list
> jsword-devel at crosswire.org
> http://www.crosswire.org/mailman/listinfo/jsword-devel
>