On Tue, Jun 8, 2010 at 7:13 AM, Galen Charlton <[email protected]> wrote:
> Hi,
>
> On Jun 7, 2010, at 11:16 PM, Mike Rylander wrote:
>> ISTM the same could be achieved without the cost of an XSLT with
>>
>> //marc:datafie...@tag='245' or (@tag='880' and
>> starts-with(./marc:subfie...@code='6']/text(),
>> '245'))]/marc:subfie...@code='a']
>>
>> I could certainly be missing something, though.
>
> Indeed it could, but using the XSLT allows the XPath to be more concise 
> without having to make the user worry about 880 fields beyond remembering to 
> set format to 'marc21expand880'.  In the particular database that inspired 
> this patch, where for at least the first pass the indexing is based on a 
> direct mapping of MARC subfields to index names per the legacy ILS's 
> definitions, there are about 180 MARC tags involved spread among 12 indexes.  
> The time required to create such indexes during ingest wasn't noticeably 
> longer that that required for the MODS-based indexes.
>

I wouldn't expect it to be slower, so that's good confirmation, but I
would expect the pure (if somewhat more complicated) raw marcxml XPath
to be noticeably faster.  A test of that would be useful information,
but not critical to the decision for core inclusion IMO.

That being said, I agree that's a reasonable tradeoff to give users --
(potentially ... until tested) better batch ingest speed (I'm not
concerned much with per-record costs of XSLT) vs simplified
configuration.

I'll commit to trunk soon.  Thanks!

-- 
Mike Rylander
 | VP, Research and Design
 | Equinox Software, Inc. / The Evergreen Experts
 | phone:  1-877-OPEN-ILS (673-6457)
 | email:  [email protected]
 | web:  http://www.esilibrary.com

Reply via email to