[
https://issues.apache.org/jira/browse/JCR-3667?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=13781230#comment-13781230
]
frank commented on JCR-3667:
----------------------------
In the issue: JCR-3476, it says
This is unnecessarily expensive because the binary is fetched. Also an empty
field is added to the index document.
While Tika v1.4 solves a issue : TIKA-1133
In some cases it is beneficial to allow empty and duplicate Tika metadata
values for multi-valued XML elements like RDF bags.
Current Jackrabbit depends on Tika v1.3, how about upgrading Tika to version
1.4 and check to see whether this problem solved or not?
> Possible regression with accepted content types when extracting and indexing
> binary values
> ------------------------------------------------------------------------------------------
>
> Key: JCR-3667
> URL: https://issues.apache.org/jira/browse/JCR-3667
> Project: Jackrabbit Content Repository
> Issue Type: Bug
> Affects Versions: 2.4.4
> Reporter: Cédric Damioli
> Fix For: 2.4.5, 2.6.4, 2.7.2
>
>
> JCR-3476 introduced a mime-type test before parsing binary values, based on
> Tika's supported parsers.
> This may lead to incorrect behaviours, with a "text/xml" not being extracted
> and indexed because the XMLParser does not declare "text/xml" as a supported
> type.
> The problem here is that there is a regression between 2.4.3 and 2.4.4,
> because the same content was previously well recognized by Tika's Detector
> and then extracted.
> Furthermore, it seems to me inconsistent on one hand to rely on the declared
> content type and on the other hand to delegate the actual type detection to
> Tika ?
> This may lead to cases where the jcr:mimeType value is set to eg.
> "application/pdf" but detected and parsed by Tika as "text/plain" with no
> error.
--
This message was sent by Atlassian JIRA
(v6.1#6144)
