Allow override mapping mime<-->parsers through config
-----------------------------------------------------
Key: TIKA-527
URL: https://issues.apache.org/jira/browse/TIKA-527
Project: Tika
Issue Type: Improvement
Components: config
Affects Versions: 0.7
Reporter: Jan Høydahl
h2. Background
As of Tika 0.7, tika-config.xml is not longer mandatory and loading 3rd party
parsers as plugins through service architecture is supported.
This introduces great flexibility, and even allows for extending Tika's file
format support by simply dropping in jar's on the classpath. This is great for
configuring Tika when it's embedded as part of another application such as Solr
or Nutch. You can easily add support for e.g. a commercial document filter with
Tika wrapper without changing Tika or the consuming application, or even
maintaining a tika-config.xml.
This serves the majority of all use cases.
h2. Problem
However, as the variety of 3rd party document parsers increases, we'll start
seeing an overlap of parsers supporting the same mime-types. A very likely
scenario is a company specialized in document filters packaging their parsers
as a Tika plugin, under whatever license they choose.
In this scenario, a system integrator (working with e.g. Solr) wants to gather
all the parsers that the particular customer needs, and then choose which
parser should handle each mime-type. She may want to let a 3rd party parser
plugin handle Word files but the Tika supplied POI parser handle Excel.
Today, the last parser plugin that gets loaded by the class-loader happens to
"win" the mime-types it supports. As it is not uncommon for one parser to
register multiple mime-types, re-claiming a subset of the types is not possible
unless you are consuming Tika directly.
.h2 Solution
Allow for an "override" style mime-to-parser mapping by configuration. To keep
the number of config files down, this is probably best done as an extension of
the tika-config.xml syntax, allowing for specifying only the *changes*, without
repeating all the mappings. Tika should look for tika-config.xml on class-path
by default even if it's not bundled by default.
Say we add a parser plugin which supports a bunch of Office formats, but their
Excel parser sucks. We want to explicitly give control over
application/vnd.ms-excel to the POI parser. Here's how that could be done by
adding support for an "append" attribute on the <parsers> and <parser> tags,
instead of repeating all mime types:
{code:xml}
<properties>
<parsers append="true">
<parser name="parse-office"
class="org.apache.tika.parser.microsoft.OfficeParser" append="true">
<mime>application/vnd.ms-excel</mime>
<mime>application/vnd.ms-excel.sheet.binary.macroenabled.12</mime>
</parser>
</parsers>
<properties>
{code}
When Tika sees append="true", it will first initialize everything as default,
and then re-do the mappings explicitly specified. If you want to remove support
for a parser by config, you could specify a <parser...> tag without an append
attribute and with no <mime> sub-tags specified.
--
This message is automatically generated by JIRA.
-
You can reply to this email to add a comment to the issue online.
