We are currently using a shim (https://github.com/tballison/hadoop-safe-tika because of binary conflicts in commons-io versions between what Hadoop supports and the more modern features that Apache Tika and Apache POI were using in commons-io.

For now, all you have to do is update the fat jar dependencies:

1. tika-core-shaded in ivy/ivy.xml

2. tika-parsers-standard-package-shaded in src/plugin/parse-tika/ivy.xml

3. The library name version for tika-parsers-standard-package-shaded in src/plugin/parse-tika/plugin.xml

4. Repeat steps 2 and 3 for the language-identifier

5. Build Nutch and run all unit tests:

   $ cd ../../../ $ ant clean runtime test

The following directions are what we used to do with thin jars. Hopefully, we'll be able to get back to these directions once we have version harmony with Hadoop and Tika/POI.

1. Upgrade Tika dependency (tika-core) in ivy/ivy.xml

2. Upgrade Tika dependency in src/plugin/parse-tika/ivy.xml

3. Upgrade Tika's own dependencies in src/plugin/parse-tika/plugin.xml

   To get the list of dependencies and their versions execute: $ cd src/plugin/parse-tika/ $ ant -f ./build-ivy.xml $ ls lib | sed ‘s/^/ <library name="/g’ | sed ‘s/$/"/>/g’

   In the plugin.xml replace all lines between

   and

   with the output of the command above.

4. (Optionally) remove overlapping dependencies between parse-tika and Nutch core dependencies:

   • check for libs present both in build/lib and build/plugins/parse-tika/ (eventually with different versions)
   • duplicated libs can be added to the exclusions of transitive dependencies in build/plugins/parse-tika/ivy.xml
   • but the library versions in ivy/ivy.xml MUST correspond to those required by Tika

5. Remove the locally “installed” dependencies in src/plugin/parse-tika/lib/:

   $ rm -rf lib/

6. Repeat steps 2-5 for the language-identifier plugin which also depends on Tika modules

   $ cd ../language-identifier/

7. Build Nutch and run all unit tests:

   $ cd ../../../ $ ant clean runtime test