[ https://issues.apache.org/jira/browse/TIKA-879?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=14257888#comment-14257888 ]
Nick Burch commented on TIKA-879: --------------------------------- I've done something a little different in r1647721 as a partial workaround - I've added a new mimetype of text/x-tika-text-based-message which is a parent of the 3 text-based message/ mimetypes (most of the message/ mimetypes are not text based). With that in place, Vladimir's test email is now correctly detected when it has a .eml extension. (I think that this parent probably is more semantically meaningful than text/plain, which is why I went for it) However, this doesn't solve the "detection without filename" issue, and some other related mail detection problems (eg multipart/signed). We might therefore want to think about adding a new mail detector, along the lines of the one suggested in this Stack Overflow question from a few weeks back - http://stackoverflow.com/questions/27397807/tika-detect-multipart-signed > Detection problem: message/rfc822 file is detected as text/plain. > ----------------------------------------------------------------- > > Key: TIKA-879 > URL: https://issues.apache.org/jira/browse/TIKA-879 > Project: Tika > Issue Type: Bug > Components: metadata, mime > Affects Versions: 1.0, 1.1, 1.2 > Environment: linux 3.2.9 > oracle jdk7, openjdk7, sun jdk6 > Reporter: Konstantin Gribov > Assignee: Nick Burch > Attachments: TIKA-879-thunderbird.eml > > > When using {{DefaultDetector}} mime type for {{.eml}} files is different (you > can test it on {{testRFC822}} and {{testRFC822_base64}} in > {{tika-parsers/src/test/resources/test-documents/}}). > Main reason for such behavior is that only magic detector is really works for > such files. Even if you set {{CONTENT_TYPE}} in metadata or some {{.eml}} > file name in {{RESOURCE_NAME_KEY}}. > As I found {{MediaTypeRegistry.isSpecializationOf("message/rfc822", > "text/plain")}} returns {{false}}, so detection by {{MimeTypes.detect(...)}} > works only by magic. -- This message was sent by Atlassian JIRA (v6.3.4#6332)