[
https://issues.apache.org/jira/browse/TIKA-1644?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=14569061#comment-14569061
]
Nick Burch commented on TIKA-1644:
----------------------------------
Some of the WMF to PDF changes actually seem to be correct. For example, the
two embedded images in 000/000913.doc do seem to be PDFs with a non-standard
WMF header on the front. Ditto for some PICT to PDF, eg image6.pict from
533/533017.doc
Some of the other changes might actually be a bug in the eval code. If the
parsing fails for some reason on the file (eg 186/186738.ppt), then we're
getting entries with a null EMBEDDED_FILE_PATH when some embedded resources
were actually extracted and mimetype detected. With the extra word list
exceptions, I think quite a few are false positives from that. Any chance you
could see if we could exclude / fix for that case?
> Mime type diffs between 1.8 and 1.9-rc1
> ---------------------------------------
>
> Key: TIKA-1644
> URL: https://issues.apache.org/jira/browse/TIKA-1644
> Project: Tika
> Issue Type: Bug
> Reporter: Tim Allison
> Attachments: mime_diffs.xlsx
>
>
> When running 1.9-rc1 against govdocs1, I found a few files whose mime-types
> have changed. I'm posting this now so that others can look...some of these
> are for the better, and some not.
> For further investigation:
> * embedded pict and wmf are now sometimes identified as pdf (TIKA-1085)
> * several .doc files are now identified as application/x-msmetafile and no
> text is being extracted
> * several .doc files are now identified as jpeg or png and no text is being
> extracted
> * several .ppt files which were being identified as various (jpeg, ppt, png,
> msoffice, word) are now being detected as excel
> Probably for the good:
> * a handful of files that were identified as text are now identified as pdf
> (TIKA-1085)
--
This message was sent by Atlassian JIRA
(v6.3.4#6332)
