Details
-
Bug
-
Status: Resolved
-
Minor
-
Resolution: Fixed
-
1.26
-
None
-
None
Description
I've encountered some Word files that have Ole10Native embeddeds which Tika extracts with strange filenames. It looks like the attachments were originally named with Chinese & Unicode characters, and the filename that Tika is giving is a cp1252 interpretation of the original UTF-8-encoded filename.
Looking closer at the Ole10Native stream of these files, it does seem like there is a UTF-8 version of the filename stored, as well as a UTF-16 version of the filename stored later on after the actual attachment data. I believe POI is returning this first UTF-8 version of the filename interpreted as if it were ANSI / cp1252.
A possible solution would for Apache POI to read and return the provided UTF-16 filename if it is present. Alternatively, Tika could check the currently returned "ANSI" name to see if it might actually be valid UTF-8.
Attached is an sample file I made which has a .msg file with name "約翰的測試文件🖖.msg" embedded in a .docx file. Tika currently extracts the attachment with filename "ç´ç¿°ç測試æ件ð.msg"
–
Regarding the Ole10Native data stream, I can't find any official documentation for its structure, but these extra three UTF-16 string properties I'm seeing at the end look to follow the following format:
- The strings are not null terminated, but instead are proceeded by a 4-byte string length value. Note that this value is the number of 16-bit code units in the UTF-16 string and not the byte length.
- The order of the 3 strings is temporary path, filename, original path. This differs from the order of the normal ANSI / UTF-8 strings near the beginning of the Ole10Native stream which is filename, original path, temporary path.
- I'm assuming these wide variants of these strings are optional and may not be present.