[PDFBOX-2823] StringIndexOutOfBoundsException when doing DateConverter.parseDate()

• Type: Bug
• Status: Closed
• Priority: Major
• Resolution: Fixed
• Affects Version/s: 1.8.9, 1.8.10
• Fix Version/s: 1.8.10
• Component/s: Parsing
• Assignee: Tilman Hausherr
• Reporter: Tilman Hausherr
• Created: 04/Jun/15 16:49
• Updated: 23/Jul/15 06:35
• Resolved: 04/Jun/15 17:12

Description

From Kevin J. in the user mailing list:

We are currently using Apache Solr / Tika to index documents for searching. The exact version that is being used is version 1.8.8 of PDFBox.

We can across a document that produced this stack trace (trimmed to the relevant part of PDFBox):

Caused by: java.lang.StringIndexOutOfBoundsException: String index out of range: 1
        at java.lang.String.charAt(String.java:658)
        at org.apache.pdfbox.util.DateConverter.parseDate(DateConverter.java:679)
        at org.apache.pdfbox.util.DateConverter.toCalendar(DateConverter.java:808)
        at org.apache.pdfbox.util.DateConverter.toCalendar(DateConverter.java:780)
        at org.apache.pdfbox.util.DateConverter.toCalendar(DateConverter.java:753)
        at org.apache.pdfbox.cos.COSDictionary.getDate(COSDictionary.java:849)
        at org.apache.pdfbox.pdmodel.PDDocumentInformation.getCreationDate(PDDocumentInformation.java:212)

Inspection of the document's binary revealed that it contained a creationDate consisting of a single white space (ASCII 0x20), which is probably illegal. I managed to create a small reproduction of the error using like so:

File file = new File("/path/to/document/bad.pdf");
InputStream stream = new FileInputStream(file);
PDFParser parser = new PDFParser(stream);
parser.parse();
PDDocumentInformation info = parser.getPDDocument().getDocumentInformation();
Calendar creationDate = info.getCreationDate();
System.out.println(creationDate.toString());

Which produces the same stack trace. I verified this against the latest build from the site on 1.8.9, and the behavior remains. This looks very similar to PDFBOX-1803, however that issue is marked as resolved in 1.8.5. So, my questions:

• Is the exception an expected behavior? Ideally Tika would just index the document anyway, the creation date isn't important to us. Tika had an issue for this, TIKA-1233, that marks it as fixed by swallowing the exception, but looking at the comments for it, they removed the try/catch in r1593983 since it is marked as fixed here.
• Is this a regression, or slightly different somehow from 1803? Shall I create a new issue or get the existing 1803 re-opened?
• The PDF that reproduces the issue can be downloaded here: https://www.dropbox.com/s/tll5rscrlt95xuc/bad.pdf?dl=0

Issue Links

• is depended upon by: TIKA-1233 PDFBox can throw StringIndexOutOfBoundsException on some dates
• relates to: TIKA-1588 Upgrade to PDFBox 1.8.10 when available