[PDFBOX-1622] TextNormalize init not thread-safe, may lead to infinite loop

• Type: Bug
• Status: Closed
• Priority: Major
• Resolution: Fixed
• Affects Version/s: 1.0.0
• Fix Version/s: 1.8.3, 2.0.0
• Component/s: Utilities
• Assignee: Andreas Lehmkühler
• Reporter: Florent Guillaume
• Created: 04/Jun/13 17:12
• Updated: 30/Nov/13 17:02
• Resolved: 09/Aug/13 18:27

Description

TextNormalize fills a static HashMap (DIACHASH) from a method (populateDiacHash) called by the TextNormalize constructor.

If the constructor is called from two different threads at the same time, then the HashMap may be written by two concurrent threads which may and will cause infinite loops.

We see the CPU at 100% and jstack shows 4 threads all stuck at:

"Thread-2" prio=10 tid=0x00007f6e94499000 nid=0x347 runnable [0x00007f6e925d6000]
java.lang.Thread.State: RUNNABLE
at java.util.HashMap.put(HashMap.java:391)
at org.apache.pdfbox.util.TextNormalize.populateDiacHash(TextNormalize.java:82)
at org.apache.pdfbox.util.TextNormalize.<init>(TextNormalize.java:41)
at org.apache.pdfbox.util.PDFTextStripper.<init>(PDFTextStripper.java:193)

A patch to fix this is attached, it just moves the initialization to a static block.

Please apply to the 1.8.3 and 2.0.0 branches.

Issue Links

• relates to: PDFBOX-449 Decomposed extended Latin Characters not normalized