[PDFBOX-5499] Performance issue since 2.0.18

• Type: Bug
• Status: Closed
• Priority: Major
• Resolution: Fixed
• Affects Version/s: 2.0.19
• Fix Version/s: 2.0.27, 3.0.0 PDFBox
• Component/s: PDModel
• Assignee: Tilman Hausherr
• Reporter: Thomas Debray Luyat
• Created: 05/Sep/22 10:48
• Updated: 29/Sep/22 17:57
• Resolved: 20/Sep/22 17:17

Description

Our PDF is parsed in less than 200ms in 2.0.18 and more then 8 seconds in 2.0.19. The same issue is still there in 2.0.26.

 

In version 2.0.19, SmallMap has been introduced. We're facing a performance issue since this modification.

We patch our code to just replace the SmallMap implementation like this:

package org.apache.pdfbox.util;

import java.util.LinkedHashMap;

public class SmallMap<K, V> extends LinkedHashMap<K, V> {
    // nothing : use the standard LinkedHashMap
}

And the performance issue disappear. 

Our test is really simple:

    long start = System.currentTimeMillis();
    try (PDDocument document = PDDocument.load(new File(inFile))) {
      // nothing : only parsing is evaluated
    }
    long duration = System.currentTimeMillis() -start;

    assertTrue(duration < 500);

 

I can understand that the SmallMap can solve issues in some cases, but it is possible to implement a factory to create this map and then allow to setup which Map implementation we want to use?

Issue Links

• relates to: PDFBOX-3284 Big Pdf parsing to text - Out of memory