Memory Issues on GetWords() and crashes with given file

[stephen-williamson]

0020.pdf

I am having a issue with a given PDF, The pdf itself is larger than most that I use pdfPig for. at round 13mb (normally my pdfs are <1mb)
It takes longer than normal to call the GetPage() method (about 5 seconds instead of instant) but it does succeed. While the GetWords() method hangs for a long time (multiple minutes) before eventfully fully crashing.

In that time, memory has shot right up, I end up with 1.5GB GC Heap Size and around 5GiB Allocation Rate looking at the diagnostics session in visual studio.

I cannot even catch the error with a try catch,
Any help would be great, even if it was just to be able to catch the crash nicely. I've attached a snapshot of the memory

var path = @"C:\Users\stephen.williamson\Downloads\0020.pdf";

using (var document = PdfDocument.Open(path))
{
    for (var i = 0; i < document.NumberOfPages; i++)
    {
        var page = document.GetPage(i + 1); //This line takes about 5 seconds
        
        var words = page.GetWords(NearestNeighbourWordExtractor.Instance); //Here it crashes but if i remove the Parameter, it willcrash on the next line instead
        var blocks = DocstrumBoundingBoxes.Instance.GetBlocks(words);
        var orderedBlocks = DefaultReadingOrderDetector.Instance.Get(blocks);

        Console.WriteLine("((TEXT SECTION))");

        foreach (var block in orderedBlocks)
        {
            Console.WriteLine("==BLOCK==");
            Console.WriteLine(block.Text);
           
            // Do something
        }
    }
}

[BobLd]

@stephen-williamson Thanks for sharing the document. The main issue I see with your document is that the page contains about 2 million letters.... NearestNeighbourWordExtractor was not designed to handle that many letters.

Fixing that involves a deep optimisation of the layout analysis algos. The document you provided will be very usefull for benchmarking though

[BobLd]

after further analysis, the letter count can be brought down to 300k by only taking in account the ones that are within the boundary of the page

Related to #681