Pdf Merge issue in itextsharp - PDF looks distorted after Merge - pdf

Question

I have a simple scenario where I extract pages from a PDF document (or split the document in two parts, if you will) and merge the parts back to a new document, with an option to add new pages in between.
However, in one particular case the resulting document differs from the original one in that couple of pages (in this case pages 4 and 5) look distorted in comparison to the source document.
How can I circumvent the distortion of the pages? The reproduction code below has been tested with iTextSharp versions 5.5.0.0 and 5.5.6.0 (latest at the moment).
You can find the input-File i used here.
void Main()
{
var pathPrefix = #"C:\temp"; // TODO change
var inputDocPath = #"input.pdf";
var part1 = ExtractPages(Path.Combine(pathPrefix, inputDocPath), 1, 2);
var outputPath1 = Path.Combine(pathPrefix, "part1.pdf");
File.WriteAllBytes(outputPath1, part1);
var part2 = ExtractPages(Path.Combine(pathPrefix, inputDocPath), 3);
var outputPath2 = Path.Combine(pathPrefix, "part2.pdf");
File.WriteAllBytes(outputPath2, part2);
var merged = Merge(new[] {
outputPath1,
outputPath2
});
var mergedPath = Path.Combine(pathPrefix, "output.pdf");
File.WriteAllBytes(mergedPath, merged);
}
//Page sizes:
// input: 8,26x11,68; 8,26x11,69; 8,26x11,69; 8,26x11,69; 8,26x11,69; 8,26x11,68; 8,26x11,68
// output: 8,26x11,68; 8,26x11,69; 8,26x11,69; 8,26x11,69; 8,26x11,69; 8,26x11,68; 8,26x11,68
public static byte[] Merge(string[] documentPaths)
{
byte[] mergedDocument;
using (MemoryStream memoryStream = new MemoryStream())
using (Document document = new Document())
{
PdfSmartCopy pdfSmartCopy = new PdfSmartCopy(document, memoryStream);
document.Open();
foreach (var docPath in documentPaths)
{
PdfReader reader = new PdfReader(docPath);
try
{
reader.ConsolidateNamedDestinations();
var numberOfPages = reader.NumberOfPages;
for (int page = 0; page < numberOfPages;)
{
PdfImportedPage pdfImportedPage = pdfSmartCopy.GetImportedPage(reader, ++page);
pdfSmartCopy.AddPage(pdfImportedPage);
}
}
finally
{
reader.Close();
}
}
document.Close();
mergedDocument = memoryStream.ToArray();
}
return mergedDocument;
}
public static byte[] ExtractPages(string pdfDocument, int startPage, int? endPage = null)
{
var reader = new PdfReader(pdfDocument);
var numberOfPages = reader.NumberOfPages;
var endPageResolved = endPage.HasValue ? endPage.Value : numberOfPages;
if (startPage > numberOfPages || endPageResolved > numberOfPages)
string.Format("Error: page indices ({0}, {1}) out of bounds. Document has {2} pages.",
startPage, endPageResolved, numberOfPages).Dump();
byte[] outputDocument;
using (var doc = new Document()) // NOTE use reader.GetPageSizeWithRotation(startPage) ?
using (var msOut = new MemoryStream())
{
var pdfCopyProvider = new PdfCopy(doc, msOut);
doc.Open();
for (var i = startPage; i <= endPageResolved; i++)
{
var page = pdfCopyProvider.GetImportedPage(reader, i);
pdfCopyProvider.AddPage(page);
}
doc.Close();
reader.Close();
outputDocument = msOut.ToArray();
}
return outputDocument;
}

Answer 1

I could reproduce the issue using your code and your test file with iTextSharp 5.5.6. Actually, though, the images are not merely distorted, they have been replaced by other ones! Inspecting the result PDF internally, one observes:
Originally page 3 through 5 each had their own respective Resource dictionary containing different entries than the ones of each other.
After split up, as pages 1 through 3 of part2.pdf, they still had different Resource dictionaries.
In the final merged result, though, page 3 through 5 all refer to the same Resource dictionary object, a copy of the resources of the original page 3!
(As page 3 contains images with the same names as the images on pages 4 and 5, this results in page 3 images being shown on pages 4 and 5.)
Somehow PdfSmartCopy seems to outsmart itself here, using PdfCopy instead creates the expected result.
I assume PdfSmartCopy falsely considers those source dictionaries identical, probably some hash collision without actual equality check.
It might be of interest to note that an equivalent test using Java and iText, SmartMerging.java, does not show the same issue, its result is as expected.
Thus, this looks like an issue of the iTextSharp port or .Net in general.