PdfReader | Fast load and parse of Pdf format in C | Document Editor library

With PDFsharp and MigraDoc this optimization is done automatically if you use them as intended.

Load the image once with PDFsharp and add it to as many pages as you like, there will be only one copy of the image in the document.

In createAndAddPage you import the page into the wrong PdfWriter:

As already mentioned in a comment, the example document "InvalidDocumentSignedTwice.pdf" has the signature not applied in an incremental update, so here it is obvious that former signatures will break. But this is not the issue of the OP's example project. Thus, the issue is processed with an eye on the actual outputs of the example project.

When validating signed PDFs Adobe Acrobat executes two types of checks:

• It checks the signature itself and whether the revision of the PDF it covers is untouched.
• (If there are additions to the PDF after the revision covered by the signature:) It checks whether changes applied in incremental updates only consist of allowed changes.

The former check is pretty stable and standard but the second one is very whimsical and prone to incorrect negative validation results. Like in your case...

In case of your example document one can simply determine that the first check must positively validate the first signature: The file with only one (valid!) signature constitutes a byte-wise starting piece of the file with two signatures, so nothing can have been broken here.

Thus, the second type of check, the fickle type, must go wrong in the case at hand.

To find out what change one has to analyze the changes done during signing. A helpful fact is that doing the same using iText 5 does not produce the issue; thus, the change that triggered the check must be in what iText 7 does differently than iText 5 here. And the main difference in this context is that iText 7 has a more thorough tagging support than iText 5 and, therefore, also adds a reference to the new signature field to the document structure tree.

This by itself does not yet trigger the whimsical check, though, it only does so here because one outline element refers to the parent structure tree element of the change as its structure element (SE). Apparently Adobe Acrobat considers the change in the associated structure element as a change of the outline link and, therefore, as a (disallowed) change of the behavior of the document revision signed by the first signature.

So is this an iText error (adding entries to the structure tree) or an Adobe Acrobat error (complaining about the additions)? Well, in a tagged PDF (and your PDF has the corresponding Marked entry set to true) the content including annotations and form fields is expected to be tagged. Thus, addition of structure tree entries for the newly added signature field and its appearance not only should be allowed but actually recommended or even required! So this appears to be an error of Adobe Acrobat.

Knowing that this appears to be an Adobe Acrobat bug is all well and good, but at the end of the day one might need a way now to sign such documents multiple times without current Adobe Acrobat calling that invalid.

It is possible to make iText believe there is no structure tree and no need to update a structure tree. This can be done by making the initialization of the document tag structure fail. For this we override the PdfDocument method TryInitTagStructure. As the iText PdfSigner creates its document object internally, we do this in an override of the PdfSigner method InitDocument.

I.e. instead of PdfSigner we use the class MySigner defined like this:

You use page.newContentStreamBefore() to add a content stream for your water mark. Thus, when the page is drawn, the watermark is drawn first and the content thereafter over it. For usual text and sparse vector graphics that is good but for full area covering content the watermark may be totally hidden.

Try to use page.newContentStreamAfter() instead. But beware, you may not want to bold the watermark text then anymore, probably even restrict to outline only, as the mark now covers content.

In a question update and a comment you asked how to

make the text transparent.

Transparency usually is achieved by selecting a matching extended graphics state, e.g.:

You could do this with a for loop. Extract the text from the pages in the loop and append them to a list.

You could do both:

To save your altered PDF to memory in an object that can be passed around (instead of writing to a file), simply create an empty instance of io.BytesIO:

With PDF4NET you can extract the XMP metadata without loading the entire document in memory:

For anyone who needs it, I solved this issue by creating a controller.