stamps are not rendered correctly

[SHartmann-Roesberg]

PDF file:
Stamp_Problem.pdf

Configuration:

• Web browser and its version: current Firefox, Chrome
• Operating system and its version: Windows 10
• PDF.js version: current (2.10.377), using viewer from demo site

Steps to reproduce the problem:
Open the attached pdf in a viewer.
If you open the pdf with Acrobat or with integrated Chrome PDF Viewer etc. it looks like in screenshot 1.
If you open it with pdf-js in Chrome, Firefox, Edge or use the integrated PDFViewer of Firefox it looks like screenshot 2.

expected behavior:

The stamps are rendered on top of the table and shown correctly

What went wrong:

the stamps are only visible on top of the table lines

[Snuffleupagus]

Interestingly, if you skip the regular operatorList[1] and only render the Annotations this actually works correctly!
Hence it seems that we're somehow not correctly resetting the state after the "normal" operatorList and before the Annotation-operatorLists are handled (in the src/display/canvas.js file).

---

[1] By replacing all of

pdf.js/src/core/document.js

Lines 346 to 355 in 2822ba4

	return partialEvaluator
	.getOperatorList({
	stream: contentStream,
	task,
	resources: this.resources,
	operatorList: opList,
	})
	.then(function () {
	return opList;
	});

with a return opList; line.

[calixteman]

I think we can easily fix this in using the hasOwnCanvas flag.

[Snuffleupagus]

I think we can easily fix this in using the hasOwnCanvas flag.

That'd seem like a work-around, rather than a "real" fix, and it seems quite important (to me at least) that we figure out why it happens since this does suggest something being wrong w.r.t. resetting the canvas-state before rendering Annotations.

Note that these documents should obviously render correctly even if no separate AnnotationLayer is actually being used.

[calixteman]

The last commands on the page are:

0 G 0 J 0 j 1.44 w 10 M []0 d 411.84 149.76 1.68 630.96 re W n
412.8 780.48 m 412.8 149.76 l S

So the clip region is very small (width = 1.68) and then we draw the annotations.
So drawing in a new canvas in using the hasOwnCanvas stuff is helping to have no clipping path: it's why it works.

I think that a correct fix would be to restore (in beginAnnotations) the context saved when we start to draw the page in order to have something clean before starting drawing the annotations.
I suppose it doesn't really make sense to use something which has been set during the page drawing (before the annotations rendering).
@Snuffleupagus, wdyt ?