mergeRotatedTranslatedPage doubt

[josejachuf]

Hello everyone

I can't understand what the process is like when applying page1.mergeRotatedTranslatedPage (page2, ...)

Assuming the two pages of equal size

1. Given page1 portrait (rotated 90 °) and page2 landscape

• Are they considered aligned bottom-left?
• What is the rotation point?
• Once rotated, it is transferred by tx and ty?
• Does all of the above apply to page2?

The following works well:

page1.mergeRotatedTranslatedPage (
    page2,
    rotation = 90,
    tx = page1.mediaBox.getWidth () / 2,
    ty = page1.mediaBox.getWidth () / 2,
    expand = True
    )

2. 1. Given page1 portrait (rotated 270 °) and page2 landscape

This works fine

page1.mergeRotatedTranslatedPage (
    page2,
    rotation = 270,
    tx = page1.mediaBox.getHeight () / 2,
    ty = page1.mediaBox.getHeight () / 2,
    expand = True
    )

3. The problem I am having with page1 portrait (rotated 270 °) with cropBox! = MediaBox

pag1.mergeRotatedTranslatedPage (
    page2,
    rotation = 270,
    tx = page1.mediaBox.getHeight () / 2 + page1.cropBox [0],
    ty = page1.mediaBox.getHeight () / 2 + page1.cropBox [1],
    expand = True
    )

more or less it comes out as I want but not as well as I wish

Could someone please explain this to me?
Thank you

[MartinThoma]

more or less it comes out as I want but not as well as I wish

What do you mean by that?

[Atlasfreak]

I too have some problems understanding how excatcly this function works. When I am trying to center a landscape page which I rotated to portrait on a bigger page I am absolutley unable to do it as I have no idea what way I need to translate it. I tried to understand the translation matrix that is applied but I was unable to grasp the concept fully.
Additionally I noticed that when the rotation is positive and not 180 degrees you need to use the page height instead of page width but when the rotation is exactly 180 degrees you need to use page width and height with tx and ty respectively.

[josejachuf]

I do not know the origin of the PDF that you use, per do you make a recommendation, use the gs command to clean inconsistent info containing the PDF before using py-pdf2

gs -o output.pdf -sDEVICE=pdfwrite -dPDFSETTINGS=/screen input.pdf

[Atlasfreak]

Test.pdf
Just rotate it and then try to center it on a page that is bigger than A4 while it is possible it is not obvious as it differs from the way you do it with none rotated pages.
With rotated pages when you change tx or ty the page is moved diagonally and not along their respective axis except when it is rotated 180 degrees.
I think this should at least be documented somewhere so people who do not know how transformation matrices work (like me) know how it works and don't need to try arbitrarily until they find something that works.
Side note I found solution for my initial problem.

[MartinThoma]

The rotation with mergeRotatedTranslatedPage is deprecated now. Please use the following pattern instead (which should give the same results):

from PyPDF2 import Transformation, PdfReader, PdfWriter

reader = PdfReader("box.pdf")
page = reader.pages[0]

x1, y1, x2, y2 = page.cropbox
assert x1 == 0
assert y1 == 0

op = Transformation().rotate(rotation=180).translate(tx=x2, ty=y2)
page.add_transformation(op)

print(page.cropbox)
print(page.mediabox)

writer = PdfWriter()
writer.add_page(page)
with open("output.pdf", "wb") as fp:
    writer.write(fp)

The center point of the coordinates is in the bottom-left corner. That means if you want to flip the image by 180 degrees, the content would go from the upper-right quadrant to the lower-left one.

You can see that those are just simple coordinate transformations. PyPDF2 does not care for this operation where the actual content is. It also doesn't care about the mediabox / cropbox / trimbox. It just applies the transformation to the coordinate system.

Did that clarify the topic?

[carecavoador]

Hi Martin. First, I want to say I really appreciate your effort in maintaining this. I really like PyPDF2 and I try to use it whenever I can.

I'm having issues with this add_transformation method. I took the time to do a minimal example to demonstrate what is going on. The page objects are being translated and rotated, but the cropbox is not.

[MartinThoma]

Wow, thank you so much for putting that much effort into investigating the issue! I'll have a look tomorrow (it's late in my timezone and this day was super busy)

[MartinThoma]

I thought about creating two more convenience methods for Page objects:

• trim(box): Fit a page box (by default box="trimbox", but you could specify "mediabox" or "cropbox" as well) to the actual content.
• normalize(): Move the content to the upper-right quadrant, e.g. making sure that there are no negative coordinates + making sure that the lowest x/y coordinates are actually 0.

Would that be useful? Do you have an opinion on the naming of those functions? Which media-

[MartinThoma]

I'm sorry, this is not as easy as I thought it would be. I will not be able to work on this feature this year. But maybe somebody wants to add a PR?

[zain910128]

Hey Martin,
Thanks for the letting us know.
Can you also provide details on what challenges you've experienced or you anticipate ?

I hope the details will give others an idea of what is involved in case anybody wants to attempt it.

[MartinThoma]

You have to (1) get the position and (2) re-compute the position of every single object.

Even if you go with the simpler solution that you just do it via the bounding boxes, you still have to get every single position. It's just too easy to forget something.

[pubpub-zz]

Just a (stupid?) question:
What is the advantage of "normalizing". the relative positions of objects are still valid and the merge_transformed_page() and other functions seem to do the job. no ?

[MartinThoma]

Not a stupid question at all :-)

When adding an overlay ( https://pypdf.readthedocs.io/en/latest/user/add-watermark.html ) it is not so easy to say how the final result will look like.

For example, say you want to add the watermark in the bottom right corner. If the original images mediabox is (0, 0) to (page_width, page_height) and the watermarks cropbox is (0, 0) to (watermark_width, watermark_height), then you just need to shift the watermark by +(page_width - watermark_width, page_height - watermark_height).

Things are a bit more complicated with the image is not centered. Still doable, though.

[Cyberware-tech]

The "minimal example" from earlier in this thread seems to be a broken link.
Can someone please specify a similar example in a more permanent form so that the issue is clear ?

[pubpub-zz]

you should find examples in https://pypdf.readthedocs.io

[zain910128]

Hey

I actually meant the "minimal example" that @carecavoador posted in the linked post as a description of the bug. The link to that example is broken.

[pubpub-zz]

This example deals with PyPDF2 which is obsolete. the core team will not support any discussion about it