Document select keeps the pdf with the same size.

[caiocesarrm]

Hi!
I have a pdf (a book with 373 pages). I select one page using doc.select([75]) and save.

doc = fitz.open(f'{base_path}[Chollet]-Deep_Learning_with_Python.pdf')
doc.select([75])
doc.save(f'{base_path}page.pdf', clean=True, garbage=4, deflate=True)

the page saved: page.pdf

After i save the file, the pdf have almost the same size, having only one page.
i believe that this is because the content is still there. Any thoughts on how can i solve this?

PyMuPDF 1.18.6.
Built for Python 3.6 on linux (64-bit).

[JorjMcKie]

There may be a number of reasons for this - even apart from a potential PyMuPDF bug 😉.

Garbage collection can only remove objects which are no longer referenced by anything whatsoever ... and that may be all sorts of this.
Other possible explanations are potentially lots of / large embedded files.
Also, that single page may still use large fonts for its text or show big images ...
The important thing to notice about select: apart from removing pages, the PDF is not changed otherwise.

Here are a few steps to get closer to an explanation:

1. Before selecting, do a page.clean_contents(sanitize=True) (or cleanContents(sanitize=True) for an older version).
   • cleaning removes any unneeded references from the page definition ... many PDFs are created in that unfortunate way ...
2. Do that page cleaning before saving.
3. Make a new PDF and insert that one page into it:

new = fitz.open()
new.insert_pdf(doc, from_page=75, to_page=75)  # or insertPDF(...)
new.save(...)

[JorjMcKie]

Found the reason for this unexspected size behaviour:
The page contains a /B entry

B array (Optional; PDF 1.1; recommended if the page contains article beads) An ar-
ray of indirect references to article beads appearing on the page (see Sec-
tion 8.3.2, “Articles”; see also implementation note 37 in Appendix H).
The beads are listed in the array in natural reading order.

In this case it references annotations, which in turn reference other pages, and yet other pages, etc.
Result: the full PDF is still there - only the page pointers have been reduced to that one page.

So you have the option 3 from previous post - which gives you below 70 KB PDF size.
OR with functions new in v1.18.7, you can - half-manually - remove (= set to PDF null) that entry:

>>> import fitz
>>> doc=fitz.open("page.pdf")
>>> page=doc[0]
>>> print(doc.xref_get_keys(page.xref))
('Annots', 'B', 'Contents', 'CropBox', 'MediaBox', 'Parent', 'Resources', 'Rotate', 'Type')
>>> doc.xref_set_key(page.xref, "B", "null")
>>> page.clean_contents()
>>> doc.save("x.pdf",garbage=4,deflate=True)

Which results in a PDF of 66 KB.

Semi-manual might also mean

for key in doc.xref_get_keys(page.xref):
    if key not in ('Annots', 'Contents', 'CropBox', 'MediaBox', 'Parent', 'Resources', 'Rotate', 'Type'):
        doc.xref_set_key(page.xref, key, "null")

So this would only keep PDF keys also included by the doc.insert_pdf method.