Is there any way to subset_fonts before inserting fonts?

[izerui]

I can know all the text before I do the page insertion, this will improve the efficiency of subset_fonts, it seems like subset_fonts doesn't support multithreading.
Creating subsets is a time-consuming action when working with documents with many pages

[JorjMcKie]

Threading is not supported at all in PyMuPDF! Please use the multiprocessing module of Python when needed.

Creating font subsets is now directly supported by MuPDF. It still is declared as "experimental", so we haven't yet included it officially.
But you can use it and we do not know of any issues. It is at least 15 times faster than the current method.
Here is what you can do instead doc.subset_fonts():

pdf = fitz._as_pdf_document(doc)  # access underlying PDF-specific level
fitz.mupdf.pdf_subset_fonts2(pdf, list(range(doc.page_count)))

[izerui]

Sorry, I wrote a separate test case and it seems like he took effect, I'll check my program again.

[JorjMcKie]

you have of course to use a save method with all available compression!!!
doc.ez_save(...).

[izerui]

Yes, just now I also found out that if I use save then the size of the save is the same, if it is ez_save then the file size will be much smaller, by the way I would like to ask again: is ez_save the same processing logic as doc.tobytes(garbage=4, deflate=True), thank you very much!

[JorjMcKie]

doc.tobytes(garbage=4, deflate=True, use_objstms=1).
Object stream support (use_objstms) provides for additional compression effects.

[izerui]

Your tips, and improve the compression effect of my final pdf, very ideal, thank you so much, obviously my negligence, not timely attention to the source code parameters of some of the description.

[izerui]

Can I use this code block in multiple threads?
This allows multiple documents to be compressed at the same time!

pdf = fitz._as_pdf_document(doc)  # access underlying PDF-specific level
fitz.mupdf.pdf_subset_fonts2(pdf, list(range(doc.page_count)))

[JorjMcKie]

NO! No multi-threading support!
Put processing different PDF's in different processes using Python multiprocessing module.
And do not forget to use doc.ez_save()!

[izerui]

Thanks, I see what you mean, I use different threads in the thread pool to compress different documents, and I haven't found any conflicting issues for the time being, can I test it for a while and see if it exposes some issues?

docs = [doc0, doc1, doc2....]
with concurrent.futures.ThreadPoolExecutor(max_workers=10) as pool:
        
        pdf = fitz._as_pdf_document(docs[i])  # access underlying PDF-specific level
          fitz.mupdf.pdf_subset_fonts2(pdf, list(range(docs[i].page_count)))

          tmp_dir = tempfile.gettempdir()
          tmp_file_path = os.path.join(tmp_dir, f"{tmp_dir}/{int(time.perf_counter() * 1000)}.pdf")
          doc.ez_save(tmp_file_path)
          doc.close()
          return fitz.open(tmp_file_path)

Forgive me for wanting to try for faster performance.

[JorjMcKie]

You can do what you want of course.

I would definitely use multiprocessing instead. It is safer and also faster than multithreading, because it is the operating system doing the parallelism - not Python. There is an example in the documentation.