Illustrates Pixmap regression introduced in PyMuPDF 1.14.13

keep_proportion.py

import fitz
import io
import requests

PDF_URL = "https://github.com/pymupdf/PyMuPDF/raw/master/examples/colordbHSV.pdf"
PNG_URL = "https://www.onlygfx.com/wp-content/uploads/2018/07/12-grunge-brush-stroke-banner-11.png"


def handle(_):
    pdf = requests.get(PDF_URL).content
    png = requests.get(PNG_URL).content
    rect = fitz.Rect(0, 0, 400, 50)
    document = fitz.open(stream=pdf, filetype="pdf")

    for page in document:
        page.insertImage(rect, pixmap=fitz.Pixmap(png), keep_proportion=True)

    return document.write()


if __name__ == "__main__":
    result = handle(None)
    with open("result.pdf", "wb") as file:
file.write(result)

hm, works fine on my Windows, but segfaults on Linux ...

Jpegs also seem to always work on Linux.
I'll try a PNG on Linux now ...

The following also works fine on Windows and Linux. The PNG is a transparent image. Using bytes instead of BytesIO also works in either case.
Something special must be hidden in your case ...

import fitz
import io

pdffile = open("showPDFpage.pdf", "rb")
pdfstream = io.BytesIO(pdffile.read())  # or pdffile.read()
doc = fitz.open("pdf", pdfstream)
page = doc[0]

imgfile = open("test.png", "rb")
imgstream = imgfile.read()  # or io.BytesIO(imgfile.read())

rect = fitz.Rect(100, 400, 300, 700)
page.insertImage(rect, stream=imgstream)
page.drawRect(rect)
doc.save("test-add.pdf")

pdffile.close()
imgfile.close()

Your example works always, if you take stream instead of pixmap.

(Not to mention, that the pixmap should be created only once, outside the loop for every page ...)

So there is a problem with specifically pixmap now ...

Segfaults on macOS Mojave (10.14.3), too. Stream and filename work.

Reverting to pip install pymupdf==1.14.12 works fine with pixmap, so there is an error introduced in 1.14.13?