Weird edge detection behavior when the blurred image border pixel extrapolation method changes

Expected behaviour

The canny edge detector should yield very similar images when the blurred image has two different pixel extrapolation methods for the border, since only the border is affected by this.

Actual behaviour

Edge detection is being affected by the selected method. When I select cv.BORDER_REPLICATE I get many more edges than when cv.BORDER_DEFAULT is selected. This does not make sense.

Steps to reproduce

Code running on Windows 11
OpenCV version : 4.10.0
Python version : 3.11.4

CODE :

img = cv.imread('../Photos/park.jpg')
#Gaussian blur
blur1 = cv.GaussianBlur(img , (7,7) , cv.BORDER_REPLICATE)
blur2 = cv.GaussianBlur(img, (7,7) , cv.BORDER_DEFAULT)
cv.imshow('Original' , img)
cv.imshow('Blurred 1' , blur1)
cv.imshow('Blurred 2' , blur2)

#Canny edge detector:
canny1 = cv.Canny(blur1 , 150,210)
cv.imshow('Canny (1) edges w/ 150-210' , canny1)

canny2 = cv.Canny(blur2 , 150 , 210)
cv.imshow('Canny (2) edges w/ 150-210' , canny2)

Start by running the supplied Python reproduction with OpenCV 4.10.0 and compare the GaussianBlur outputs and subsequent Canny results for BORDER_REPLICATE and BORDER_DEFAULT. Trace how the GaussianBlur border argument is handled and determine whether the differing edge results are expected; done means the behavior has a clear resolution and is corrected or documented.