Performance drop of MiniMagick::Image.open vs. MiniMagick::Image.new

[tudorpavel]

My use case is that I want to create image previews from the first page of multi-page PDF document uploads.

I've used @davidchristiandy's solution from here.

version :cover do
  process :convert_cover => :png
end

def convert_cover(format)
  manipulate! do |img| # this is ::MiniMagick::Image instance
    img.format(format.to_s.downcase, 0)
    img
  end
end

The problem I noticed is a significant performance drop for large PDFs like books with tens/hundreds of pages which didn't make sense to me. Uploading one such book took about 9 seconds to process. So, I began tracing the problem.

First, I converted the first page directly using imagemagick and it ran in less than 1 second.

$ convert book.pdf[0] test-imagemagick.png

Then, I used MiniMagick directly from a rails console and it also ran in less than 1 second.

img = MiniMagick::Image.new('book.pdf') # takes about 1 second
img.format('png', 0)

Looking through the implementation i found that manipulate! uses MiniMagick::Image.open.

When I ran that in rails console, I had found the performance drop.

img = MiniMagick::Image.open('book.pdf') # takes about 9 seconds

Turns out, the MiniMagick::Image.open method makes a copy of the loaded file instead of manipulating it directly and that's most likely the reason for the performance drop.

My question is, couldn't manipulate! just use MiniMagick::Image.new since the original file is being overridden anyway?

Also, having different versions in the Uploader already makes copies with prefixes such as cover_ before running manipulate!, doesn't it?

There might be other use cases that would be affected by this change, but I'm not sure. If I have more time, I can investigate the implications further.