MiniMagick::Image#mime_type outputting wrong value for PDFs with IM 7.0.8.64

[adamcreekroad]

I'm not sure if this is a bug here that is just now manifesting, or a bug in IM (or both?). According to their changelog, it looks like they somewhat recently revised the PDF reader.

Given a PDF exported by Adobe Illustrator, the identify command now outputs
Mt Moon.pdf AI 810x1242 810x1242+0+0 16-bit ColorSeparation CMYK 3.83776MiB 0.010u 0:00.010

What it used to output is
Mt Moon.pdf PDF 810x1242 810x1242+0+0 16-bit ColorSeparation CMYK 3.83776MiB 0.400u 0:00.389

With -verbose added on, they both give the correct mime-type:
Image: Mt Moon.pdf Format: AI (Adobe Illustrator CS2) Mime type: application/pdf

Image: Mt Moon.pdf Format: PDF (Portable Document Format) Mime type: application/pdf

The mime_type method in MiniMagick doesn't actually use that value, it takes the format (AI or PDF) and just adds it onto image/.

def mime_type
  "image/#{self["format"].downcase}"
end

So what we now get is image/ai instead of image/pdf. Even image/pdf isn't technically right, it should be application/pdf

An easy way to get the proper value

def mime_type
  self["data"]["mimeType"]
end

This would of course be breaking to everyone expecting the old behavior of outputting image/pdf.

[janko]

That was a poor decision to build the MIME type this way (by prepending image/), as ImageMagick can work with lots of formats that don't have image/* MIME types, like PDFs that you mentioned.

It should get changed to output correct values, maybe with a new major release then.

[hwo411]

The same problem is with svg, actually.

It says image/svg while it should be image/svg+xml.

The solution from the original method indeed works. Would be great to have this fixed.

[janko]

This cannot be fixed in a backwards compatible way, as

def mime_type
  self["data"]["mimeType"]
end

will be significantly slower, because the whole image has to be read, unlike with current implementation where only the magic header needs to be read.

I don't think ImageMagick is sufficiently good for determining MIME types, there are Marcel and MimeMagic that are able to do this by just reading the header file. So, I will be deprecating MiniMagick::Image#mime_type.