Proper exif metadata formatting

[karlch]

During #212 we came to the conclusion that proper formatting of the exif metadata is non-trivial. Piexif seems to have no option to give nice correctly formatted information, only the raw data.

We have a few options here:

• Try to get this functionality into piexif. This would be awesome as others could benifit from the implementation, but certainly not guaranteed that upstream wants to support it. Maybe worth trying though.
• Implement a wrapper ourselves and extend this on-demand. We could have a dictionary mapping exif keys to their formatted version, e.g.:

  formatters = {
  piexif.ExifIFD: {
      piexif.ExifIFD.ExposureTime: lambda value: f"{value:d} s",
  }
  }

• Switch to another exif tool. This is not my favourite option but could be a last resort. We would probably have to duplicate the current code with both exif dependencies and deprecate piexif for a while. It could be removed for v1.0.0. For this we would need to find a much better yet still not too heavy alternative that is really worth it.

Any ideas and suggestions are very welcome!

[karlch]

Having some simple proof-of-concept and at least the values of the default keys in the different metadata key-sets properly format before releasing v0.8.0 would make a lot of sense to me. I will probably take a shot at implementing this directly in vimiv and ask you @jcjgraf for some help once the basics are there. Does this sound like a plan to you?

[jcjgraf]

I think it definitely makes sense to get the formatting right for the next release.

Doing this formatting ourselves, based on the raw data provided by piexif seems a lot of work do me. While it may be reasonable to to it for the most frequently used metadata, it still very tedious. Especially, since the the formatter has to be extended whenever someone needs a new certain new metadata tag.

So my question is, does it really make sense to urgently stick to piexif instead of switching to a more elaborated package? exiftool as well as exiv2 are both amazing tools. While they are not written in python themselves, there exist python wrappers for both.

Obviously both provide formatted metadata output by default (besides many other advantages).

Extracting metadata using PyExifTool (wrapper for exiftool) is quite simple

import exiftool

image = "test.CR2"

with exiftool.ExifTool() as et:
    metadata = et.get_metadata(image)

    print(metadata)

metadata is in this case a dict with all metadata found in the image. The metadata tag is the key and formatted metadata is the value.

In order to extract only certain tags, one can use et.get_tags(["firstTag, secondTag"], image). which returns a similarly structured dict.

While exiv2 seems a bit less powerful that exiftool, its wrapper py3exiv2 works similarly to the example above.

In my view, switching to such an elaborated package is a way cleaner solution that bloading vimiv with formatting code. Also, piexifs documentation does not even state what image types are supported (CR2 certainly works, but I have no clue what other formats are supported). Also, the mentioned tools keep receiving updates for newly released cameras/lenses etc.

Concerning possible disadvantages, the only ones I can imagine is heaving two dependencies instead of one (the actual tool + the wrapper), and a larger size of the dependencies (12MB for exiftool and 9MB for exiv2).

Anyways, let me know what you think about it.

And whatever solution we decide on, I propose to into this issue. This way you can deal with #293 and everything else required for v0.8.0.

[karlch]

To me this currently seems like a lose-lose situation. Let me summarize the pros and cons of the different tools in my opinion. Disclaimer: I haven't used any of them extensively.

piexif

• + pure python
• + package available in arch linux community repository
• - limited options, especially no formatting

PyExifTool

• + very powerful, especially supports formatting
• - wrapper around a command line tool
• - python package only available in the AUR, AUR package does not build for me and depends on half of the python universe for building

pyexiv2

• + powerful, especially supports formatting
• - wrapper around a C++ library, still better than wrapping the command line
• - python package only available in the AUR

Just from this I would rule out using PyExifTool. This leaves us with piexif and pyexiv2.

If you feel like the benefits from using a wrapper for exiv2 are large, feel free to go that route in a PR. We would still need to support piexif for at least one release with a deprecation warning to allow users to prepare for a transition though. For this, it might make sense to decouple the usage even more from the underlying library in an additional module, i.e. imutils.exif_wrapper. This will most certainly be quite an ugly effort, especially with our lazy loading and the optional features related to exif.

[jcjgraf]

Now, after #311 has been merged, this issue can be closed I guess. Is there anything else which needs to be done for v0.8.0?

[karlch]

Yes, makes sense! I have started working on #220 as it makes a lot of sense to have this with the new metadata flexibility, I think. I will probably ask you to take a quick look before merging, but this should be straightforward compared to #311. After this, v0.8.0 should be good to go :tada: