It doesn't actually do anything, but it does provide a layer for it to start working.

They can be very CPU intensive, and people may not want to suffer that. Having the option to turn them off isn't a bad idea.

This is a different approach to what #35 describes. The same config gives hints as to what should be in the config, however the default config should be a bit safer. For instance, the default config has no storage paths - so nothing can actually be uploaded without a config. This is intentional.

Adds a feature described in #43

Adds #43

Fixes #37

* Use a swap image for resizing the frame (otherwise the frame gets corrupted)
* Use a linear resample filter (better output quality)
* Redraw the frame using a different technique
* Set the bounds of the image on the image config prior to re-encoding (otherwise the image has a bunch of whitespace)

This finishes #28

Adds #13

Adds #14

This prevents multiple downloads an other conflicts. This does mean that there may be a spike in traffic as all the parties requesting the media download it at the same time. Adds #22

and include instructions on how to build this on alpine

This might fix alpine?

This doesn't work on two grounds:
* There is significant pixelation
* The resulting image size is wrong (but the frames are correct)

This is part of #28 and needs some cleaning up before it's an active code path. Pushing now because this code path won't be activated under normal circumstances.

This is the first part of #28. It defaults to false to maintain synapse parity and normal expected behaviour. This commit does the basic gif->png thumbnailing (forced when required), but does not handle resizing gifs.

Adds #27

Crop causes the thumbnail to become insanely large in Riot. For synapse parity, we'll use scale by default.

This will make more sense when the resource locks come into play - they'll need a place to live.

Currently not used though

Fixes #25 and https://github.com/vector-im/riot-web/issues/2585 (when using matrix-media-repo)

Fixes #26