WebP animation decoder requires all frames to be decoded up front

[Shnatsel]

As of v0.25.4, WebPDecoder::into_frames() decodes the entire animation up front, as opposed letting the user read frames one by one.

This is a blocker for Oculante to migrate away from libwebp-sys + webp-animation to image's memory-safe WebP animation decoder: woelper/oculante#252 (comment)

This will also be an issue for wondermagick which will need to support decoding a set of specific frames, and always decoding all of the frames will be slow.

[fintelia]

Are you sure? Looking at the code, it seems that next_frame is called lazily within webp::FramesInner::next each time a new frame is requested. Though if you call collect on the iterator it'll of course decode all the frames.

image-webp does scan the chunk headers for each frame of the animation when creating the decoder (to determine the animation runtime and determine whether the image uses lossy compression), but I'd expect that to be comparatively cheap.

[Shnatsel]

That was an incorrect assumption, the performance bottleneck is elsewhere: woelper/oculante#252 (comment)

I'll investigate and open an issue with a profile.