Black outlines in image are removed when using trim()

[curiousdustin]

version: sharp@0.25.4

What are the steps to reproduce?

Attempt to trim the attached example file.

sharp(inputFilePath)
    .trim()
    .toFile(outputFilePath)

Input

Output

What is the expected behaviour?

Black outlines would be preserved.

What is the output of running npx envinfo --binaries --system?

  System:
    OS: macOS 10.15.4
    CPU: (8) x64 Intel(R) Core(TM) i7-4870HQ CPU @ 2.50GHz
    Memory: 47.95 MB / 16.00 GB
    Shell: 3.2.57 - /bin/bash
  Binaries:
    Node: 13.12.0 - ~/.nvm/versions/node/v13.12.0/bin/node
    Yarn: 1.22.4 - /usr/local/bin/yarn
    npm: 6.14.5 - ~/.nvm/versions/node/v13.12.0/bin/npm
    Watchman: 4.9.0 - /usr/local/bin/watchman```

[curiousdustin]

This seems related to #1246, #1797, and #1597, which are all closed but my issue persists.

[lovell]

The black outline in this image is part of the alpha channel - please see #2166 for a future possible enhancement to address this.

[curiousdustin]

Is there another manipulation I can do prior to trimming to work around this issue? To somehow move the outline out of the alpha channel?

[curiousdustin]

I have found that if I flatten and removeAlpha it trims as expected. Except of course then I have the background color from the flatten.

I was then able to using the offsets mentioned in the docs, to crop (extract) the original image with alpha.

The info response Object will contain trimOffsetLeft and trimOffsetTop properties.

const flattened = await sharp(inputFilePath)
    .flatten({ background: { r: 255, g: 0, b: 255 } })
    .removeAlpha()
    .toBuffer();

sharp(flattened)
  .trim()
  .toBuffer((err, data, info) => {
    const crop = {
      left: -info.trimOffsetLeft,
      top: -info.trimOffsetTop,
      width: info.width,
      height: info.height,
    };

    return sharp(inputFilePath)
      .extract(crop)
      .toFile(outputFilePath)
  });

This is probably a sluggish work around, but it appears to work well enough for my use case.

[curiousdustin]

Closing in favor of #2166