StripExif doesn't work

[christiandavilakoobin]

Describe the bug
Until 8.0.2, we can remove exif info using setting the edit rotate value to "null". Now, this no longer works, and we there is the option stripExif. This was added on the PR #635 but it does not seems to work. It tries to overwrite all the exif adding a "Software" key, but it does not remove all the other ones as intended:

if (edits?.stripExif === true) {
        // Removes all EXIF by inserting minimal EXIF tag. Leaves ICC untouched.
        image.keepIccProfile().withExif({
          IFD0: {
            Software: 'Dynamic Image Transformation for Amazon CloudFront'
          }
        });
        delete edits.stripExif;
      }

It just ADDS the Software key but keeping the old ones.

To Reproduce
'edits' => [
'stripExif' => true,
'stripIcc' => true,
'resize' => [
'width' => 285,
'height' => 285,
'fit' => 'cover',
'position' => 'entropy',
'serial' => 81
],
]

Expected behavior

Please complete the following information about the solution:

• Version: 8.0.3
• Region: eu-west-1
• Was the solution modified from the version published on this repository? No

Screenshots

[gsingh04]

Please allow us to look into it and update you.

[bryceaug]

@christiandavilakoobin Do you have the sample image you used for your comparison? Would you be able to make it available to us, or provide a similar image that produces the same behavior?

I ask because in your second screenshot, additional tags like Creator Tool, Create Date , Title, etc that were not present in the 1st screenshot have been inserted. This is not consistent with how DIT behaves when stripping EXIF.

Thanks!

[christiandavilakoobin]

Hi @bryceaug!

I'm trying with this image:

Thanks!

[bryceaug]

Hey @christiandavilakoobin,

While we do support the transformations stripIcc and stripExif in DIT's ECS based implementation, these transformations keys are not supported in our Lambda implementation. This means from your B64 encoded payload, the stripIcc and stripExif keys are not being used.

The recommended way to strip EXIF data is still to supply a rotate: null transformation in your B64 encoded payload. If you want to also strip the color space, you'd need to use our Thumbor style request format, and supply /filters:strip_exif()/ and /filters:strip_icc()/.

Let me know if you continue to see issues once you swap out stripExif for rotate: null

[christiandavilakoobin]

Hi @bryceaug! I did use the rotate: null but it stopped working. Looking the source code, it seems that block was removed in the same PR.

[bryceaug]

Hey @christiandavilakoobin, apologies.

I misspoke in my previous comment. We did indeed remove Exif stripping when supplying rotate: null. Our intentions are to more closely align with Sharp, which has a rotate: null -> autoOrient() behavior baked in. While it applies orientation, and does remove the orientation tag, it does not strip all Exif. You can find information on this here.

Also, while not explicitly mentioned on our documentation, like you attempted, you can invoke Exif stripping with stripExif: true in your B64 encoded payload. We will update the documentation to cover this shortly.

That being said, after replicating the issue with your sample image, I can confirm there is a defect with Exif stripping with regards to how DIT, Sharp, and ultimately libvips interact with each other. The end result is that our configurations are applied additively to the Sharp instance, which prevents us from resetting the keepMetadata bitmask through Sharp, that libvips ultimately consumes. We will resolve this in our next update, and I can update this thread when that becomes available.

In the meantime, you can downgrade to a previous version, such as 8.0.2, and use either the rotate: null transformation, or the Thumbor /filters:strip_exif()/ filter.

[christiandavilakoobin]

Thanks! I'll stick onto 8.0.2 until this is fixed.