hasSizeMismatch auto-failure should be made optional

[phawxby]

We develop on windows but visual regression runs on Linux. The the most-part this is fine, however every now and then we find a component might render 1px bigger or smaller in terms of crop, it's otherwise identical.

https://github.com/americanexpress/jest-image-snapshot/blob/master/src/diff-snapshot.js#L130

[anescobar1991]

So you're proposing a flag to disable that behavior? I wouldn't be opposed to that.

In terms of your issue at hand have you tried running the tests inside a docker container? That is what we do to prevent differences between operating systems. I believe CodeSandbox does that same thing.

[sbekrin]

Hey @phawxby, I had similar issue and can suggest few things:

1. Use selenium/puppeteer server for consistent output. Running VR tests in different environments is unreliable and would likely to fail due to different text-rendering and image anti-aliasing;
2. Auto-crop whitespace area before comparing images. It's possible to do with Jimp.

Hope that helps 🙌

[phawxby]

Thanks for the help guys. I ended up getting this solved last week using Docker, and helpfully CircleCI publish their docker images publicly so I'm actually generating the image snapshots using the exact same docker image that's I'm using during CI, very helpful.

As for auto-cropping, I'm using puppeteer element screenshot of which natively crops the screenshot to just that of the target element.

I think I've got this one all covered. Thanks

[booxood]

Hope can add hasSizeMismatch can be optional, because sometimes want to ignore 1px or 2px differences.

@anescobar1991 Will accept PR? Add one option, no 'harm', right?