Ability to choose alternative comparison library

[phawxby]

Looking at github commits it appears blink-diff is pretty much dead, no activity in over a year and this issue suggests the author has left Yahoo.

Being able to pick a more modern comparison library while retaining blink-diff as the default would probably make sense.

Resemble-js is an option but again, there hasn't been much activity in over a year.

So I'm thinking pixelmatch would make sense, although it has less features than the other libraries it has an active community and by far the most popular image comparison library on npm.

Thoughts?

[anescobar1991]

That is a very good point, and one that I have already spent some time thinking about.

However the blocker I have with most other libraries is that they need to be able to run synchronously and they should not require any external dependencies like canvas.

I would love to be able to use a better maintained library but these are the constraints that currently hold me to blink-diff. For the time being though I am quite limited on time but feel free to explore this option and open a PR and/or keep discussing here until I can find some time to implement a solution!

[anescobar1991]

Solved by #19

[cdeutsch]

After a week of using v2 with pixelmatch, I like blink-diff better.

I've been getting a lot of inconsistent results with pixelmatch. I often run with 0 as the threshold, and it will pickup tiny diffs in one run but not the next.

A side note, it doesn't look like pixelmatch is in active development either (maybe neither really need to be). Last significant commit to pixelmatch was over a year ago.

I like the original idea of picking a comparison library versus just switching wholesale to pixelmatch, but obviously that's a lot more work and complicates things.

For now I'll just go back to v1.0.1 which was meeting my needs so far, just figured I'd share my experience.

Love this library by the way! 👍

[phawxby]

@cdeutsch do you know of any others comparably fast and in active development?

Also are you able to provide any test cases? The only similar cases I have seen relate to slight font rendering differences between Puppeteer on Windows verses our Linux based CI environment. For that reason I implemented failureThreshold and failureThresholdType to allow for a few percentage points of difference between snapshots.