Swift Snapshot Testing in Github Actions
I have a Swift package called Knob that creates a circular knob control for iOS and macOS (and apparently tvOS too). For testing, I use the excellent SnapshotTesting package from Point•Free that supports a nice collection of testable types like NSView
and UIView
, and artifact representations such as PNG representations from those views. It works great on my laptop, but when run on Github, the tests fail because the PNG images that the tests generate are not of the same format as they are on my laptop. This is primarily due to Retina scaling, though there could also be colorspace issues and rendering environments (Metal vs CPU) coming into play as well.
The key to making the tests work in a Github continuous-integration (CI) workflow is to partition the snapshot artifacts into those that come from a dev environment such as my laptop, and those that comee from Github workflow. To that end, I followed a comment made by another user of the SnapshotTesting library and modified the artifact names based on the environment the test was running in. Here is the function I use to assert that the rendering of my test knob is the same as before:
func assertSnapshot(file: StaticString = #file, testName: String = #function, line: UInt = #line) throws {
knob.layoutSubtreeIfNeeded()
knob.display()
let snapshotEnv = ProcessInfo.processInfo.environment["SNAPSHOT_ENV"] ?? "dev"
let failure = verifySnapshot(matching: knob,
as: .image(precision: 1.0, perceptualPrecision: 1.0),
named: snapshotEnv,
record: isRecording,
snapshotDirectory: nil,
file: file,
testName: testName,
line: line)
guard let message = failure else { return }
XCTFail(message, file: file, line: line)
}
The start of the method asks the knob to render itself prior to being used for a snapshot – for some reason on Github without this I was seeing blank snapshots, and my tests were not exercising some custom NSBezierPath
routines.
The Github CI workflow script set the SNAPSHOT_ENV
value to be “ci” before it starts the bnild and test stage of the workflow:
- name: Build and Test
run: make
env:
SNAPSHOT_ARTIFACTS: "$PWD/.snapshots"
SNAPSHOT_ENV: ci
It also sets SNAPSHOT_ARTIFACTS
so that if the SnapshotTesting library detects a difference, it will save the artifact for the test failure in a location that we can use later on to upload to Github as a workflow artifact.
The ability to set SNAPSHOT_ARTIFACTS
is nice, but unfortunately if the test failes because no artifact exists to compare against, the SnapshotTesting library ignores it and instead plops the image file into the repo location where it expected to find it, which is not very useful for me. To overcome this, I added a step in the CI workflow that saves the entire contents of this location if the test step failed:
- name: Copy Snapshots
if: ${{ failure() }}
run: |
mkdir .snapshots
cp -r $(find . -name Knob-macOSTests)/__Snapshots__ .snapshots/
Now, when a new test fails because there is not a “ci” tagged image, I can download the artifacts, locate the new “ci” image file and manually add it to the repo so that a subsequent commit will make it available for the CI tests. Not ideal, but it works well enough for the times when I add a new test case.
If tests fail because of differences fonund in an existing artifact, the SnapshotTesting framework will do the right thing and deposit the artifact from the failed test into the SNAPSHOT_ARTIFACTS
directory. This allows me to evaluate the difference and hopefully fix why the test failed.
Finally, here is the CI step that uploads any image artifacts. Note that it only executes if the build/test stage fails, just like the copy step above.
- name: Upload Snapshot Failures
uses: actions/upload-artifact@v3
if: ${{ failure() }}
with:
name: snapshots
path: .snapshots/
The entire CI script is viewable at CI.yaml