Allow rtesting to use custom diff func

[scothis]

Test cases can now specify a custom diff function which can be used to either use custom cmp options, or replace cmp with some other differ.

The existing set of cmp options are matched using the DiffReason. Custom funcs should assume additional DiffReasons will be defined in the future. The DefaultDiff func is used if a custom Diff is not defined. Custom funcs may delegate to the default for DiffReasons they choose not to modify.

Refs #589

[scothis]

I'm unsure this is the best approach to fully decouple the existing behavior without either coupling the public API to cmp, or introducing each as a custom func.

[mamachanko]

An alternative approach could be a Differ interface with a default implementation where each "diff reason" maps to a method, e.g. Differ#DiffRaw(actual, expected) string, Differ#DiffTrackRequest(actual, expected) string etc. We'd get a bit of method sprawl, but it may be easier to provide a custom differ without accidentally missing a diff reason. At the same time, maintainers need to take care to always call the right diff method when asserting test case. To some degree that's already the case, what with the different opts / reasons.

type X509CertificateDiffer struct {
	rtesting.DefaultDiffer
}

func (d *X509CertificateDiffer) DiffStashedValue(expected, actual interface{}) string {
	return cmp.Diff(expected, actual, cmp.AllowUnexported(big.Int{}))
}

var _ rtesting.Differ = X509CertificateDiffer{}

[scothis]

One of the things I like about using dedicated methods is that additional type safety can be introduced, or in the case of stash, additional metadata can be added like the stash key.

[codecov]

Codecov Report

Attention: Patch coverage is 57.33333% with 32 lines in your changes missing coverage. Please review.

Project coverage is 59.35%. Comparing base (df0175b) to head (655f2ce).
Report is 5 commits behind head on main.

Files with missing lines	Patch %	Lines
testing/diff.go	64.44%	16 Missing ⚠️
testing/subreconciler.go	0.00%	7 Missing ⚠️
testing/webhook.go	0.00%	5 Missing ⚠️
testing/reconciler.go	0.00%	4 Missing ⚠️

Additional details and impacted files

@@            Coverage Diff             @@
##             main     #590      +/-   ##
==========================================
+ Coverage   58.26%   59.35%   +1.09%     
==========================================
  Files          33       35       +2     
  Lines        3800     3981     +181     
==========================================
+ Hits         2214     2363     +149     
- Misses       1492     1522      +30     
- Partials       94       96       +2     

☔ View full report in Codecov by Sentry.
📢 Have feedback on the report? Share it here.

[mamachanko]

I think this should work.

However, Diff is starting to compete with testing#SubReconcilerTestCase's
Verify and VerifyStashedValue and testing#ReconcilerTestCase's Verify field. Their propositions are muddling a bit, aren't they? Afaik Diff should be able to accomplish everything that Verify* is designed to do.

Maybe we should start deprecating Verify* in favour of Diff.

[mamachanko]

An alternative approach could be a Differ interface with a default implementation where each "diff reason" maps to a method, e.g. Differ#DiffRaw(actual, expected) string, Differ#DiffTrackRequest(actual, expected) string etc. We'd get a bit of method sprawl, but it may be easier to provide a custom differ without accidentally missing a diff reason. At the same time, maintainers need to take care to always call the right diff method when asserting test case. To some degree that's already the case, what with the different opts / reasons.

type X509CertificateDiffer struct {
	rtesting.DefaultDiffer
}

func (d *X509CertificateDiffer) DiffStashedValue(expected, actual interface{}) string {
	return cmp.Diff(expected, actual, cmp.AllowUnexported(big.Int{}))
}

var _ rtesting.Differ = X509CertificateDiffer{}

[mamachanko]

➕1️⃣ for Differ!

LGMT