perf: fmt.Errorf used instead of errors.New

[egonelbre]

Throughout the whole codebase there are many places that use fmt.Errorf whereas errors.New can be used. Using fmt.Errorf is roughly 4x slower than errors.New.

This can be easily benchmarked with:

var result error

func BenchmarkFmtError(b *testing.B) {
	for range b.N {
		result = fmt.Errorf("something")
	}
}

func BenchmarkErrorsNew(b *testing.B) {
	for range b.N {
		result = errors.New("something")
	}
}

On my computer I'm getting:

BenchmarkFmtError-32             9535638               123.8 ns/op            32 B/op          2 allocs/op
BenchmarkErrorsNew-32           32106765                33.87 ns/op           16 B/op          1 allocs/op

This performance issue can be automatically fixed with:

gofmt -w -r "fmt.Errorf(s) -> errors.New(s)" .

Those error paths usually aren't hot paths, however it might impact the size of code generated.

[codyoss]

Thanks for the detailed report. This is a good suggestion. It looks like a good number of these are in our generated code so we can patch our generator to remove a lot of these and gofmt the rest.

[quartzmo]

@codyoss Can this be closed now?

[egonelbre]

There are still 137 cases of fmt.Errorf being used in the repo. This can be easily seen by running gofmt -w -r "fmt.Errorf(s) -> errors.New(s)" on the whole repository.