hashicorp / errwrap Goto Github PK

Looking at the doc and trying out the tool I noticed that the Wrapf function is deprecated but the documentation still refers to use it instead of fmt.Errorf. At this point the README is deprecated too, right?

For context, we have similar wrapping code here: https://github.com/cloudfoundry/bosh-agent/blob/master/errors/errors.go#L52

I'm wondering why you guys decided to allow the user to specify where (and if) the error string from the wrapped error gets included in the new error message.

Do you have special use cases where you need to format them differently? It seems like having a standard pattern would be sufficient, like fmt.Sprintf("%s: %s", msg, err.Error()).

We found that if more complex formatting was required the user could easily traverse the error chain and craft their own custom message format, like we did here: https://github.com/cloudfoundry/bosh-init/blob/master/ui/fmt/error.go

https://github.com/hashicorp/errwrap#basic-usage:

	// ...
	if errwrap.Contains(err, ErrNotExist) {
		// Do something
	}
	// ...

errwrap.Contains type signature is func Contains(err error, msg string) bool. However ErrNotExist has the same name as os.ErrNotExist and seems to contain an error based on naming. This is conflicting usage. Is the type signature wrong, or the usage documentation?

It might be useful to have Wrap and Wrapf equivalents that return nil when given nil input, allowing the called to avoid preceding these calls with a nil check. Would you accept a PR for this?

Should this package be deprecated in favor of https://go.dev/doc/go1.20#errors?

Should this package simplify and wrap the native implementation? If so, would anything be left? What benefits remain after such a refactor?

Similar to:

• #9