Validate custom time.Duration types

[pgporada]

We can now validate custom time.Duration types defined as type MyDuration time.Duration with all of the baked in tags such as gt, gte, eq, lt, lte, etc.

In an application, the caller will need to define their custom duration, instantiate a new validator, register the custom duration type, then configure the validation tags where applicable. What that looks like in practice is

package example

// Sometimes we need a custom type because we have application specific methods
type MyDuration time.Duration

func (m *MyDuration) UnmarshalJson() ...
func (m *MyDuration) MarshalJson() ...

package config

type MyConfig struct {
  TLS TLSConfig
  ConnectMaxIdleTime example.MyDuration `validate:"omitempty,lte=60s"`
  ConnectionLifetime example.MyDuration `validate:"required,gte=5s"`
}

func ValidateConfig(mc MyConfig, in io,Reader) {
	// Initialize the validator and load any custom tags.
	validate := validator.New()
	for tag, v := range cv.Validators {
		err := validate.RegisterValidation(tag, v)
		if err != nil {
			return err
		}
	}

	// Register example.MyDuration so the validator understands it and map it to an existing type.
	validate.RegisterCustomType(time.Duration(0), example.MyDuration(0))

       err := validate.Struct(mc)
}

[pgporada]

Thank you @aaomidi for helping me figure this out.

[beautifulentropy]

Is this rename necessary?

Suggested change

	val := reflect.ValueOf(struct{}{})
	v := reflect.ValueOf(struct{}{})

[pgporada]

Yes because of the new input parameter v *Validate. That parameter is needed to meet the function signature defined here. Without the rename I get an a "no new variables on left side" error.

[beautifulentropy]

Here and below, v *Validate param is being added but never used. I'm not understanding why this is the case.

[pgporada]

All bakedInValidators have to match the Func function signature defined here. Should this parameter be func isMAC(_ *Validate, fl FieldLevel) bool instead since I don't use v?

[aarongable]

Having not taken the time to fully understand the whole validator package myself, it appears that there's something very weird going on where some things happen at the *Validate instance level, and others happen at the global level. This existing inconsistency is leading this PR to become very large while attempting to accomplish a seemingly-straightforward goal.

I don't think that this PR is the right place to do a massive refactoring of this library to allow everything to happen at the instance level. If we want to do that (and maybe we do!) it should be as a separate PR, where we can consider multiple different ways of making that happen (e.g. making lots of current top-level helper functions instead be methods on that type, which seems cleaner than having those methods take an instance as their first argument).

[pgporada]

Closing in favor of letsencrypt/boulder#7564