CascadeMode improvements

CascadeMode.StopOnFirstFailure only affects rules in the same RuleFor chain, even when set at the validator level. This causes confusion:

CascadeMode = CascadeMode.StopOnFirstFailure;
RuleFor(x => x.Property).NotNull().NotEqual("bar");
RuleFor(x => x.AnotherProperty).NotNull().NotEqual("baz");

…so if the NotNull fails on the first rule, then the NotEqual will not run. However, the second rule will still run because these are distinct/independent rule chains. IE, this is the equivalent of:

RuleFor(x => x.Property).Cascade(CascadeMode.StopOnFirstFailure).NotNull().NotEqual("bar");
RuleFor(x => x.AnotherProperty).Cascade(CascadeMode.StopOnFirstFailure).NotNull().NotEqual("baz");

This is by design, but it isn’t intuitive and often causes confusion. I don’t think changing this behaviour is a good idea as it would be an extremely subtle breaking change that wouldn’t be caught by the compiler. Even with a major version bump this is still a very subtle change.

Investigate alternative approaches, which might involve adding new options to the CascadeMode enum to enable this behaviour.

For example, could add a new enum entry CascadeMode.FailFast. This would do the same as StopOnFirstFailure, plus also bail out after the first failure from any rule.

[weitzhandler]

I don’t think changing this behaviour is a good idea as it would be an extremely subtle breaking change that wouldn’t be caught by the compiler.

Agreed 100%. The documentation should probably be elaborated, but the functionality should be left intact.

Investigate alternative approaches, which might involve adding new options to the CascadeMode enum to enable this behaviour.

For example, could add a new enum entry CascadeMode.FailFast. This would do the same as StopOnFirstFailure, plus also bail out after the first failure from any rule.

That’s a great idea.

We could choose to mark StopOnFirstFailure as Obsolete and add another more self-explanatory enum member of an equal value, accompanying the naming style of its fail-fast/short-circuit counterpart.

This sounds like a fairly easy assignment and I’d be happy to contribute.

[weitzhandler]

Oh I understand. I actually understood it the other day and managed to reconfuse myself!

If you want to discuss cascading between rulesets, we can do that, but that isn’t the goal here. We can open a separate issue to discuss and track that, but it’s a larger issue that needs a separate discussion.

Yes, definitely, that’s what I’m aiming for!

If that isn’t something you’re interested in working on then that’s fine, but please let me know now.

The difference between chained and unchained validations are not much of a difference to me.
I might find time to work on it over the coming weeks, but it’s much lower priority than rulesets.