Provide rule when a default parameter value should be used about rho

[arouel]

I added a first implementation with commit 35a1f81.

[bryce-anderson]

I like the feature, but would like to see if we can keep the AST clean by moving the validation to the parsers. The question arises what to do when a value parses, but fails validation? In that case I think we need a ternary representation for the return value. Something like

sealed trait ParseResult[T] {
    def flatMap(... blah blah... ) ...
}
case class ParseFail(error: String) extends ParseResult[Nothing]
case class ValidationFail(error: String) extends ParseResult[Nothing]
case class Success[T](value: T) extends ParseResult[T]

[arouel]

Sounds good. We should do that improvement. Than I can also solve other issues more easily.

[bryce-anderson]

There are two schools of thought concerning handling error in the presence of a default.
1: On failure use the default
2: On failure fail.
Currently we are doing 1, but I like 2 better. I think returning a unknown value can result in unexpected results especially when an interface changes. I've made a branch to do that, particularly commit 1436324. The tests are not all updated to the new result, but I'll get that done soon.
Is there a show stopper with this change?

[bryce-anderson]

The tests are brought into compliance with commit 7c2bbc7.

[arouel]

I'm with you and prefer the 2 over 1. This makes my use cases much easier instead of working with Option everywhere.

[bryce-anderson]

I feel like this should be closed. If thats not true, feel free to reopen.

[arouel]

I thought a bit about the current behavior and have the following question: Do we want to use the default value in case a number is expected but cannot be parsed? Maybe this is more handy?

[bryce-anderson]

We have the following possibilities when querying for an Int with id foo:

• /hello/world Use the default.
• /hello/world?foo=1 Use the result.
• /hello/world?foo I say use the default as no value was given.
• /hello/world?foo= I say this is invalid because you've passed an invalid value, "".
• /hello/world?foo=one Should fail, this is an invalid query. Otherwise I would consider any result spooky.

In summary, I think we should only use the default when a value has not been presented, and never when an invalid value has been presented. The client should be alerted to their mistake so they can fix their behavior, not given a result which they may not have intended to fetch. Are there good reasons for behavior contrary to that above?

[arouel]

Make sense to behave as you described it. Maybe we can improve the HTTP response by providing a better status code. A JSON-representation with a message would be nice but should be defined by the user itself.

Okay, now I have it. A user should be able to define a strategy/function in case the value is not parseable. Does it make sense?

[bryce-anderson]

Yes, a failure strategy would be nice. We should make an issue for that and
start a branch.
On Jul 23, 2014 9:37 AM, "André Rouél" [email protected] wrote:

Make sense to behave as you described it. Maybe we can improve the HTTP
response by providing a better status code. A JSON-representation with a
message would be nice but should be defined by the user itself.

Okay, now I have it. A user should be able to define a strategy/function
in case the value is not parseable. Does it make sense?

—
Reply to this email directly or view it on GitHub
#12 (comment).

[arouel]

We work in a new branch for that feature.