New "if" keyword does not work for else conditions

[anderseknert]

I would expect if to work the same following an else condition as it works for any other conditional assignment:

package play

import future.keywords.if

foo := true if {
	input.foo
} else := true if {
	input.bar
} else := false

Alternatively, without explicit assignment of true:

foo if {
	input.foo
} else if {
	input.bar
} else := false

Actual outcome:

1 error occurred: policy.rego:7: rego_parse_error: unexpected if keyword: expected else value term or rule body
	} else if {
	       ^

[srenatus]

I would expect if to work the same following an else condition as it works for any other conditional assignment

Hmm I wouldn't. 😅 I suppose it depends on whether you see the else as "attached" to the if or not. 🤔

[anderseknert]

Interesting 😄 To me it's just another form of (ordered) conditional assignment, ie. "... else foo is true if { ... }" If not read that way, how is it supposed to be read? 🙂

[srenatus]

"foo is true if ...; else it is true [if] ...; else it is false"

Hmm yeah it reads weird without the if.

[stale[bot]]

This issue has been automatically marked as inactive because it has not had any activity in the last 30 days.

[aarcamp]

I would expect if to work the same following an else condition as it works for any other conditional assignment

Same expectation. I was very happy when the "if" keyword was added to the grammar (it especially helps improve readability for those new to Rego), but the lack of "else if" leaves a gap here.

[srenatus]

🤔 The tricky thing here is forward compatibility: old OPA versions compiling new Rego policies. So far, we've been able to say that if an opa version knows the if future keyword, feeding it policies using that import is OK. If we introduce the else if construct without any new guard, we lose that: policies that use else+if would fail on some OPA versions with future kw capability for if and succeed with others (the more recent ones).

[anderseknert]

That's a good point. I guess something like import future.keywords.elseif could be a way forward, but it would certainly be more ergonomic to skip that requirement. I guess we could also consider having more advanced conditions be part of an import, perhaps when looking into https://github.com/open-policy-agent/opa/issues/5105