rego vs wasm: x == arr[x]

[srenatus]

This one is hard to summarise, but the example is clear:

package play

p {
	some foo
	foo == input.foos[foo]
}

echo '{"foos": ["foo1", "foo2"] }' | opa eval -fpretty -d pol.rego -I data.play.p

topdown	wasm
undefined	true

Note that it's important that the RHS is input.foos, since if the array was known, we'd get a compile time error:

package play

p {
	foos := ["foo1", "foo2"]
	some foo
	foo == foos[foo]
}

$ opa eval -fpretty -d 2e.rego -I data.play.p
1 error occurred: 2e.rego:6: rego_type_error: match error
	left  : number
	right : string

[srenatus]

FWIW it's compiled into that rego

p = true { __local0__ = input.foos[__local0__] }

The planner will plan this:

| *ir.Funcs Funcs (1 funcs)
| | *ir.Func g0.data.test.p (2 params: [Local<0> Local<1>], 3 blocks, path: [g0 test p])
| | | *ir.Block Block (3 statements)
| | | | *ir.ResetLocalStmt &{Target:Local<3> Location:{File:0 Col:1 Row:2 file:module-0.rego text:p}}
| | | | *ir.DotStmt &{Source:{Value:Local<0>} Key:{Value:String<1>} Target:Local<4> Location:{File:0 Col:3 Row:4 file:module-0.rego text:foo == input.foos[foo]}}
| | | | *ir.ScanStmt &{Source:Local<4> Key:Local<5> Value:Local<6> Block:Block (3 statements) Location:{File:0 Col:3 Row:4 file:module-0.rego text:foo == input.foos[foo]}}
| | | | | *ir.Block Block (3 statements)
| | | | | | *ir.AssignVarStmt &{Source:{Value:Local<5>} Target:Local<7> Location:{File:0 Col:3 Row:4 file:module-0.rego text:foo == input.foos[foo]}}
| | | | | | *ir.AssignVarStmt &{Source:{Value:Local<6>} Target:Local<8> Location:{File:0 Col:3 Row:4 file:module-0.rego text:foo == input.foos[foo]}}
| | | | | | *ir.AssignVarOnceStmt &{Source:{Value:Bool<true>} Target:Local<3> Location:{File:0 Col:1 Row:2 file:module-0.rego text:p}}
| | | *ir.Block Block (2 statements)
| | | | *ir.IsDefinedStmt &{Source:Local<3> Location:{File:0 Col:1 Row:2 file:module-0.rego text:p}}
| | | | *ir.AssignVarOnceStmt &{Source:{Value:Local<3>} Target:Local<2> Location:{File:0 Col:1 Row:2 file:module-0.rego text:p}}
| | | *ir.Block Block (1 statements)
| | | | *ir.ReturnLocalStmt &{Source:Local<2> Location:{File:0 Col:1 Row:2 file:module-0.rego text:p}}

The relevant section is

| | | | *ir.ScanStmt &{Source:Local<4> Key:Local<5> Value:Local<6> Block:Block (3 statements) Location:{File:0 Col:3 Row:4 file:module-0.rego text:foo == input.foos[foo]}}
| | | | | *ir.Block Block (3 statements)
| | | | | | *ir.AssignVarStmt &{Source:{Value:Local<5>} Target:Local<7> Location:{File:0 Col:3 Row:4 file:module-0.rego text:foo == input.foos[foo]}}
| | | | | | *ir.AssignVarStmt &{Source:{Value:Local<6>} Target:Local<8> Location:{File:0 Col:3 Row:4 file:module-0.rego text:foo == input.foos[foo]}}
| | | | | | *ir.AssignVarOnceStmt &{Source:{Value:Bool<true>} Target:Local<3> Location:{File:0 Col:1 Row:2 file:module-0.rego text:p}}

which is simply kind of wrong. They key Local<5> is unified with a new var, Local<7> (the first AssignVarStmt). The second one is the result of unifying the value Local<6> with another new var, Local<8>. When the second AssignVarStmt is planned, we should really notice that we don't need a new var but should unify with Local<7>... 🤔