add a test for eval?.(code)

[mysticatea]

This PR adds a test for eval?.(code). It's indirect eval.

  12.3.9.2 Runtime Semantics: ChainEvaluation
    OptionalChain: ?. Arguments
      1. Let thisChain be this OptionalChain.
      2. Let tailCall be IsInTailPosition(thisChain).
      3. Return ? EvaluateCall(baseValue, baseReference, Arguments, tailCall).

https://tc39.es/ecma262/#sec-optional-chaining-chain-evaluation

[leobalter]

Calling implementors to check this out: @syg @codehag @anba @kmiller68 @rkirsling.

I hope we can avoid a new de-facto direct eval.

[anba]

@mysticatea (thank you!) filed https://bugzilla.mozilla.org/show_bug.cgi?id=1647169 against SpiderMonkey, which we already fixed today.

[syg]

I actually feel like this would be the least confusing if it were a direct eval.

[leobalter]

I'm currently reviewing the current tests for Optional Chaining. I can update the tests from indirect to direct eval if we have a matching PR in ECMA-262. Just let me know!

[rkirsling]

To echo @syg, I find it surprising (both as a reviewer and as an implementer) that this would be an exception to the "code with ?. behaves the same as code without it when the LHS is non-nullish" rule of thumb.

tc39/proposal-optional-chaining#21 suggests that this was not an accident, but given that browser-hosted engines all implemented and shipped what we expected the spec to say, I believe we need a proper discussion in order to resolve this divergence (even if we decide to leave the spec as it is).

I'm not sure how this discussion should be conducted (or if there's a good precedent for such a situation)—does a spec issue thread suffice or should we add a topic to the next meeting agenda?

[leobalter]

I'm not sure how this discussion should be conducted (or if there's a good precedent for such a situation)—does a spec issue thread suffice or should we add a topic to the next meeting agenda?

I believe so. We will need someone to champion this work at ECMA-262. For this specific change I believe it needs a Normative PR making sure that eval?.(x) will trigger PerformEval with the direct flag set as true.

[ljharb]

We can have the discussion without the normative PR, but filing it does accelerate things in the event the change gets consensus.