define for optional chaining is different from esbuild

[hi-ogawa]

I don't have a specific use case in mind, but I was testing a similar example from evanw/esbuild#3551 which I reported on esbuild in the past and found a minor difference/inconsistency.

reproduction: https://github.com/hi-ogawa/reproductions/tree/main/rolldown-define-optional-chain

• input

{
  define: {
    'globalThis.process.env': '{}'
  }
}

console.log(globalThis.process.env)
console.log(globalThis.process?.env)
console.log(globalThis.process?.env.ok)
console.log((globalThis.process?.env).ok)

• output (oxc, rolldown)

console.log({});
console.log(globalThis.process?.env);
console.log({}.ok);
console.log((globalThis.process?.env).ok);

• output (esbuild)

var define_globalThis_process_env_default = {};
console.log(define_globalThis_process_env_default);
console.log(define_globalThis_process_env_default);
console.log(define_globalThis_process_env_default.ok);
console.log(define_globalThis_process_env_default.ok);

[IWANABETHATGUY]

Most of the part should be done with #7433 ?

[IWANABETHATGUY]

The only difference should not relevant to optional chain? Because for a normal StaticMemberExpr we still will not generate a shared object, we will replace each DotDefineMemberExpr with a ObjectExpr

[hi-ogawa]

Thanks for the fix!
I think we can close. I've never wondered why esbuild uses shared object, but I think that's for later when that alignment turns out to be important.