If it's a bug that the typechecker does not produce an error for a missing $$, we can close this PR and file an issue in https://github.com/facebook/hhvm.

facebook/hhvm#8698

There is already an issue tracking this 😉.
Please don't let this convince you to withdraw this linter. Pipes without $$ are fishy it should be flagged. A lint rule could be deployed far quicker than a language change. I would be surprised if www would typecheck after 2.5 years since the unintentional (?) removal of the check. Strongly in favor of shipping this linter, even if it becomes a language level error later.

Two of the test suites are failing because they could not run hhvm and two are failing with this error:

 Server launched with the following command:
  	'/usr/local/Cellar/hhvm-nightly/2022.05.03/bin/hh_server' '-d' '/Users/runner/work/hhast/hhast' '--waiting-client' '8'
  Running in daemon mode
  Server connection took over 2.0 seconds. Refreshing...
  Typing[4297] Was expecting an array or collection but type is unknown [1]
  -> It is unknown because the type of the lambda parameter could not be determined. Please add a type hint to the parameter [2]
  
  src/__Private/XHProf.hack:32:53
        30 |     invariant(self::$enabled === true, "Can't disable twice");
        31 |     self::$enabled = false;
  [1,2] 32 |     $raw = Dict\map(\xhprof_disable(), $v ==> (int) $v['wt']);
        33 | 
        34 |     $inclusive = dict[];
  
  1 error found.
  Error: Process completed with exit code 2.

This seems like an unrelated

Yes, I ran into that too. You can add a dict<arraykey, mixed> before $v and explicitly reformat. This makes the tests pass again without conflict with my branches.

6c00e36 is the diff you can apply.

Retrigger the CI

Retriggered CI again to get the build to pass on macOS. I think this PR is now ready for review/discussion.