It was like this before but it meant importing a bunch of stuff into TcHsType which caused import cycles. Doing it this way means all the code to do with splices is in the same module rather than being split out into the zonker as well.

It needs to be orig_expr as this is used to print the message to users which describes what the splice expands to.

Grumble. OK.

But, I think q_expr and orig_expr should pretty-print identically.

say what each of these fields are.

I'm still very suspicious about why we need the GhcRn version.

Add new_ev_binds to tcg_env, so that you don't need to pass them to zonkTcGblEnv

Worth commenting that this step could in principle give rise to more finalisers... or not?

If you are going to take a TcGblEnv as argument, surely you should return one as a result! That makes the client code much cleaner.

What's the conclusion here? Can we drop orig_expr?

What's happening here? Explain!

Give a type signature for test_foo

Explain why! Something like

To run the splice, we must compile `test_foo` all the way to byte code.  But at the moment when the type checker is looking at the splice, test_foo has type `Q (alpha -> alpha)`  (NB: I think that's not quite right), and we certainly can't compile code involving unification variables!

We could default `alpha` to `Any` but then we infer `qux :: Any -> Any` which definitely is not what we want.  Moreover, if we had
  qux = [$$(test_foo), (\x -> x +1::Int)]
then `alpha` would have to be `Int`.

Conclusion: we must defer taking decisions about `alpha` until the typechecker is done; and *then* we can run the splice.  It's fine to do it later, because we know it'll produce type-correct code.

splice

Right I tried it out. Before we get:

testsuite/tests/th/T15471.hs:7:10-17: Splicing expression
    test_foo ======> foo1
testsuite/tests/th/T15471.hs:9:12-29: Splicing expression
    list_foo [|| y_a1jc ||] ======> [y_a1jc, y_a1jc]

and after we get

testsuite/tests/th/T15471.hs:7:10-17: Splicing expression
    test_foo ======> foo1
testsuite/tests/th/T15471.hs:9:12-29: Splicing expression
    list_foo Language.Haskell.TH.Syntax.unsafeTExpCoerce [|| y_a1jc ||]
  ======>
    [y_a1jc, y_a1jc]

So the two are not the same and I will keep the old behaviour.

It's just an artifact of the fact that tcTopSpliceExpr doesn't zonk the result any more. Not sure what you want me to explain?

Any thoughts?

So the two are not the same and I will keep the old behaviour.

OK. Perhaps add that as a comment to explain why we need orig_expr?

Did you act on this?

Don't forget to fix this typo

Yes, the answer is that no more finalisers can occur because of some checks in the implementation of addTopDecls. I added a comment about this.

Yes.

Did this as well.

done.