Validation validates Hugr with wrong op input type

[tatiana-s]

Calling validate in Guppy (which leads to read_hugr_envelope in Selene underneath which calls package.validate()) validates a Hugr which should be invalid.

In this particular case there was a bug in the Guppy compiler where a struct was passed to an op that takes an array as input (discard_all_borrowed should only take a borrow array as input, not a tuple):

So this test for example (where the above diagram snippet is from and which used the buggy code, full file attached below) passed fine (and the bug was only discovered later during execution):

 def test_hugr_gen(validate):
    @guppy
    def main() -> None:
        for x in array(1, 2, 3):
            pass

    validate(main.compile())

array_iter.txt

[acl-cqc]

Is this to do with the entrypoint?
I think I've got a case here where the Hugr validates because the offending bit is outside the entrypoint...? (Change the entrypoint to be the module-root and validation then reports the error)

[acl-cqc]

This file: https://github.com/CQCL/tket2/blob/343e3d84607ce6d24795d0098af06eb24085f617/test_files/squashing_inline.hugr

I think that this succeeds:

use crate::extension::REGISTRY; // in tket2

let reader = BufReader::new(include_bytes!("/path/to/test_files/squashing_inline.hugr").as_slice());
let mut hugr = Hugr::load(reader, Some(&REGISTRY)).unwrap();
hugr.set_entrypoint(Node::from(NodeIndex::new(1176)));
hugr.validate().unwrap()

whereas if you instead hugr.set_entrypoint(hugr.module_root()) then validation fails. Can you reproduce like so?