feat: support wrapped plugin ops

[afinch7]

Add support for op wrapping pattern used in cli:

#[no_mangle]
pub fn deno_plugin_init(interface: &mut dyn Interface) {
  // this won't work when `register_op` takes a function pointer(`fn`)
  interface.register_op("example", wrap_op(op_example));
}

fn wrap_op<D>(d: D) -> Box<dyn Fn(&mut dyn Interface, Value, Option<ZeroCopyBuf>) -> Op> 
where
  D: Fn(&mut dyn Interface, Value, Option<ZeroCopyBuf>) -> Op,
{
  // code for wrapping op
}

fn op_example(
  _interface: &mut dyn Interface,
  control: &[u8],
  zero_copy: Option<ZeroCopyBuf>,
) -> Op {
  // Op code
}

fixes #5478

[afinch7]

@piscisaureus Any idea why this only fails on windows?

[CLAassistant]

All committers have signed the CLA.

[bartlomieju]

Thanks @afinch7 could you please rebase this branch and add some test case?

[afinch7]

Going to rebase this and add a test case with json ops.

[afinch7]

I decided the json op test was too complicated, so I just did something simpler.

[afinch7]

I think I finally figured out what causes that segfault on windows. The default drop implementation for the closure created in cli/ops/plugin.rs for each op drops the captured plugin_lib value first. I don't know if there is a way to change the order of these actions here.

I think the "correct" solution here is to make sure that all of a plugins ops are unregistered first when it is dropped.

[afinch7]

@bartlomieju This is ready for review.

[bartlomieju]

@afinch7 thanks, I'll leave review to @piscisaureus as he's refactoring the ops now.

[piscisaureus]

The issue I have with this is that I see that you're wrapping ops but I have no idea why this is needed at all..

[afinch7]

It's not needed, but it's not currently possible to implement anything like register_op_minimal for plugin ops. I would just use a generic in Interface, but as far as I know that's not possible.

[afinch7]

With the recent changes to the op registry I found it was actually fairly easy to implement my "correct" solution of actually unregistering a plugins ops when it is closed.

[kitsonk]

This is quite stale now. We should either rebase or close.

[bartlomieju]

I'm gonna close this PR without a merge because it's stale. In light of #8490 it's not clear if we want to make those changes, remove plugins completely or rework them.