Anchor v2 defined arg struct requires IdlAccountType impl

[rute-sophie]

I'm trying out anchor-v2 and came up with an issue when creating custom argument structs.

I followed the example here and derived #[derive(Clone, Copy, wincode::SchemaRead, wincode::SchemaWrite)].

Compilation fails with

error[E0277]: Ensure that `instructions::my_instr::MyInstrArgs` has an `#[account]` attribute
  --> programs/my-program/src/lib.rs:18:15
   |
18 |         args: MyInstrArgs
   |               ^^^^^^^^^^^^^^^^ unsatisfied trait bound
   |
help: the trait `IdlAccountType` is not implemented for `instructions::my_instr::MyInstrArgs`

Was able to find two workarounds, both are very duck-tapey and not the desired solution:
1- Deriving #[account] - this one is odd and requires padding to the structure. Also not an account.
2- anchor build --no-idl - this is problematic as it skips idl generation.

Seems to be an issue with IDL generation for anchor v2.

[eteen12]

Traced this — the example you followed is actually what's wrong. The args struct needs #[derive(IdlType)] on top of the wincode derives, that's what gives it the IdlAccountType impl the IDL build is asking for:

#[derive(Clone, Copy, IdlType, wincode::SchemaRead, wincode::SchemaWrite)]
pub struct MyInstrArgs { ... }

It's in the prelude already. No #[account] padding hack or --no-idl needed.

The example in tests-v2 gets away without it because that crate ships a pre-written IDL json and never actually runs idl build, so nothing ever caught it. And the error message suggesting #[account] definitely didn't help.

Opened #4854 to fix the example, make the error point at IdlType, and add coverage so the arg path actually goes through idl build in CI.

[rute-sophie]

Thank you for the suggestion. It worked!

[eteen12]

@jamie-osec #4854 should close this issue,no ?