How to pass install arguments from dfx?

I’m trying to pass some arguments when installing / upgrading a canister from dfx and I can’t seem to figure it out.

In the rust canister I have this code:

#[derive(CandidType, Deserialize, Debug, Default)]
struct SendArgs {
    greet: String,
}

#[init]
fn init() {
    let call_arg = ic_cdk::api::call::arg_data::<(Option<SendArgs>,)>().0;
    ic_cdk::print(format!("Init: {:?}", call_arg));
[...]
}

When installing the code from another rust canister, like here, this works and I can read call_arg. I just can’t seem to figure out how to send this from dfx.

Things I’ve tried:

dfx deploy ic_test --argument '(record {greet="Test"})' --mode reinstall
dfx deploy ic_test --argument '(opt record {greet="Test"})' --mode reinstall

dfx canister install --mode reinstall --argument '(opt record {greet="test"})' ic_test

All I get is
[Canister rrkah-fqaaa-aaaaa-aaaaq-cai] Init: None

I also tried arg_data::<(Option,)>() and arg_data::<(String,)>() without any success. I think the issue is with how I send the data from dfx.

For context, I’m trying to call ::arg_data from post_update, so we can use a neat pattern that @Seb showed us during the Friday call.

1 Like

How about something like this?

#[init]
fn init(foo: String) {
  ic_cdk::print(format!("foo: {:?}", foo));
}
#[init]
fn init(foo: String) {
    ic_cdk::print(format!("Init: {:?}", foo));
}

+

dfx start --clean
dfx canister create ic_test
dfx canister install --argument '("Test")' ic_test

[OR]

dfx deploy ic_test --argument '("Test")'

Output
Error: The Replica returned an error: code 5, message: "Canister rrkah-fqaaa-aaaaa-aaaaq-cai trapped explicitly: Custom(No more values on the wire, the expected type text is not opt, reserved or null)"

#[init]
fn init(foo: Option<String>) {
    ic_cdk::print(format!("Init: {:?}", foo));
}

Same inputs, output is always:

[Canister rrkah-fqaaa-aaaaa-aaaaq-cai] Init: None

Maybe @senior.joinu or @nomeata can shed some light on this.

Can you try opt "Test"?

dfx deploy ic_test --argument ‘(opt “Test”)’ with init(foo: String):

Error: The Replica returned an error: code 5, message: “Canister rrkah-fqaaa-aaaaa-aaaaq-cai trapped explicitly: Custom(No more values on the wire, the expected type text is not opt, reserved or null)”

dfx deploy ic_test --argument ‘(opt “Test”)’ with init(foo: Option):

No error, but output is None.

I have a feeling it’s something that dfx does when adding the parameter, I’m checking the repo now

A-ha! Some progress. Following my hunch that dfx does something weird (rust to rust works, afterall) I tried sending raw arguments, like so:

$ didc encode '("Test")'
4449444c0001710454657374

$ dfx deploy ic_test --argument 4449444c0001710454657374 --argument-type raw

And now I get [Canister rrkah-fqaaa-aaaaa-aaaaq-cai] Init: "Test"

And now my old code works as well

[Canister rrkah-fqaaa-aaaaa-aaaaq-cai] PU: Some("Test")
[Canister rrkah-fqaaa-aaaaa-aaaaq-cai] Post upgrade ended

So both post_upgrade(args: T) and post_upgrade(){ic_cdk::api::call::arg_data()} work fine. It’s just that dfx has some issues when sending idl. Works with raw, though it would be a bit awkward to also use didc…

Please create an issue :slightly_smiling_face:

done

…20char 20 more 20202020

1 Like

Posting here as well so this gets documented for other people trying the same things:

@chenyan figured out that the problem was my did file. It needs to include the params at a service level, so dfx can work. (e.g. service : (text) → { … })

Old did file:

service : {
  add : (text) -> (bool);
  getAll : () -> (vec text) query;
  greet : (text) -> (text) query;
}

New did file:

service : (text) -> {
  add : (text) -> (bool);
  getAll : () -> (vec text) query;
  greet : (text) -> (text) query;
}

Just a heads up, if you’re currently using the __get_candid_interface_tmp_hack trick to __export_service() this doesn’t pick up the fact that you need those params in the init fn so it doesn’t add the proper definition. Waiting on a response from github to see if this was ever intended, or it’s something just not needed.

3 Likes

Right, so now I’m a bit more clear on what happened. I am using the following pattern to output the did file:

// Auto export the candid interface
candid::export_service!();

#[query(name = "__get_candid_interface_tmp_hack")]
#[candid_method(query, rename = "__get_candid_interface_tmp_hack")]
fn __export_did_tmp_() -> String {
    __export_service()
}

#[cfg(test)]
mod tests {
    use super::*;

    #[test]
    fn check_candid() {
        let expected = String::from_utf8(std::fs::read("ic_test.did").unwrap()).unwrap();

        let actual = __export_service();

        if actual != expected {
            println!("{}", actual);
        }

        assert_eq!(
            actual, expected,
            "Generated candid definition does not match expected did file"
        );
    }
}

The tests makes sure that I get an error every time the candid interface changes, however I didn’t know how to decorate the init fn so the proper signature got used. I had tried using #[candid_method(update)] but that didn’t work, as it would add the init() fn to the methods, but not update the service signature. The proper way to annotate the init() fn is this: #[candid_method(init)]

This now produces the correct did singnature:

service : (text) -> {
  add : (text) -> (bool);
  getAll : () -> (vec text) query;
  greet : (text) -> (text) query;
}
2 Likes

Where is ic_test.did relative to the file containing the Rust code that tries to read it?

I have lib.rs in the same directory as the .did file but get a "No such file or directory". I tried a few different paths.

For the example above? It’s in the same dir as the Cargo.toml

$ tree
.
├── Cargo.lock
├── Cargo.toml
├── dfx.json
├── README.md
└── src
    └── ic_test
        ├── Cargo.toml
        ├── ic_test.did
        └── src
            ├── env.rs
            ├── lib.rs
            └── lifetime.rs

3 directories, 9 files

This just solved my problem, thank you

3 Likes