InstructionData::data allocates on every call

[cavemanloverboy]

At present, both in the client side and cpi context, using anchor's InstructionData::data() allocates a new `Vec on each call.

For a use case such as

for _ in 0..ITERATIONS {
    // cpi or tx build using InstructionDataImplementor { ... }.data()
}

its simply not necessary.

I propose a new trait method InstructionData::write_to, which would allow developers to write to an existing allocation.

pub trait InstructionData: Discriminator + AnchorSerialize {
    fn data(&self) -> Vec<u8> {
        let mut d = Self::discriminator().to_vec();
        d.append(&mut self.try_to_vec().expect("Should always serialize"));
        d
    }
    /// Clears `vec` and writes instruction data to it
    ///
    /// We use a Vec<u8> here because of the additional flexibility of re-allocation (only if necessary),
    /// and because the data field in `Instruction` expects a `Vec<u8>`.
    fn write_to(&self, mut vec: &mut Vec<u8>) {
        // Clear vector
        vec.clear();

        // Write discriminator and then Self
        borsh::to_writer(&mut vec, &Self::DISCRIMINATOR)
            .expect("Discriminator is infallibly serializable");
        borsh::to_writer(vec, &self).expect("InstructionData should be infallibly serializable");
    }
}

Example usage

let mut instruction = Instruction { ... };
for _ in 0..ITERATIONS {
    // update ix program_id, accounts if needed

    // update instruction data
    InstructionDataImplementor { .. }.write_to(&mut instruction.data);

    // cpi or tx build
}

[acheroncrypto]

Resolved in #2733