SDK Install script does take Linux-aarch64 into consideration.

[ocluf]

I want create a starter canister repo with a devcontainer so that everybody can easily clone it and get started without worrying about needing to install the toolchain.

However since I don't have an intel laptop but an M1 it seems that the install script fails at get_architecture() when I try to run

RUN sh -ci "$(curl -fsSL https://internetcomputer.org/install.sh)"

I expected to see this happen:

The sdk getting installed in my docker image

Instead, this happened:

Error exit code 1: dfinity-sdk: unknown CPU type: aarch64

[sesi200]

dfx currently requires you to have rosetta installed. You can install it by running softwareupdate --install-rosetta in your terminal. With rosetta installed, x86_64 will be an available architecture and dfx should run without any problems.

I made a note to update docs and install script to provide instructions in that case. It can easily be detected and we should provide proper error messages in that case.

[ocluf]

The error didn't surface on my device, but when I try to build a docker container with a dev environment. I got it working by adding the --platform=linux/x86_64 flag in:

FROM --platform=linux/x86_64 mcr.microsoft.com/vscode/devcontainers/rust:0-${VARIANT}

The build is quite slow though due to needing to emulate x86_64 with QEMU. I could publish an image but devs might rather build it from scratch so they can trust it better. I also don't know how it would perform on a macbook air.

I'm unblocked for now so this issue can be closed. I'll see if I can find a cleaner way to choose the target platform so everybody can use the .devcontainer configuration without needing to make manual changes. Do you know why dfx can't be build for ARM? I know lots of people at the foundation have ARM macs as well so I assume it has been tried.

[ocluf]

Hmm sadly I did not get it working in the end. The docker container successfully was build but when I try to run a local replica I get the following error.

dfx start -v Running dfx start for version 0.12.1 There is no project-specific network 'local' defined in /workspaces/starter/dfx.json Using the default definition for the 'local' shared network because /root/.config/dfx/networks.json does not exist. Local server configuration: bind address: 127.0.0.1:4943 bitcoin: disabled canister http: enabled log level: Error replica: subnet type: Application log level: Error data directory: /root/.local/share/dfx/network/local scope: shared

Version v0.12.1 installed successfully. Starting canister http adapter... canister http adapter ready Starting replica... thread 'replica-actor' panicked at 'called Result::unwrap() on an Err value: Cannot start the replica: Timeout', src/dfx/src/actors/replica.rs:376:78 note: run with RUST_BACKTRACE=1 environment variable to display a backtrace qemu: uncaught target signal 6 (Aborted) - core dumped Aborted

Has someone at the foundation successfully ran a replica in a docker container on an ARM macbook?

[domwoe]

Hey @ocluf ,

see this discussion: https://github.com/dfinity/sdk/discussions/2642

It worked (slowly) with an older replica version, but not anymore. @mbeekman wanted to try with nix builds but I'm not sure if he got it working in the end.

[ocluf]

Thanks @domwoe that seems like the right thread to continue the discussion I'll leave a comment there in this one can be closed