Skip to content

Latest commit

 

History

History
132 lines (98 loc) · 8.43 KB

README.md

File metadata and controls

132 lines (98 loc) · 8.43 KB

CosmWasm Optimizing Compiler

This is a Docker build with a locked set of dependencies to produce reproducible builds of cosmwasm smart contracts. It also does heavy optimization on the build size, using binary stripping and wasm-opt.

Image Description x86_64 images (default) ARM images (experimental1)
optimizer Combines rust-optimizer and workspace-optimizer in a single image cosmwasm/optimizer
DockerHub
cosmwasm/optimizer-arm64
DockerHub
rust-optimizer deprecated Single contract builds cosmwasm/rust-optimizer
DockerHub
cosmwasm/rust-optimizer-arm64
DockerHub
workspace-optimizer deprecated Multi-contract workspaces (e.g. cosmwasm-plus) cosmwasm/workspace-optimizer
DockerHub
cosmwasm/workspace-optimizer-arm64
DockerHub

1 ARM images do not produce the same output as the default images and are discouraged for production use. See Notice below.

Usage

This works for most cases, for monorepo builds see advanced

The easiest way is to simply use the published docker image. You must set the local path to the smart contract you wish to compile and it will produce an artifacts directory with <crate_name>.wasm and checksums.txt containing the hashes. This is just one file.

Run it a few times on different computers and use sha256sum to prove to yourself that this is consistent. I challenge you to produce a smaller build that works with the cosmwasm integration tests (and if you do, please make an issue/PR):

docker run --rm -v "$(pwd)":/code \
  --mount type=volume,source="$(basename "$(pwd)")_cache",target=/target \
  --mount type=volume,source=registry_cache,target=/usr/local/cargo/registry \
  cosmwasm/optimizer:0.16.0

Demo this with cosmwasm-examples (going into eg. erc20 subdir before running), with cosmwasm-plus, or with a sample app from cosmwasm-template.

Note that we use one registry cache (to avoid excessive downloads), but the target cache is a different volume per contract that we compile. This means no interference between contracts, but very fast recompile times when making minor adjustments to a contract you had previously created an optimized build for.

Mono Repos

Contracts as Workspace Members

This is designed for cosmwasm-plus samples. We use a separate docker image

Sometime you want many contracts to be related and import common functionality. This is exactly the case of cosmwasm-plus. In such a case, we can often not just compile from root, as the compile order is not deterministic and there are feature flags shared among the repos. This has lead to issues in the past.

For this use-case we made a second docker image, which will compile all the contracts/* folders inside the workspace and do so one-by-one in alphabetical order. It will then add all the generated wasm files to an artifacts directory with a checksum, just like the basic docker image (same output format).

To compile all contracts in the workspace deterministically, you can run:

docker run --rm -v "$(pwd)":/code \
  --mount type=volume,source="$(basename "$(pwd)")_cache",target=/target \
  --mount type=volume,source=registry_cache,target=/usr/local/cargo/registry \
  cosmwasm/optimizer:0.16.0

The downside is that to verify one contract in the workspace, you need to compile them all, but the majority of the build time is in dependencies, which are shared and cached between the various contracts and thus the time is sub-linear with respect to number of contracts.

Contracts excluded from Workspace

This is designed for cosmwasm samples. You cannot provide automatic verification for these

If you have a more complex build environment, you need to pass a few more arguments to define how to run the build process.

cosmwasm has a root workspace and many contracts under ./contracts/*, which are excluded in the top-level Cargo.toml. In this case, we compile each contract separately with it's own cache. However, since they may refer to packages via path (../../packages/std), we need to run the script in the repo root. In this case, we can use the optimize.sh command:

docker run --rm -v "$(pwd)":/code \
  --mount type=volume,source="devcontract_cache_burner",target=/code/contracts/burner/target \
  --mount type=volume,source=registry_cache,target=/usr/local/cargo/registry \
  cosmwasm/optimizer:0.16.0 ./contracts/burner

Caches

The build system uses the folder /target in its local file system for all Rust compilation results. This ensures there is no conflict with the target folder of the source repository. It also means that for each compilation, Cargo will have an empty target folder and have to re-compile all contracts and all dependencies. In order to avoid this, you can optionally mount a Docker volume into the file system, as highlighted here:

 docker run --rm -v "$(pwd)":/code \
+  --mount type=volume,source="$(basename "$(pwd)")_cache",target=/target \
   --mount type=volume,source=registry_cache,target=/usr/local/cargo/registry \
   cosmwasm/rust-optimizer:0.15.0

Using this cache is considered best practice and included in all our example call snippets.

Before version 0.13.0, the target folder was located at /code/target. This caused situations in which rust-optimizer/workspace-optimizer wrote to the target folder of the host in an unintended way. By ensuring the target location is not a subfolder of the mounted source code we can avoid those sort of problems.

Development

Take a look at the Makefile You can edit the Dockerfile (in a fork), and run make build to compile it.

Notice

This has been tested on Linux (Ubuntu / Debian). There are currently versions of both optimizers for two processor architectures: Intel/Amd 64-bits, and Arm 64-bits (these run natively on Mac M1 machines).

However, the native Arm version produces different wasm artifacts than the Intel version. Given that that impacts reproducibility, non-Intel images contain a "-arm64" suffix to differentiate them.

Arm images are released to ease development and testing on Mac M1 machines. For release / production use, only contracts built with the Intel optimizers must be used.