github.com/magma/magma Go module

This is the home for all new Golang code for all magma deployment targets.

Goals:

1. Single go.mod defines all dependencies in one place and keeps versions in sync.
2. Code can easily be shared with natual package paths that match file paths. No replace directives necessary in go.mod.
3. Make it easy to find shared libs/patterns for use across magma targets.
4. Other projects can easily import github.com/magma/magma/src/go code for use in derivative works.

Notes:

1. The go.mod is placed at $MAGMA_ROOT/src/go to avoid name collisions with existing Go directories.
2. We eventually want a canonical golang import path of magmacore.org/magma

Bazel integration

Generating Bazel build files

When updating or adding a new dependency, run

bazel run //:gazelle -- update-repos -from_file=src/go/go.mod -to_macro=bazel/go_repositories.bzl%go_repositories

To generate BUILD.bazel files, run

bazel run //:gazelle

Running tests

To run all the tests, run

bazel test //src/go/...

Dependencies

Only pull in dependencies when they are worth their weight in tech debt. Strongly review libraries before adding as a dependency along the following considerations:

1. Does the dependency have a license compatible with BSD-3?
2. How large is the dependency and its transitive dependencies? Weigh the total increase in go.sum against the provided functionality.
3. Does the dependency have good test coverage and is it well maintained?
4. What alternatives were considered?

Add the considerations in this README when adding the dependency to go.mod.

Protobuf

Google/Go authors BSD-3 license.

A large/heavy dependency, but provides a solid, well-tested, and production- hardened IDL. Good semantics for forwards/backwards compatibility and highly performant. Front-runner for de facto standard IDL/serdes library in modern software engineering projects.

Considered alternatives: Thrift, Swagger, JSON Schema.

gRPC

Apache License 2.0

A large/heavy dependency that provides both RPC service IDL and base client/server implementation. Tightly coupled with protobuf.

Considered alternatives: Thrift, Swagger, JSON-RPC

Uber Zap

MIT license.

Medium weight, pulls in a few transitive dependencies, but has a large overlap with testify. Provides scoped/named loggers and parameterized logging while achieving better performance than other libraries proven via benchmarks.

Very well tested: 98% code coverage; performance benchmarks included in CI

Considered alternatives: built-in log, glog, klog

testify

MIT license.

Lightweight, for test purposes only. Good test coverage and widely adopted.

Downside is main usage eschews compile-time type checking in favor of runtime reflection-based equality checking.

Considered alternatives: regular Golang code

GoMock

Apache License 2.0

Lightweight test tool to codegen verifiable mocks from interfaces. Generated EXPECT() functions are type-defined, which engages compile-time checking. Strict call count and ordering expectation. These explicit requirements make GoMock superior to mockery.

Considered alternatives: mockery, hand-coded mock implementations

Dependency Injection

TBD -- manual DI or library e.g. Uber dig/fx

External Dependencies / SDKs

In most cases, we will want to abstract dependencies via a clean interface package. In this context, "clean" means the interface package is defined entirely with Golang built-ins and other clean magma interface packages. i.e. there are no 3rd party library dependencies, except for test helpers (e.g. gomock).