Add full Go Support

[djgilcrease]

Description of Changes

Architecture decisions

• Code generation over reflection — Go's reflect package is not available in WASM. The stdb-gen tool (https://gitlab.com/digitalxero/stdb-gen) reads //stdb: directives and generates all serialization, table accessors, and module definitions at compile time.
• //go:wasmexport — Requires Go 1.24+. Exports are in server/runtime/exports.go behind //go:build wasip1.
• Panic-based error handling in server — Insert/Delete/Update operations panic on error (matching Rust SDK convention), caught by recover() in the WASM call reducer wrapper.
• WASI stubs — Go's wasip1 target requires WASI imports. Minimal stubs redirect stdout/stderr to the host logger and stub out clock/args/environ.

Summary

This PR adds first-class Go language support to SpacetimeDB, including a complete server-side WASM module SDK, a WebSocket client SDK, Rust-side codegen for Go bindings, CLI build toolchain integration, project templates, and a comprehensive test suite covering all 66 integration tests.

Highlights

• Server SDK (sdks/go/server/) — Write SpacetimeDB modules in Go, compiled to WASM via GOOS=wasip1 GOARCH=wasm with //go:wasmexport (Go 1.24+). Supports tables, reducers, procedures, views, event tables, scheduled reducers, lifecycle hooks, row-level security, and HTTP from procedures.
• Client SDK (sdks/go/client/) — Channel-based WebSocket client with BSATN protocol v2, brotli compression, client-side table cache (xsync lock-free maps), subscriptions, and reducer callbacks.
• BSATN library (sdks/go/bsatn/) — Complete binary serialization matching the SpacetimeDB wire format (LE integers, u32-prefixed strings/arrays, u8 sum tags, sequential product fields).
• Code generator (stdb-gen) — Replaces reflection with compile-time code generation via //stdb: directives. Generates type-safe table accessors, BSATN serialization, index methods, and module definition.
• Rust codegen (crates/codegen/src/go.rs) — Implements the Lang trait for generating Go client bindings from module schemas.
• CLI integration (crates/cli/src/tasks/go.rs) — spacetime build and spacetime init support for Go modules.
• WASI stubs (crates/core/src/host/wasmtime/wasi_stubs.rs) — Minimal WASI Preview 1 shims (fd_write for stdout/stderr, clock, args, environ) to support Go's wasip1 runtime.
• Templates — basic-go and chat-console-go starter projects for spacetime init.
• All 66 integration tests pass — 50 core SDK tests, 7 procedure tests, 6 view tests, 3 event table tests.

Templates (templates/)

• basic-go/ — Minimal starter with a Person table and add/say_hello reducers
• chat-console-go/ — Chat app with User/Message tables, connect/disconnect lifecycle, and a terminal client

API and ABI breaking changes

None

Expected complexity level and risk

2 - does require go knowlage, but it fully integrates with the existing integration tests and benchmarks
This should be relatively isolated from existing code, except for the updates to the cli to add go language support.

Rust changes (crates/) — 18 files, ~1,900 lines

• crates/codegen/src/go.rs (1,133 lines) — Go client binding codegen implementing Lang trait
• crates/cli/src/tasks/go.rs (123 lines) — Go module build task (stdb-gen → go generate → go build)
• crates/cli/src/subcommands/init.rs — Go template support for spacetime init
• crates/cli/src/subcommands/generate.rs — Go language option for spacetime generate
• crates/core/src/host/wasmtime/wasi_stubs.rs (272 lines) — WASI Preview 1 stubs for Go WASM modules
• crates/testing/ — Go module build support in test harness
• crates/smoketests/ — Go module smoke test

Testing

• All 50 core Go SDK integration tests pass (cargo test -p spacetimedb-sdk -- go::)
• All 7 Go procedure tests pass (cargo test -p spacetimedb-sdk -- go_procedures::)
• All 6 Go view tests pass (cargo test -p spacetimedb-sdk -- go_view::)
• All 3 Go event table tests pass (cargo test -p spacetimedb-sdk -- go_event_table_tests::)
• Go unit tests pass (cd sdks/go && go test ./...)
• Go benchmarks run and are competitive with Rust module performance
• Smoke test verifies Go toolchain detection
• CI workflow validates on PR (This will also need to be updated to tag changes to sdks/go with sdks/go/v#.#.# so that go clients/servers will be able to properly import the code.

Test modules (modules/) — 46 files, ~4,500 lines

Module	Coverage
sdk-test-go/	73 tables, 200+ reducers — full parity with Rust sdk-test module
sdk-test-connect-disconnect-go/	Connect/disconnect lifecycle callbacks
sdk-test-procedure-go/	8 procedures (return values, transactions, panics, HTTP, scheduled, UUID v7)
sdk-test-view-go/	4 views (auth/anonymous, Option/Vec returns)
sdk-test-event-table-go/	Event tables (emit, multiple events, non-persistence)
benchmarks-go/	3 suites: circles (game physics), ia_loop (AI state), synthetic (index strategies)
module-test-go/	Comprehensive: 11 tables, 15 reducers, 5 procedures, 1 view, scoped types
perf-test-go/	Multi-column index performance

Benchmark Results: Rust vs Go

All benchmarks run on the same machine. Circles and IA Loop use debug mode; Index Scans use release mode.
Go modules built with: GOOS=wasip1 GOARCH=wasm go build -buildmode=c-shared

Circles Benchmark (Debug mode)

Operation	Rows	Rust	Go	Go/Rust
insert_bulk_food	50	459µs	517µs	1.13x
insert_bulk_entity	50	207µs	198µs	0.96x
insert_bulk_circle	500	1.302ms	979µs	0.75x
cross_join_circle_food	25,000	7.799ms	9.993ms	1.28x
cross_join_all	1,250,000	331.984ms	263.614ms	0.79x

IA Loop Benchmark (Debug mode)

Operation	Rows	Rust	Go	Go/Rust
insert_bulk_position	20,000	48.872ms	25.778ms	0.53x
insert_bulk_velocity	10,000	19.615ms	10.891ms	0.56x
update_position_all	20,000	81.697ms	42.038ms	0.51x
update_position_with_velocity	10,000	63.440ms	85.632ms	1.35x
insert_world	5,000	71.591ms	43.645ms	0.61x
game_loop_enemy_ia	100	184.563ms	88.021ms	0.48x

Index Scans (Release mode, 1.2M rows)

Index	Rust	Go	Go/Rust
{id} (unique PK)	27.6µs	55.4µs	2.01x
{chunk} (btree)	75.5µs	54.0µs	0.72x
{x, z, dimension} (3-col btree)	17.1µs	47.5µs	2.78x
{x, z} (2-col prefix)	90.7µs	61.6µs	0.68x
load_location_table (1.2M inserts)	1.961s	2.629s	1.34x

Key Takeaways

• Go wins most bulk insert/update workloads (often ~2x faster in debug mode)
• Go wins the massive cross_join_all (1.25M rows) by 21%
• Mixed results on index scans — Go wins on btree range scans, Rust wins on unique PK lookups and multi-column exact matches
• Values bolded where Go is faster (ratio < 1.0)

[CLAassistant]

All committers have signed the CLA.