#tarantool-protobuf

A protoc plugin and pure-Lua runtime for using Protocol Buffers (proto3) and gRPC service stubs from Tarantool.

Tarantool ships an in-tree require('protobuf') module, but it is encode-only and has no support for map, oneof, services, or any decode path. This project fills those gaps with:

• protoc-gen-tarantool — a protoc plugin (Go) that turns .proto files into Lua modules.
• runtime/pb — a pure Lua + LuaJIT-FFI runtime the generated code uses for the wire format. Named pb rather than protobuf to avoid colliding with Tarantool's built-in module.

#Status

Proto3 conformance is closed: every Required.* and Recommended.* test in both the binary+JSON and text-format suites passes.

#Install

Until the repository is published on its canonical remote, the source.url in tarantool-protobuf-scm-1.rockspec is aspirational and tt rocks install <url> will fail. Install from a local checkout instead:

git clone <this-repo> && cd tarantool-protobuf
tt rocks make tarantool-protobuf-scm-1.rockspec

That puts the pb.* runtime modules under .rocks/share/tarantool/. The Go plugin still has to be built separately — see "Quick start" below.

#Quick start

# 1. Build the plugin and generate the example.
just gen

# 2. Run the round-trip test in Tarantool.
just test

The plugin emits one .lua file per .proto. By default the output path mirrors the proto package (package foo.bar; baz.proto → foo/bar/baz_pb.lua), required as foo.bar.baz_pb. Two ways to override:

// 1. Per-file, via a proto option:
import "tarantool/tarantool.proto";
option (tarantool.lua_package) = "myapp.proto.foo";

# 2. Plugin-wide, via the `prefix=` plugin parameter. Every generated
#    module is prepended with this namespace and cross-file imports rewrite
#    to match. Equivalent to applying `option (tarantool.lua_package)` to
#    every input file, but without touching the .proto.
protoc --tarantool_out=out \
       --tarantool_opt=prefix=myapp.proto \
       file.proto
# -> out/myapp/proto/<pkg>/<file>_pb.lua, required as
#    "myapp.proto.<pkg>.<file>_pb"

prefix= and (tarantool.lua_package) compose: when both are set, the prefix is prepended to the option's value. See docs/reference/cli.md for the full mapping.

For a full walk-through that takes a fresh .proto to a Tarantool process encoding and decoding it, see docs/howto/01-first-message.md.

#Vendoring an upstream .proto schema

If you're vendoring someone else's .proto into your project (etcd, prometheus, opentelemetry, pprof, …), protoc-gen-tarantool plus a small preprocessor is the typical path.

Upstream protos commonly import annotation extensions that only the original generator consumes — versionpb, google.api, gogoproto, grpc.gateway.protoc_gen_openapiv2. Mainline protoc won't parse a file with an unresolved import, so the choice is between vendoring the extension .proto files (lots of additional surface, no wire effect) or stripping the imports and their attached options before generating. Stripping is the lower-cost path — these annotations affect nothing on the wire.

A drop-in preprocessor (one python3 script, no dependencies) should drop import "versionpb/..."; / google/api/... / gogo.proto / protoc-gen-openapiv2/... lines, drop single-line and brace-balanced option (foo.bar) = ...; blocks at file/message/field scope, and drop inline field options [(foo.bar) = "..."]. Reference: tarantool-etcd's proto/_strip_annotations.py (~60 lines).

The same preprocessor is also where you rewrite cross-package imports to a flat layout: e.g. import "etcd/api/mvccpb/kv.proto" → import "mvccpb/kv.proto", so a single protoc -I proto resolves every file without mirroring the upstream subdirectory tree.

Putting it together:

mkdir -p proto/<pkg>
for f in upstream/<path>/*.proto; do
    python3 strip_annotations.py < "$f" > proto/<pkg>/"$(basename "$f")"
done
protoc -I proto --tarantool_out=prefix=myapp.proto:out $(find proto -name '*.proto')

One pitfall: this plugin is proto3-only. If an upstream schema mixes proto2 and proto3, either skip the proto2 files (provided your proto3 side doesn't import them) or wait for proto2 support — see PLAN.md's deferred non-goals.

#Generated API

For each message Foo the plugin emits:

local M = require('myapp.proto.foo')

M.Foo_descriptor       -- the descriptor table consumed by the runtime
M.Foo_new(t)           -- returns t (or {}); placeholder for future validation
M.Foo_encode(t)        -- table -> wire bytes (string)
M.Foo_decode(b)        -- wire bytes (string) -> table
M.Foo_decode_lazy(b)   -- wire bytes -> MessageView (zero-copy view)
M.Foo_text(t, opts)    -- table -> protoc-style text format (debug printer)
M.Foo_has_<field>(t)   -- only emitted for proto3 explicit-`optional` fields
M.Foo_clear_<field>(t) -- same
M.Foo_fields           -- strict {field = "field", ...} for lazy-view callers
M.Foo_oneofs           -- strict {oneof_group = "oneof_group", ...}, when any

Lazy-view field-name arguments should be routed through M.Foo_fields / M.Foo_oneofs rather than passed as string literals — typos error at the read site instead of silently returning nil. See docs/api-modes.md.

Text-format decoding is exposed on the runtime as pb.text.decode(desc, text, opts) (no per-message wrapper — it's used from a few places, like the conformance runner, and didn't warrant codegen surface).

For each enum Color:

M.Color_descriptor          -- { name, by_name, by_value }
M.Color                     -- alias for by_name: M.Color.RED -> 0

Repeated fields are Lua arrays (1-based, contiguous). 64-bit integers (int64, uint64, fixed64, sfixed64, sint64) are LuaJIT int64_t / uint64_t cdata — lossless and the same convention used by Tarantool's net.box, msgpack, and built-in protobuf modules.

#Migrating from a Lua proto library that auto-down-casts int64

If you're moving from a library that hands back Lua numbers (silently losing precision past 2^53), expect a sweep wherever a cdata value crosses into a primitive that doesn't accept it. The four patterns that catch out every migrator:

local id = msg.user_id  -- cdata: uint64_t

-- 1. log / printf format verbs: %d on cdata raises an error.
log.info('user %d signed in', tonumber(id))

-- 2. numeric for-loop bounds: `for i = 1, n` requires a Lua number.
for i = 1, tonumber(msg.row_count) do ... end

-- 3. string.format with %d / %x: same as log.
local hex = string.format('%016x', tonumber(id))

-- 4. table keys: cdata is hashed by identity, not value, so two
--    distinct cdata for the same number won't collide. Either convert
--    to number (if it fits) or use tostring(id) as the key.
cache[tonumber(id)] = row

box.tuple / net.box / msgpack / Tarantool's protobuf all accept cdata int64 directly — those paths don't need a tonumber(). The boundary is Lua primitives that expect a number. Past 2^53 (≈ 9e15), tonumber() silently truncates; if your IDs can be that large, keep them as cdata or stringify with tostring(id):gsub('U?LL$', '').

Three API modes live side-by-side. Same descriptor, three call shapes — the inline (full) generated API is the default, the descriptor-driven runtime API is for dynamic schemas, and the lazy API is a zero-copy view for sparse reads and proxy / router workloads. See docs/api-modes.md for when to pick which, with measured trade-offs.

#Layout

cmd/
  protoc-gen-tarantool/      Go plugin (the codegen)
    main.go                  reads CodeGeneratorRequest, hands off to gen
    internal/gen/            per-message emission for both modes
  protoc-gen-tarantool-doc/  separate Go plugin that emits Markdown docs
  conformance/               Lua conformance dispatch (loaded by the runner)
  conformance-runner.lua     stdin/stdout framing for `conformance_test_runner`

runtime/pb/                  pure-Lua runtime (`require('pb')`)
  init.lua                   public surface
  wire.lua                   varint / zigzag / fixed / float / LEN primitives
  codec.lua                  descriptor-driven encode/decode
  lazy.lua                   zero-copy MessageView / ArrayView / MapView
  text.lua                   text format encode + decode (proto3)
  json.lua                   proto3 JSON encode + decode (strict)
  wkt.lua                    Timestamp / Duration / Empty / Wrappers /
                             Struct / Value / ListValue / Any / FieldMask
  grpc.lua                   transport interface + loopback / multiplex
  parser.lua                 pure-Lua proto3 schema parser (.proto → AST)
  dynamic.lua                AST → descriptor module
  fileset.lua                FileDescriptorSet bytes → descriptor module
  descriptor_pb.lua          hand-built descriptors of descriptor.proto

options/tarantool/           custom proto file options
  tarantool.proto            (tarantool.lua_package) — Lua require path override

examples/proto/              demo .proto inputs
examples/expected/           generated output for full + runtime modes
                             (both committed for inspection + parametrized tests)
test/                        luatest groups, conformance regressions, fixtures
docs/                        codegen notes, API mode comparison, design briefs
bench/                       per-helper bench + JIT-trace gate + alloc baseline

#Documentation

Start at docs/index.md — the documentation map, grouped by what you're trying to do (getting started, reference, specs, internals). The two reference pages worth knowing by name:

• docs/api-modes.md — when to use Foo_encode (full), pb.encode(desc, t) (runtime / reflect), or pb.decode_lazy(desc, b) (lazy view). Measured allocation + throughput trade-offs.
• docs/codegen.md — plugin internals, descriptor shape contract, how to add a new wire type or scalar, the LuaJIT hot-path rules generated code observes.

Roadmap is in PLAN.md; cross-codebase invariants in CLAUDE.md.

#Conformance

cmd/conformance-runner.lua speaks the Google protobuf conformance protocol on stdin/stdout. Drive it with the canonical conformance_test_runner binary like so:

just gen
conformance_test_runner --enforce_recommended \
    tarantool cmd/conformance-runner.lua

Homebrew's protobuf package does not ship conformance_test_runner, so a Dockerfile under docker/conformance.Dockerfile builds it from upstream protobuf source and bundles Tarantool. Run the full suite locally with:

just conformance

(Mounts the repo into the container — generated Lua from just gen on the host is what gets tested.) Known failures live in test/conformance/known_failures.txt (binary + JSON suite) and test/conformance/known_failures_text.txt (text-format suite); both are empty for the proto3 suites as of 2026-05-16.

Current baseline (2026-05-16, protobuf v34.1):

The 1313 + 18 skipped tests all target protobuf_test_messages.proto2.TestAllTypesProto2, which we don't generate Lua for. Proto2 support is a separate slice — see docs/codegen.md#proto2-deferral for what it would take.

The runner supports protobuf_test_messages.proto3.TestAllTypesProto3 in binary, JSON, and text-format input/output, including the JSON_IGNORE_UNKNOWN_PARSING_TEST category (forwarded as ignore_unknown_fields=true to pb.json.decode). The self-test in test/conformance_test.lua exercises the runner with crafted requests on every just test run.

#Benchmarks

just bench           # print throughput + alloc per op (5 sizes × 2 modes)
just bench-baseline  # overwrite bench/baseline.json (run on a quiet machine)
just bench-compare   # exit 1 if any alloc-per-op regressed >5% vs baseline

The committed bench/baseline.json tracks only allocation per op — that's reproducible across machines because it counts bytes, not time. Throughput in stderr is informational; it swings 30%+ on a contended CPU.

Current baseline (LuaJIT 2.1, hello.Person):

Lazy decode trades a much higher index-build cost (one MessageView table + four SoA arrays) for near-zero allocation on subsequent field reads — best when you touch a small fraction of fields, or when you re-encode mostly-unchanged messages (proxy / router workloads). See docs/api-modes.md for the bench numbers and the cross-over point.

#Why named pb instead of protobuf?

Tarantool's loader prefers the built-in require('protobuf') over any filesystem module of the same name. Trying to override it would break code that uses the built-in's encode API. pb is short, unambiguous, and lives alongside the built-in.

#License

BSD 2-Clause. See LICENSE.