Google protobuf support for Lua

English | 中文

This project offers a C module for Lua (5.1, 5.2, 5.3, 5.4 and LuaJIT) manipulating Google's protobuf protocol, both for version 2 and 3 syntax and semantics. It splits to the lower-level and the high-level parts for different goals.

For converting between binary protobuf data with Lua tables, using pb.load() loads the compiled protobuf schema content (*.pb file) generated by Google protobuf's compiler named protoc and call pb.encode()/pb.decode().

Or use these modules to manipulate the raw wire format in lower-level way:

• pb.slice: a wire format decoding module.
• pb.buffer: a buffer implement that use to encode basic types into protobuf's wire format. It can be used to support streaming decode protobuf data.
• pb.conv: a module converting integers in the protobuf wire format.
• pb.io: a module access stdin/stdout or other files in binary mode.

If you don't want to depend Google's protobuf compiler, protoc.lua is a pure Lua module translating text-based protobuf schema content into the *.pb binary format.

Install

To install, you could just use luarocks:

luarocks install lua-protobuf

If you want to build it from source, just clone the repo and use luarocks:

git clone https://github.com/starwing/lua-protobuf
luarocks make rockspecs/lua-protobuf-scm-1.rockspec

If you don't have luarocks, use hererocks to install Lua and luarocks:

pip install hererocks
git clone https://github.com/starwing/lua-protobuf
hererocks -j 2.0 -rlatest .
bin/luarocks make lua-protobuf/rockspecs/lua-protobuf-scm-1.rockspec CFLAGS="-fPIC -Wall -Wextra" LIBFLAGS="-shared"
cp protoc.lua pb.so ..

Or you can build it by hand, it only has a pure Lua module protoc.lua and a pair of C source: pb.h and pb.c. Notice that in order to build the pb C module, you need Lua header file and/or libary file installed. replace $LUA_HEADERS and $LUA_LIBS below to real install locations.

To build it on macOS, use your favor compiler:

gcc -O2 -shared -undefined dynamic_lookup -I "$LUA_HEADERS" pb.c -o pb.so

On Linux, use the nearly same command:

gcc -O2 -shared -fPIC -I "$LUA_HEADERS" pb.c -o pb.so

On Windows, you could use MinGW or MSVC, create a *.sln project or build it on the command line (notice the Lua_BUILD_AS_DLL flag):

cl /O2 /LD /Fepb.dll /I "$LUA_HEADERS" /DLUA_BUILD_AS_DLL pb.c "$LUA_LIBS"

Example

local pb = require "pb"
local protoc = require "protoc"

-- load schema from text (just for demo, use protoc.new() in real world)
assert(protoc:load [[
   message Phone {
      optional string name        = 1;
      optional int64  phonenumber = 2;
   }
   message Person {
      optional string name     = 1;
      optional int32  age      = 2;
      optional string address  = 3;
      repeated Phone  contacts = 4;
   } ]])

-- lua table data
local data = {
   name = "ilse",
   age  = 18,
   contacts = {
      { name = "alice", phonenumber = 12312341234 },
      { name = "bob",   phonenumber = 45645674567 }
   }
}

-- encode lua table data into binary format in lua string and return
local bytes = assert(pb.encode("Person", data))
print(pb.tohex(bytes))

-- and decode the binary data back into lua table
local data2 = assert(pb.decode("Person", bytes))
print(require "serpent".block(data2))

Use case

Usage

protoc Module

To parse a text schema content, create a compiler instance first:

local p = protoc.new()

Then, set some options to the compiler, e.g. the unknown handlers:

-- set some hooks
p.unknown_import = function(self, module_name) ... end
p.unknown_type   = function(self, type_name) ... end
-- ... and options
p.include_imports = true

The unknown_import and unknown_type handle could be true, string or a function. Seting it to true means all non-exist modules and types are given a default value without triggering an error; A string means a Lua pattern that indicates whether an unknown module or type should raise an error, e.g.

p.unknown_type = "Foo.*"

means all types prefixed by Foo will be treat as existing type and do not trigger errors.

If these are functions, the unknown type and module name will be passed to functions. For module handler, it should return a DescriptorProto Table produced by p:load() functions, for type handler, it should return a type name and type, such as message or enum, e.g.

function p:unknown_import(name)
  -- if can not find "foo.proto", load "my_foo.proto" instead
  return p:parsefile("my_"..name)
end

function p:unknown_type(name)
  -- if cannot find "Type", treat it as ".MyType" and is a message type return ".My"..name, "message"
end

After setting options, use load() or compile() or parse() function to get result.

pb Module

pb module has high-level routines to manipulate protobuf messages.

In below table of functions, we have several types that have special means:

• type: a string that indicates the protobuf message type, ".Foo" means the type in a proto definition that has not package statement declared. "foo.Foo" means the type in a proto definition that declared package foo;

• data: could be string, pb.Slice value or pb.Buffer value.

• iterator: a function that can use in Lua for in statement, e.g.

  for name in pb.types() do
    print(name)
  end

NOTICE: Only pb.load() returns error on failure, do check the result it returns. Other routines raise a error when failure for convenience.

Schema loading

pb.load() accepts the schema binary data and returns a boolean indicates the result of loading, success or failure, and a offset reading in schema so far that is useful to figure out the reason of failure.

Type mapping

Type Information

Using pb.(type|field)[s]() functions retrieve type information for loaded messages.

pb.type() returns multiple informations for specified type:

• name : the full qualifier name of type, e.g. ".package.TypeName"
• basename: the type name without package prefix, e.g. "TypeName"
• "map" | "enum" | "message": whether the type is a map_entry type, enum type or message type.

pb.types() returns a iterators, behavior like call pb.type() on every types of all messages.

print(pb.type "MyType")

-- list all types that loaded into pb
for name, basename, type in pb.types() do
  print(name, basename, type)
end

pb.field() returns information of the specified field for one type:

• name: the name of the field
• number: number of field in the schema
• type: field type
• default value: if no default value, nil
• "packed"|"repeated"| "optional": label of the field, optional or repeated, required is not supported
• [oneof_name, oneof_index]: if this is a oneof field, this is the oneof name and index

And pb.fields() iterates all fields in a message:

print(pb.field("MyType", "the_first_field"))

-- notice that you needn't receive all return values from iterator
for name, number, type in pb.fields "MyType" do
  print(name, number, type)
end

pb.enum() maps from enum name and value:

protoc:load [[
enum Color { Red = 1; Green = 2; Blue = 3 }
]]
print(pb.enum("Color", "Red")) --> 1
print(pb.enum("Color", 2)) --> "Green"

Default Values

Using pb.defaults() to get or set a table with all default values from a message. this table will be used as the metatable of the corresponding decoded message table when setting use_default_metatable option.

You could also call pb.defaults with "*map" or "*array" to get the default metatable for map and array when decode a message. This settings will bepass use_default_metatable option.

To clear a default metatable, just pass nil as second argument to pb.defaults().

   check_load [[
      message TestDefault {
         optional int32 defaulted_int = 10 [ default = 777 ];
         optional bool defaulted_bool = 11 [ default = true ];
         optional string defaulted_str = 12 [ default = "foo" ];
         optional float defaulted_num = 13 [ default = 0.125 ];
      } ]]
   print(require "serpent".block(pb.defaults "TestDefault"))
-- output:
-- {
--   defaulted_bool = true,
--   defaulted_int = 777,
--   defaulted_num = 0.125,
--   defaulted_str = "foo"
-- } --[[table: 0x7f8c1e52b050]]

Hooks

If set pb.option "enable_hooks", the hook function will be enabled. you could use pb.hook() and pb.encode_hook to set or get a decode or encode hook function, respectively: call it with type name directly get current setted hook; call it with two arguments to set a hook; and call it with nil as the second argument to remove the hook. in all case, the original one will be returned.

After the hook function setted and hook enabled, the decode function will be called after a message get decoded and encode functions will be called before the message is encoded. So you could get all values in the table passed to hook function. That's the only argument of hook.

If you need type name in hook functions, use this helper:

local function make_hook(name, func)
  return pb.hook(name, function(t)
    return func(name, t)
  end)
end

Options

Setting options to change the behavior of other routines. These options are supported currently:

Note: The string returned by int64_as_string or int64_as_hexstring will prefix a '#' character. Because Lua may convert between string with number, prefix a '#' makes Lua return the string as-is.

all routines in all module accepts '#' prefix string/hex string as arguments regardless of the option setting.

Multiple State

pb module support multiple states. A state is a database that contains all type information of registered messages. You can retrieve current state by pb.state(), or set new state by pb.state(newstate).

Use pb.state(nil) to discard current state, but not to set a new one (the following routines call that use the state will create a new default state automatedly). Use pb.state() to retrieve current state without setting a new one. e.g.

local old = pb.state(nil)
-- if you use protoc.lua, call protoc.reload() here.
assert(pb.load(...))
-- do someting ...
pb.state(old)

Notice that if you use protoc.lua module, it will register some message to the state, so you should call proto.reload() after setting a new state.

pb.io Module

pb.io module reads binary data from a file or stdin/stdout, pb.io.read() reads binary data from a file, or stdin if no file name given as the first parameter.

pb.io.write() and pb.io.dump() are same as Lua's io.write() except they write binary data. the former writes data to stdout, and the latter writes data to a file specified by the first parameter as the file name.

All these functions return a true value when success, and return nil, errmsg when an error occurs.

pb.conv Module

pb.conv provide functions to convert between numbers.

pb.slice Module

Slice object parse binary protobuf data in a low-level way. Use slice.new() to create a slice object, with the optional offset i and j to access a subpart of the original data (named a view).

As protobuf usually nest sub message with in a range of slice, a slice object has a stack itself to support this. Calling s:enter(i, j) saves current position and enters next level with the optional offset i and j just as slice.new(). calling s:leave() restore the prior view. s:level() returns the current level, and s:level(n) returns the current position, the start and the end position information of the nth level. calling s:enter() without parameter will read a length delimited type value from the slice and enter the view in reading value. Using #a to get the count of bytes remains in current view.

local s = slice.new("<data here>")
local tag = s:unpack "v"
if tag%8 == 2 then -- tag has a type of string/bytes? maybe it's a sub-message.
  s:enter() -- read following bytes value, and enter the view of bytes value.
  -- do something with bytes value, e.g. reads a lot of fixed32 integers from bytes.
  local t = {}
  while #s > 0 do
    t[#t+1] = s:unpack "d"
  end
  s:leave() -- after done, leave bytes value and ready to read next value.
end

To read values from slice, use slice.unpack(), it use a format string to control how to read into a slice as below table (same format character are also used in buffer.pack()). Notice that you can use pb.typefmt() to convert between format and protobuf type names (returned from pb.field()).

And extra format can be used to control the read cursor in one slice.unpack() process:

e.g. If you want to read a varint value twice, you can write it as:

local v1, v2 = s:unpack("v*v", 1)
-- v: reads a `varint` value
-- *: receive the second parameter 1 and set it to the current cursor position, i.e. restore the cursor to the head of the view
-- v: reads the first `varint` value again

All routines in pb.slice module:

pb.buffer Module

Buffer module used to construct a protobuf data format stream in a low-level way. It's just a bytes data buffer. using buffer.pack() to append values to the buffer, and buffer.result() to get the encoded raw data, or buffer.tohex() to get the human-readable hex digit value of data.

buffer.pack() use the same format syntax with slice.unpack(), and support '()' format means the inner value will be encoded as a length delimited value, i.e. a message value encoded format.

parenthesis can be nested.

e.g.

b:pack("(vvv)", 1, 2, 3) -- get a bytes value that contains three varint value.

buffer.pack() also support '#' format, it means prepends a length into the buffer.

e.g.

b:pack("#", 5) -- prepends a varint length #b-5+1 at offset 5

All routines in pb.buffer module: