Protocol Buffers - Google's data interchange format (grpc依赖) https://developers.google.com/protocol-buffers/
You can not select more than 25 topics Topics must start with a letter or number, can include dashes ('-') and can be up to 35 characters long.

515 lines
16 KiB

--[[
Code to generate a C API in:
foo.proto -> foo.upb.h
foo.upb.c
This code is evolving very quickly and so there are lots of little things
that aren't perfect right now. As it settles a little more, the code
quality should improve.
--]]
local upb = require "upb"
local dump_cinit = require "dump_cinit"
local export = {}
local typemap = {
[upb.TYPE_BOOL] = "bool",
[upb.TYPE_FLOAT] = "float",
[upb.TYPE_INT32] = "int32_t",
[upb.TYPE_UINT32] = "uint32_t",
[upb.TYPE_DOUBLE] = "double",
[upb.TYPE_INT64] = "int64_t",
[upb.TYPE_UINT64] = "uint64_t",
[upb.TYPE_STRING] = "upb_stringview",
[upb.TYPE_BYTES] = "upb_stringview",
}
function strip_proto(filename)
return string.gsub(filename, '%.proto$','')
end
local function join(...)
return table.concat({...}, ".")
end
local function to_cident(...)
return string.gsub(join(...), "[%./]", "_")
end
local function to_preproc(...)
return string.upper(to_cident(...))
end
-- Strips away last path element, ie:
-- foo.Bar.Baz -> foo.Bar
local function remove_name(name)
local package_end = 0
for i=1,string.len(name) do
if string.byte(name, i) == string.byte(".", 1) then
package_end = i - 1
end
end
return string.sub(name, 1, package_end)
end
local function enum_value_symbol(enumdef, name)
return to_cident(remove_name(enumdef:full_name())) .. "_" .. name
end
local function dump_enum_vals(enumdef, append)
local enum_vals = {}
for k, v in enumdef:values() do
enum_vals[#enum_vals + 1] = {k, v}
end
table.sort(enum_vals, function(a, b) return a[2] < b[2] end)
-- protobuf convention is that enum values are scoped at the level of the
-- enum itself, to follow C++. Ie, if you have the enum:
-- message Foo {
-- enum E {
-- VAL1 = 1;
-- VAL2 = 2;
-- }
-- }
--
-- The name of VAL1 is Foo.VAL1, not Foo.E.VAL1.
--
-- This seems a bit sketchy, but people often name their enum values
-- accordingly, ie:
--
-- enum Foo {
-- FOO_VAL1 = 1;
-- FOO_VAL2 = 2;
-- }
--
-- So if we don't respect this also, we end up with constants that look like:
--
-- GOOGLE_PROTOBUF_FIELDDESCRIPTORPROTO_TYPE_TYPE_DOUBLE = 1
--
-- (notice the duplicated "TYPE").
local cident = to_cident(remove_name(enumdef:full_name()))
for i, pair in ipairs(enum_vals) do
k, v = pair[1], pair[2]
append(' %s = %d', enum_value_symbol(enumdef, k), v)
if i == #enum_vals then
append('\n')
else
append(',\n')
end
end
end
local function field_default(field)
if field:type() == upb.TYPE_MESSAGE then
return "NULL"
elseif field:type() == upb.TYPE_STRING or
field:type() == upb.TYPE_BYTES then
local default = field:default() or ""
return string.format('upb_stringview_make("%s", strlen("%s"))', field:default(), field:default())
elseif field:type() == upb.TYPE_ENUM then
return enum_value_symbol(field:subdef(), field:default())
else
return field:default();
end
end
local function ctype(field)
if field:label() == upb.LABEL_REPEATED then
return "upb_array*"
elseif field:type() == upb.TYPE_MESSAGE then
if field:containing_type():file() == field:subdef():file() then
return to_cident(field:subdef():full_name()) .. "*"
else
return "struct " .. to_cident(field:subdef():full_name()) .. "*"
end
elseif field:type() == upb.TYPE_ENUM then
return to_cident(field:subdef():full_name())
else
return typemap[field:type()] or "void*"
end
end
local function emit_file_warning(filedef, append)
append('/* This file was generated by upbc (the upb compiler) from the input\n')
append(' * file:\n')
append(' *\n')
append(' * %s\n', filedef:name())
append(' *\n')
append(' * Do not edit -- your changes will be discarded when the file is\n')
append(' * regenerated. */\n\n')
end
local function field_layout_rank(field)
-- Order:
-- 1, 2, 3. primitive fields (8, 4, 1 byte)
-- 4. string fields
-- 5. submessage fields
-- 6. repeated fields
--
-- This has the following nice properties:
--
-- 1. padding alignment is (nearly) minimized.
-- 2. fields that might have defaults (1-4) are segregated
-- from fields that are always zero-initialized (5-7).
--
-- We skip oneof fields, because they are emitted in a separate pass.
local rank
if field:containing_oneof() then
rank = 100 -- These go last (actually we skip them).
elseif field:label() == upb.LABEL_REPEATED then
rank = 6
elseif field:type() == upb.TYPE_MESSAGE then
rank = 5
elseif field:type() == upb.TYPE_STRING or field:type() == upb.TYPE_BYTES then
rank = 4
elseif field:type() == upb.TYPE_BOOL then
rank = 3
elseif field:type() == upb.TYPE_FLOAT or
field:type() == upb.TYPE_INT32 or
field:type() == upb.TYPE_UINT32 then
rank = 2
else
rank = 1
end
-- Break ties with field number.
return (rank * 2^29) + field:number()
end
local function has_hasbit(field)
if field:containing_type():file():syntax() == upb.SYNTAX_PROTO2 then
return field:label() ~= upb.LABEL_REPEATED and not field:containing_oneof()
else
return false
end
end
local function write_h_file(filedef, append)
emit_file_warning(filedef, append)
local basename_preproc = to_preproc(filedef:name())
append('#ifndef %s_UPB_H_\n', basename_preproc)
append('#define %s_UPB_H_\n\n', basename_preproc)
append('#include "upb/msg.h"\n\n')
append('UPB_BEGIN_EXTERN_C\n\n')
for msg in filedef:defs(upb.DEF_MSG) do
-- TODO(haberman): forward declare C++ type names so we can use
-- UPB_DECLARE_TYPE().
local msgname = to_cident(msg:full_name())
append('struct %s;\n', msgname)
append('typedef struct %s %s;\n', msgname, msgname)
end
append("/* Enums */\n\n")
for _, def in ipairs(sorted_defs(filedef:defs(upb.DEF_ENUM))) do
local cident = to_cident(def:full_name())
append('typedef enum {\n')
dump_enum_vals(def, append)
append('} %s;\n\n', cident)
end
for msg in filedef:defs(upb.DEF_MSG) do
local msgname = to_cident(msg:full_name())
append('/* %s message definition. */\n', msgname)
append('extern const upb_msglayout_msginit_v1 %s_msginit;\n', msgname)
append('%s *%s_new(upb_env *env);\n', msgname, msgname)
append('%s *%s_parsenew(upb_stringview buf, upb_env *env);\n',
msgname, msgname)
append('char *%s_serialize(%s *msg, upb_env *env, size_t *len);\n',
msgname, msgname)
append('void %s_free(%s *msg, upb_env *env);\n', msgname, msgname)
append('\n')
append('/* %s getters. */\n', msgname)
local setters, get_setters = dump_cinit.str_appender()
for field in msg:fields() do
local fieldname = to_cident(field:name())
if field:type() == upb.TYPE_MESSAGE and
field:subdef():file() ~= filedef then
-- Forward declaration for message type declared in another file.
append('struct %s;\n', to_cident(field:subdef():full_name()))
end
if field:label() == upb.LABEL_REPEATED then
else
local typename = ctype(field)
append('%s %s_%s(const %s *msg);\n',
typename, msgname, fieldname, msgname)
setters('void %s_set_%s(%s *msg, %s value);\n',
msgname, fieldname, msgname, typename)
end
end
for oneof in msg:oneofs() do
local fullname = to_cident(oneof:containing_type():full_name() .. "." .. oneof:name())
append('typedef enum {\n')
for field in oneof:fields() do
append(' %s = %d,\n', fullname .. "_" .. field:name(), field:number())
end
append(' %s_NOT_SET = 0,\n', fullname)
append('} %s_oneofcases;\n', fullname)
append('%s_oneofcases %s_case(const %s *msg);\n', fullname, fullname, msgname)
end
append('\n')
append('/* %s setters. */\n', msgname)
append(get_setters())
append('\n')
append('\n')
end
append('UPB_END_EXTERN_C')
append('\n')
append('\n')
append('#endif /* %s_UPB_H_ */\n', basename_preproc)
end
local function write_c_file(filedef, hfilename, append)
emit_file_warning(filedef, append)
append('#include <stddef.h>\n')
append('#include "upb/decode.h"\n')
append('#include "upb/encode.h"\n')
append('#include "upb/msg.h"\n')
append('#include "upb/upb.h"\n')
append('#include "%s"\n\n', hfilename)
for dep in filedef:dependencies() do
local outbase = strip_proto(dep:name())
append('#include "%s.upb.h"\n', outbase)
end
append('\n')
for msg in filedef:defs(upb.DEF_MSG) do
local msgname = to_cident(msg:full_name())
local fields_array_ref = "NULL"
local submsgs_array_ref = "NULL"
local oneofs_array_ref = "NULL"
local field_count = 0
local submsg_count = 0
local submsg_set = {}
local submsg_indexes = {}
local hasbit_count = 0
local hasbit_indexes = {}
local oneof_count = 0
local oneof_indexes = {}
-- Create a layout order for oneofs.
local oneofs_layout_order = {}
for oneof in msg:oneofs() do
table.insert(oneofs_layout_order, oneof)
end
table.sort(oneofs_layout_order, function(a, b)
return a:name() < b:name()
end)
for _, oneof in ipairs(oneofs_layout_order) do
oneof_indexes[oneof] = oneof_count
oneof_count = oneof_count + 1
end
-- Create a layout order for fields. We use this order for the struct and
-- for offsets, but our list of fields we keep in field number order.
local fields_layout_order = {}
for field in msg:fields() do
table.insert(fields_layout_order, field)
end
table.sort(fields_layout_order, function(a, b)
return field_layout_rank(a) < field_layout_rank(b)
end)
-- Another sorted array in field number order.
local fields_number_order = {}
for field in msg:fields() do
table.insert(fields_number_order, field)
end
table.sort(fields_number_order, function(a, b)
return a:number() < b:number()
end)
append('struct %s {\n', msgname)
-- Non-oneof fields.
for _, field in ipairs(fields_layout_order) do
field_count = field_count + 1
if field:type() == upb.TYPE_MESSAGE then
submsg_count = submsg_count + 1
submsg_set[field:subdef()] = true
end
if field:containing_oneof() then
-- Handled below.
else
if has_hasbit(field) then
hasbit_indexes[field] = hasbit_count
hasbit_count = hasbit_count + 1
end
append(' %s %s;\n', ctype(field), field:name())
end
end
-- Oneof fields.
for oneof in msg:oneofs() do
local fullname = to_cident(oneof:containing_type():full_name() .. "." .. oneof:name())
append(' union {\n')
for field in oneof:fields() do
append(' %s %s;\n', ctype(field), field:name())
end
append(' } %s;\n', oneof:name())
append(' %s_oneofcases %s_case;\n', fullname, oneof:name())
end
append('};\n\n')
if oneof_count > 0 then
local oneofs_array_name = msgname .. "_oneofs"
oneofs_array_ref = "&" .. oneofs_array_name .. "[0]"
append('static const upb_msglayout_oneofinit_v1 %s[%s] = {\n',
oneofs_array_name, oneof_count)
for _, oneof in ipairs(oneofs_layout_order) do
append(' {offsetof(%s, %s), offsetof(%s, %s_case)},\n',
msgname, oneof:name(), msgname, oneof:name())
end
append('};\n\n')
end
if submsg_count > 0 then
-- TODO(haberman): could save a little bit of space by only generating a
-- "submsgs" array for every strongly-connected component.
local submsgs_array_name = msgname .. "_submsgs"
submsgs_array_ref = "&" .. submsgs_array_name .. "[0]"
append('static const upb_msglayout_msginit_v1 *const %s[%s] = {\n',
submsgs_array_name, submsg_count)
-- Create a deterministically-sorted array of submessage entries.
local submsg_array = {}
for k, v in pairs(submsg_set) do
table.insert(submsg_array, k)
end
table.sort(submsg_array, function(a, b)
return a:full_name() < b:full_name()
end)
for i, submsg in ipairs(submsg_array) do
append(' &%s_msginit,\n', to_cident(submsg:full_name()))
submsg_indexes[submsg] = i - 1
end
append('};\n\n')
end
if field_count > 0 then
local fields_array_name = msgname .. "__fields"
fields_array_ref = "&" .. fields_array_name .. "[0]"
append('static const upb_msglayout_fieldinit_v1 %s[%s] = {\n',
fields_array_name, field_count)
for _, field in ipairs(fields_number_order) do
local submsg_index = "UPB_NO_SUBMSG"
local oneof_index = "UPB_NOT_IN_ONEOF"
if field:type() == upb.TYPE_MESSAGE then
submsg_index = submsg_indexes[field:subdef()]
end
if field:containing_oneof() then
oneof_index = oneof_indexes[field:containing_oneof()]
end
append(' {%s, offsetof(%s, %s), %s, %s, %s, %s, %s},\n',
field:number(),
msgname,
(field:containing_oneof() and field:containing_oneof():name()) or field:name(),
hasbit_indexes[field] or "UPB_NO_HASBIT",
oneof_index,
submsg_index,
field:descriptor_type(),
field:label())
end
append('};\n\n')
end
append('const upb_msglayout_msginit_v1 %s_msginit = {\n', msgname)
append(' %s,\n', submsgs_array_ref)
append(' %s,\n', fields_array_ref)
append(' %s,\n', oneofs_array_ref)
append(' NULL, /* TODO. default_msg */\n')
append(' UPB_ALIGNED_SIZEOF(%s), %s, %s, %s, %s\n',
msgname, field_count,
0, -- TODO: oneof_count
'false', -- TODO: extendable
msg:file():syntax() == upb.SYNTAX_PROTO2
)
append('};\n\n')
append('%s *%s_new(upb_env *env) {\n', msgname, msgname)
append(' %s *msg = upb_env_malloc(env, sizeof(*msg));\n',
msgname)
append(' memset(msg, 0, sizeof(*msg)); /* TODO: defaults */\n')
append(' return msg;\n')
append('}\n')
append('%s *%s_parsenew(upb_stringview buf, upb_env *env) {\n',
msgname, msgname)
append(' %s *msg = %s_new(env);\n', msgname, msgname)
append(' if (upb_decode(buf, msg, &%s_msginit, env)) {\n', msgname)
append(' return msg;\n')
append(' } else {\n')
append(' return NULL;\n')
append(' }\n')
append('}\n')
append('char *%s_serialize(%s *msg, upb_env *env, size_t *size) {\n',
msgname, msgname)
append(' return upb_encode(msg, &%s_msginit, env, size);\n', msgname)
append('}\n')
for field in msg:fields() do
local typename = ctype(field)
append('%s %s_%s(const %s *msg) {\n',
typename, msgname, field:name(), msgname);
if field:containing_oneof() then
local oneof = field:containing_oneof()
append(' return msg->%s_case == %s ? msg->%s.%s : %s;\n',
oneof:name(), field:number(), oneof:name(), field:name(),
field_default(field))
else
append(' return msg->%s;\n', field:name())
end
append('}\n')
append('void %s_set_%s(%s *msg, %s value) {\n',
msgname, field:name(), msgname, typename);
if field:containing_oneof() then
local oneof = field:containing_oneof()
append(' msg->%s.%s = value;\n', oneof:name(), field:name())
append(' msg->%s_case = %s;\n', oneof:name(), field:number())
else
append(' msg->%s = value;\n', field:name())
end
append('}\n')
end
for oneof in msg:oneofs() do
local fullname = to_cident(oneof:containing_type():full_name() .. "." .. oneof:name())
append('%s_oneofcases %s_case(const %s *msg) {\n', fullname, fullname, msgname)
append(' return msg->%s_case;\n', oneof:name())
append('}\n')
end
end
end
function export.write_gencode(filedef, hfilename, append_h, append_c)
write_h_file(filedef, append_h)
write_c_file(filedef, hfilename, append_c)
end
return export