[ruby/json] JSON.dump: write directly into the provided IO

Ref: https://github.com/ruby/json/issues/524

Rather than to buffer everything in memory.

Unfortunately Ruby doesn't provide an API to write into
and IO without first allocating a string, which is a bit
wasteful.

https://github.com/ruby/json/commit/f017af6c0a
This commit is contained in:
Jean Boussier 2024-11-05 19:57:46 +01:00 committed by Hiroshi SHIBATA
parent 519701657f
commit ee0de3fd4e
5 changed files with 98 additions and 52 deletions

View File

@ -46,9 +46,11 @@ typedef struct FBufferStruct {
unsigned long len; unsigned long len;
unsigned long capa; unsigned long capa;
char *ptr; char *ptr;
VALUE io;
} FBuffer; } FBuffer;
#define FBUFFER_STACK_SIZE 512 #define FBUFFER_STACK_SIZE 512
#define FBUFFER_IO_BUFFER_SIZE (16384 - 1)
#define FBUFFER_INITIAL_LENGTH_DEFAULT 1024 #define FBUFFER_INITIAL_LENGTH_DEFAULT 1024
#define FBUFFER_PTR(fb) ((fb)->ptr) #define FBUFFER_PTR(fb) ((fb)->ptr)
@ -66,7 +68,7 @@ static void fbuffer_append_long(FBuffer *fb, long number);
#endif #endif
static inline void fbuffer_append_char(FBuffer *fb, char newchr); static inline void fbuffer_append_char(FBuffer *fb, char newchr);
#ifdef JSON_GENERATOR #ifdef JSON_GENERATOR
static VALUE fbuffer_to_s(FBuffer *fb); static VALUE fbuffer_finalize(FBuffer *fb);
#endif #endif
static void fbuffer_stack_init(FBuffer *fb, unsigned long initial_length, char *stack_buffer, long stack_buffer_size) static void fbuffer_stack_init(FBuffer *fb, unsigned long initial_length, char *stack_buffer, long stack_buffer_size)
@ -86,24 +88,19 @@ static void fbuffer_free(FBuffer *fb)
} }
} }
#ifndef JSON_GENERATOR
static void fbuffer_clear(FBuffer *fb) static void fbuffer_clear(FBuffer *fb)
{ {
fb->len = 0; fb->len = 0;
} }
#endif
static void fbuffer_do_inc_capa(FBuffer *fb, unsigned long requested) static void fbuffer_flush(FBuffer *fb)
{ {
unsigned long required; rb_io_write(fb->io, rb_utf8_str_new(fb->ptr, fb->len));
fbuffer_clear(fb);
if (RB_UNLIKELY(!fb->ptr)) { }
fb->ptr = ALLOC_N(char, fb->initial_length);
fb->capa = fb->initial_length;
}
for (required = fb->capa; requested > required - fb->len; required <<= 1);
static void fbuffer_realloc(FBuffer *fb, unsigned long required)
{
if (required > fb->capa) { if (required > fb->capa) {
if (fb->type == FBUFFER_STACK_ALLOCATED) { if (fb->type == FBUFFER_STACK_ALLOCATED) {
const char *old_buffer = fb->ptr; const char *old_buffer = fb->ptr;
@ -117,6 +114,32 @@ static void fbuffer_do_inc_capa(FBuffer *fb, unsigned long requested)
} }
} }
static void fbuffer_do_inc_capa(FBuffer *fb, unsigned long requested)
{
if (RB_UNLIKELY(fb->io)) {
if (fb->capa < FBUFFER_IO_BUFFER_SIZE) {
fbuffer_realloc(fb, FBUFFER_IO_BUFFER_SIZE);
} else {
fbuffer_flush(fb);
}
if (RB_LIKELY(requested < fb->capa)) {
return;
}
}
unsigned long required;
if (RB_UNLIKELY(!fb->ptr)) {
fb->ptr = ALLOC_N(char, fb->initial_length);
fb->capa = fb->initial_length;
}
for (required = fb->capa; requested > required - fb->len; required <<= 1);
fbuffer_realloc(fb, required);
}
static inline void fbuffer_inc_capa(FBuffer *fb, unsigned long requested) static inline void fbuffer_inc_capa(FBuffer *fb, unsigned long requested)
{ {
if (RB_UNLIKELY(requested > fb->capa - fb->len)) { if (RB_UNLIKELY(requested > fb->capa - fb->len)) {
@ -174,11 +197,18 @@ static void fbuffer_append_long(FBuffer *fb, long number)
fbuffer_append(fb, buffer_end - len, len); fbuffer_append(fb, buffer_end - len, len);
} }
static VALUE fbuffer_to_s(FBuffer *fb) static VALUE fbuffer_finalize(FBuffer *fb)
{ {
VALUE result = rb_utf8_str_new(FBUFFER_PTR(fb), FBUFFER_LEN(fb)); if (fb->io) {
fbuffer_free(fb); fbuffer_flush(fb);
return result; fbuffer_free(fb);
rb_io_flush(fb->io);
return fb->io;
} else {
VALUE result = rb_utf8_str_new(FBUFFER_PTR(fb), FBUFFER_LEN(fb));
fbuffer_free(fb);
return result;
}
} }
#endif #endif
#endif #endif

View File

@ -54,7 +54,7 @@ struct generate_json_data {
}; };
static VALUE cState_from_state_s(VALUE self, VALUE opts); static VALUE cState_from_state_s(VALUE self, VALUE opts);
static VALUE cState_partial_generate(VALUE self, VALUE obj, generator_func); static VALUE cState_partial_generate(VALUE self, VALUE obj, generator_func, VALUE io);
static void generate_json(FBuffer *buffer, struct generate_json_data *data, JSON_Generator_State *state, VALUE obj); static void generate_json(FBuffer *buffer, struct generate_json_data *data, JSON_Generator_State *state, VALUE obj);
static void generate_json_object(FBuffer *buffer, struct generate_json_data *data, JSON_Generator_State *state, VALUE obj); static void generate_json_object(FBuffer *buffer, struct generate_json_data *data, JSON_Generator_State *state, VALUE obj);
static void generate_json_array(FBuffer *buffer, struct generate_json_data *data, JSON_Generator_State *state, VALUE obj); static void generate_json_array(FBuffer *buffer, struct generate_json_data *data, JSON_Generator_State *state, VALUE obj);
@ -453,7 +453,7 @@ static VALUE mHash_to_json(int argc, VALUE *argv, VALUE self)
{ {
rb_check_arity(argc, 0, 1); rb_check_arity(argc, 0, 1);
VALUE Vstate = cState_from_state_s(cState, argc == 1 ? argv[0] : Qnil); VALUE Vstate = cState_from_state_s(cState, argc == 1 ? argv[0] : Qnil);
return cState_partial_generate(Vstate, self, generate_json_object); return cState_partial_generate(Vstate, self, generate_json_object, Qfalse);
} }
/* /*
@ -467,7 +467,7 @@ static VALUE mHash_to_json(int argc, VALUE *argv, VALUE self)
static VALUE mArray_to_json(int argc, VALUE *argv, VALUE self) { static VALUE mArray_to_json(int argc, VALUE *argv, VALUE self) {
rb_check_arity(argc, 0, 1); rb_check_arity(argc, 0, 1);
VALUE Vstate = cState_from_state_s(cState, argc == 1 ? argv[0] : Qnil); VALUE Vstate = cState_from_state_s(cState, argc == 1 ? argv[0] : Qnil);
return cState_partial_generate(Vstate, self, generate_json_array); return cState_partial_generate(Vstate, self, generate_json_array, Qfalse);
} }
#ifdef RUBY_INTEGER_UNIFICATION #ifdef RUBY_INTEGER_UNIFICATION
@ -480,7 +480,7 @@ static VALUE mInteger_to_json(int argc, VALUE *argv, VALUE self)
{ {
rb_check_arity(argc, 0, 1); rb_check_arity(argc, 0, 1);
VALUE Vstate = cState_from_state_s(cState, argc == 1 ? argv[0] : Qnil); VALUE Vstate = cState_from_state_s(cState, argc == 1 ? argv[0] : Qnil);
return cState_partial_generate(Vstate, self, generate_json_integer); return cState_partial_generate(Vstate, self, generate_json_integer, Qfalse);
} }
#else #else
@ -493,7 +493,7 @@ static VALUE mFixnum_to_json(int argc, VALUE *argv, VALUE self)
{ {
rb_check_arity(argc, 0, 1); rb_check_arity(argc, 0, 1);
VALUE Vstate = cState_from_state_s(cState, argc == 1 ? argv[0] : Qnil); VALUE Vstate = cState_from_state_s(cState, argc == 1 ? argv[0] : Qnil);
return cState_partial_generate(Vstate, self, generate_json_fixnum); return cState_partial_generate(Vstate, self, generate_json_fixnum, Qfalse);
} }
/* /*
@ -505,7 +505,7 @@ static VALUE mBignum_to_json(int argc, VALUE *argv, VALUE self)
{ {
rb_check_arity(argc, 0, 1); rb_check_arity(argc, 0, 1);
VALUE Vstate = cState_from_state_s(cState, argc == 1 ? argv[0] : Qnil); VALUE Vstate = cState_from_state_s(cState, argc == 1 ? argv[0] : Qnil);
return cState_partial_generate(Vstate, self, generate_json_bignum); return cState_partial_generate(Vstate, self, generate_json_bignum, Qfalse);
} }
#endif #endif
@ -518,7 +518,7 @@ static VALUE mFloat_to_json(int argc, VALUE *argv, VALUE self)
{ {
rb_check_arity(argc, 0, 1); rb_check_arity(argc, 0, 1);
VALUE Vstate = cState_from_state_s(cState, argc == 1 ? argv[0] : Qnil); VALUE Vstate = cState_from_state_s(cState, argc == 1 ? argv[0] : Qnil);
return cState_partial_generate(Vstate, self, generate_json_float); return cState_partial_generate(Vstate, self, generate_json_float, Qfalse);
} }
/* /*
@ -543,7 +543,7 @@ static VALUE mString_to_json(int argc, VALUE *argv, VALUE self)
{ {
rb_check_arity(argc, 0, 1); rb_check_arity(argc, 0, 1);
VALUE Vstate = cState_from_state_s(cState, argc == 1 ? argv[0] : Qnil); VALUE Vstate = cState_from_state_s(cState, argc == 1 ? argv[0] : Qnil);
return cState_partial_generate(Vstate, self, generate_json_string); return cState_partial_generate(Vstate, self, generate_json_string, Qfalse);
} }
/* /*
@ -638,7 +638,7 @@ static VALUE mObject_to_json(int argc, VALUE *argv, VALUE self)
rb_scan_args(argc, argv, "01", &state); rb_scan_args(argc, argv, "01", &state);
Check_Type(string, T_STRING); Check_Type(string, T_STRING);
state = cState_from_state_s(cState, state); state = cState_from_state_s(cState, state);
return cState_partial_generate(state, string, generate_json_string); return cState_partial_generate(state, string, generate_json_string, Qfalse);
} }
static void State_mark(void *ptr) static void State_mark(void *ptr)
@ -1045,12 +1045,14 @@ static VALUE generate_json_rescue(VALUE d, VALUE exc)
return Qundef; return Qundef;
} }
static VALUE cState_partial_generate(VALUE self, VALUE obj, generator_func func) static VALUE cState_partial_generate(VALUE self, VALUE obj, generator_func func, VALUE io)
{ {
GET_STATE(self); GET_STATE(self);
char stack_buffer[FBUFFER_STACK_SIZE]; char stack_buffer[FBUFFER_STACK_SIZE];
FBuffer buffer = {0}; FBuffer buffer = {
.io = RTEST(io) ? io : Qfalse,
};
fbuffer_stack_init(&buffer, state->buffer_initial_length, stack_buffer, FBUFFER_STACK_SIZE); fbuffer_stack_init(&buffer, state->buffer_initial_length, stack_buffer, FBUFFER_STACK_SIZE);
struct generate_json_data data = { struct generate_json_data data = {
@ -1062,19 +1064,12 @@ static VALUE cState_partial_generate(VALUE self, VALUE obj, generator_func func)
}; };
rb_rescue(generate_json_try, (VALUE)&data, generate_json_rescue, (VALUE)&data); rb_rescue(generate_json_try, (VALUE)&data, generate_json_rescue, (VALUE)&data);
return fbuffer_to_s(&buffer); return fbuffer_finalize(&buffer);
} }
/* static VALUE cState_generate(VALUE self, VALUE obj, VALUE io)
* call-seq: generate(obj)
*
* Generates a valid JSON document from object +obj+ and returns the
* result. If no valid JSON document can be created this method raises a
* GeneratorError exception.
*/
static VALUE cState_generate(VALUE self, VALUE obj)
{ {
VALUE result = cState_partial_generate(self, obj, generate_json); VALUE result = cState_partial_generate(self, obj, generate_json, io);
GET_STATE(self); GET_STATE(self);
(void)state; (void)state;
return result; return result;
@ -1502,14 +1497,16 @@ static VALUE cState_configure(VALUE self, VALUE opts)
return self; return self;
} }
static VALUE cState_m_generate(VALUE klass, VALUE obj, VALUE opts) static VALUE cState_m_generate(VALUE klass, VALUE obj, VALUE opts, VALUE io)
{ {
JSON_Generator_State state = {0}; JSON_Generator_State state = {0};
state_init(&state); state_init(&state);
configure_state(&state, opts); configure_state(&state, opts);
char stack_buffer[FBUFFER_STACK_SIZE]; char stack_buffer[FBUFFER_STACK_SIZE];
FBuffer buffer = {0}; FBuffer buffer = {
.io = RTEST(io) ? io : Qfalse,
};
fbuffer_stack_init(&buffer, state.buffer_initial_length, stack_buffer, FBUFFER_STACK_SIZE); fbuffer_stack_init(&buffer, state.buffer_initial_length, stack_buffer, FBUFFER_STACK_SIZE);
struct generate_json_data data = { struct generate_json_data data = {
@ -1521,7 +1518,7 @@ static VALUE cState_m_generate(VALUE klass, VALUE obj, VALUE opts)
}; };
rb_rescue(generate_json_try, (VALUE)&data, generate_json_rescue, (VALUE)&data); rb_rescue(generate_json_try, (VALUE)&data, generate_json_rescue, (VALUE)&data);
return fbuffer_to_s(&buffer); return fbuffer_finalize(&buffer);
} }
/* /*
@ -1583,9 +1580,9 @@ void Init_generator(void)
rb_define_method(cState, "depth=", cState_depth_set, 1); rb_define_method(cState, "depth=", cState_depth_set, 1);
rb_define_method(cState, "buffer_initial_length", cState_buffer_initial_length, 0); rb_define_method(cState, "buffer_initial_length", cState_buffer_initial_length, 0);
rb_define_method(cState, "buffer_initial_length=", cState_buffer_initial_length_set, 1); rb_define_method(cState, "buffer_initial_length=", cState_buffer_initial_length_set, 1);
rb_define_method(cState, "generate", cState_generate, 1); rb_define_private_method(cState, "_generate", cState_generate, 2);
rb_define_singleton_method(cState, "generate", cState_m_generate, 2); rb_define_singleton_method(cState, "generate", cState_m_generate, 3);
VALUE mGeneratorMethods = rb_define_module_under(mGenerator, "GeneratorMethods"); VALUE mGeneratorMethods = rb_define_module_under(mGenerator, "GeneratorMethods");

View File

@ -286,7 +286,7 @@ module JSON
if State === opts if State === opts
opts.generate(obj) opts.generate(obj)
else else
State.generate(obj, opts) State.generate(obj, opts, nil)
end end
end end
@ -801,18 +801,15 @@ module JSON
opts = opts.merge(:max_nesting => limit) if limit opts = opts.merge(:max_nesting => limit) if limit
opts = merge_dump_options(opts, **kwargs) if kwargs opts = merge_dump_options(opts, **kwargs) if kwargs
result = begin begin
generate(obj, opts) if State === opts
opts.generate(obj, anIO)
else
State.generate(obj, opts, anIO)
end
rescue JSON::NestingError rescue JSON::NestingError
raise ArgumentError, "exceed depth limit" raise ArgumentError, "exceed depth limit"
end end
if anIO.nil?
result
else
anIO.write result
anIO
end
end end
# Encodes string using String.encode. # Encodes string using String.encode.

View File

@ -47,6 +47,17 @@ module JSON
alias_method :merge, :configure alias_method :merge, :configure
# call-seq:
# generate(obj) -> String
# generate(obj, anIO) -> anIO
#
# Generates a valid JSON document from object +obj+ and returns the
# result. If no valid JSON document can be created this method raises a
# GeneratorError exception.
def generate(obj, io = nil)
_generate(obj, io)
end
# call-seq: to_h # call-seq: to_h
# #
# Returns the configuration instance variables as a hash, that can be # Returns the configuration instance variables as a hash, that can be

View File

@ -162,6 +162,17 @@ class JSONCommonInterfaceTest < Test::Unit::TestCase
assert_equal too_deep, dump(obj, strict: false) assert_equal too_deep, dump(obj, strict: false)
end end
def test_dump_in_io
io = StringIO.new
assert_same io, JSON.dump([1], io)
assert_equal "[1]", io.string
big_object = ["a" * 10, "b" * 40, { foo: 1.23 }] * 5000
io.rewind
assert_same io, JSON.dump(big_object, io)
assert_equal JSON.dump(big_object), io.string
end
def test_dump_should_modify_defaults def test_dump_should_modify_defaults
max_nesting = JSON.dump_default_options[:max_nesting] max_nesting = JSON.dump_default_options[:max_nesting]
dump([], StringIO.new, 10) dump([], StringIO.new, 10)