chaokunyang commented on code in PR #1413: URL: https://github.com/apache/incubator-fury/pull/1413#discussion_r1542829185
########## docs/protocols/xlang_object_graph_spec.md: ########## @@ -0,0 +1,692 @@ +# Cross language object graph serialization + +Fury xlang serialization is an automatic object serialization framework that supports reference and polymorphism. +Fury will convert an object from/to fury xlang serialization binary format. +Fury has two core concepts for xlang serialization: + +- **Fury xlang binary format** +- **Framework implemented in different languages to convert object to/from Fury xlang binary format** + +The serialization format is a dynamic binary format. The dynamics and reference/polymorphism support make Fury flexible, +much more easy to use, but +also introduce more complexities compared to static serialization frameworks. So the format will be more complex. + +## Type Systems + +### Data Types + +- bool: a boolean value (true or false). +- int8: a 8-bit signed integer. +- int16: a 16-bit signed integer. +- int32: a 32-bit signed integer. +- var_int32: a 32-bit signed integer which use fury var_int32 encoding. +- fixed_int32: a 32-bit signed integer which use two's complement encoding. +- int64: a 64-bit signed integer. +- var_int64: a 64-bit signed integer which use fury PVL encoding. +- sli_int64: a 64-bit signed integer which use fury SLI encoding. +- fixed_int64: a 64-bit signed integer which use two's complement encoding. +- float16: a 16-bit floating point number. +- float32: a 32-bit floating point number. +- float64: a 64-bit floating point number including NaN and Infinity. +- string: a text string encoded using Latin1/UTF16/UTF-8 encoding. +- enum: a data type consisting of a set of named values. Rust enum with non-predefined field values are not supported as + an enum. +- list: a sequence of objects. +- set: an unordered set of unique elements. +- map: a map of key-value pairs. Mutable types such as `list/map/set/array/tensor/arrow` are not allowed as key of map. +- time types: + - duration: an absolute length of time, independent of any calendar/timezone, as a count of nanoseconds. + - timestamp: a point in time, independent of any calendar/timezone, as a count of nanoseconds. The count is relative + to an epoch at UTC midnight on January 1, 1970. +- decimal: exact decimal value represented as an integer value in two's complement. +- binary: an variable-length array of bytes. +- array type: only allow numeric components. Other arrays will be taken as List. The implementation should support the + interoperability between array and list. + - array: multidimensional array which every sub-array can have different sizes but all have same type. + - bool_array: one dimensional int16 array. + - int16_array: one dimensional int16 array. + - int32_array: one dimensional int32 array. + - int64_array: one dimensional int64 array. + - float16_array: one dimensional half_float_16 array. + - float32_array: one dimensional float32 array. + - float64_array: one dimensional float64 array. +- tensor: a multidimensional dense array of fixed-size values such as a NumPy ndarray. +- sparse tensor: a multidimensional array whose elements are almost all zeros. +- arrow record batch: an arrow [record batch](https://arrow.apache.org/docs/cpp/tables.html#record-batches) object. +- arrow table: an arrow [table](https://arrow.apache.org/docs/cpp/tables.html#tables) object. + +Note: + +- Unsigned int/long are not added here, since not every language support those types. + +### Type disambiguation + +Due to differences between type systems of languages, those types can't be mapped one-to-one between languages. When +deserializing, Fury use the target data structure type and the data type in the data jointly to determine how to +deserialize and populate the target data structure. For example: + +```java +class Foo { + int[] intArray; + Object[] objects; + List<Object> objectList; +} + +class Foo2 { + int[] intArray; + List<Object> objects; + List<Object> objectList; +} +``` + +`intArray` has an `int32_array` type. But both `objects` and `objectList` fields in the serialize data have `list` data +type. When deserializing, the implementation will create an `Object` array for `objects`, but create a `ArrayList` +for `objectList` to populate its elements. And the serialized data of `Foo` can be deserialized into `Foo2` too. + +Users can also provide meta hints for fields of a type, or the type whole. Here is an example in java which use +annotation to provide such information. + +```java + +@TypeInfo(fieldsNullable = false, trackingRef = false, polymorphic = false) +class Foo { + @FieldInfo(trackingRef = false) + int[] intArray; + @FieldInfo(polymorphic = true) + Object object; + @FieldInfo(tagId = 1, nullable = true) + List<Object> objectList; +} +``` + +Such information can be provided in other languages too: + +- cpp: use macro and template. +- golang: use struct tag. +- python: use typehint. +- rust: use macro. + +### Type ID + +All internal data types are expressed using an ID in range `-64~-1`. Users can use `0~32703` for representing their +types. At runtime, all type ids are added by `64`, and then encoded as an unsigned varint. + +### Type mapping + +See [Type mapping](../guide/xlang_type_mapping.md) + +## Spec overview + +Here is the overall format: + +``` +| fury header | object ref meta | object type meta | object value data | +``` + +The data are serialized using little endian byte order overall. If bytes swap is costly for some object, +Fury will write the byte order for that object into the data instead of converting it to little endian. + +## Fury header + +Fury header consists starts one byte: + +``` +| 4 bits | 1 bit | 1 bit | 1 bit | 1 bit | optional 4 bytes | ++---------------+-------+-------+--------+-------+------------------------------------+ +| reserved bits | oob | xlang | endian | null | unsigned int for meta start offset | +``` + +- null flag: 1 when object is null, 0 otherwise. If an object is null, other bits won't be set. +- endian flag: 1 when data is encoded by little endian, 0 for big endian. +- xlang flag: 1 when serialization uses xlang format, 0 when serialization uses Fury java format. +- oob flag: 1 when passed `BufferCallback` is not null, 0 otherwise. + +If meta share mode is enabled, an uncompressed unsigned int is appended to indicate the start offset of metadata. + +## Reference Meta + +Reference tracking handles whether the object is null, and whether to track reference for the object by writing +corresponding flags and maintaining internal state. + +Reference flags: + +| Flag | Byte Value | Description | +|---------------------|------------|---------------------------------------------------------------------------------------------------------------------------------------------------------| +| NULL FLAG | `-3` | This flag indicates the object is a null value. We don't use another byte to indicate REF, so that we can save one byte. | +| REF FLAG | `-2` | This flag indicates the object is already serialized previously, and fury will write a ref id with unsigned varint format instead of serialize it again | +| NOT_NULL VALUE FLAG | `-1` | This flag indicates the object is a non-null value and fury doesn't track ref for this type of object. | +| REF VALUE FLAG | `0` | This flag indicates the object is referencable and the first time to serialize. | + +When reference tracking is disabled globally or for specific types, or for certain types within a particular +context(e.g., a field of a type), only the `NULL` and `NOT_NULL VALUE` flags will be used for reference meta. + +For languages which doesn't support reference such as rust, reference tracking must be disabled for correct +deserialization by fury rust implementation. + +For languages whose object values are not null by default: + +- In rust, Fury takes `Option:None` as a null value +- In c++, Fury takes `std::nullopt` as a null value +- In golang, Fury takes `null interface/pointer` as a null value + +If one want to deserialize in languages like `Java/Python/JavaScript`, he should mark the type with all fields +not-null by default, or using schema-evolution mode to carry the not-null fields info in the data. + +## Type Meta + +For every type to be serialized, it must be registered with an optional ID first. The registered type will have a +user-provided or an auto-growing unsigned int i.e. `type_id`. The registration can be used for security check and type +identification. The id of user registered type will be added by `64` to make space for Fury internal data types. + +Depending on whether meta share mode and registration is enabled for current type, Fury will write type meta +differently. + +### Schema consistent + +If schema consistent mode is enabled globally or enabled for current type, type meta will be written as a fury unsigned +varint of `type_id`. + +### Schema evolution + +If schema evolution mode is enabled globally or enabled for current type, type meta will be written as follows: + +- If meta share mode is not enabled: + - If type meta hasn't been written before, the data will be written as: + ``` + | unsigned varint: 0b11111111 | type def | + ``` + - If type meta has been written before, the data will be written as: + ``` + | unsigned varint: written index << 1 | type def | Review Comment: Good catch, I missed it -- This is an automated message from the Apache Git Service. To respond to the message, please log on to GitHub and use the URL above to go to the specific comment. To unsubscribe, e-mail: commits-unsubscr...@fury.apache.org For queries about this service, please contact Infrastructure at: us...@infra.apache.org --------------------------------------------------------------------- To unsubscribe, e-mail: commits-unsubscr...@fury.apache.org For additional commands, e-mail: commits-h...@fury.apache.org
