A Simple Document Format

ASDF is a cache oriented string based JSON representation. Besides, it is a convenient Json Library for D that gets out of your way. ASDF is specially geared towards transforming high volumes of JSON dataframes, either to new JSON Objects or to custom data types.

Why ASDF?

asdf was originally developed at Tamedia to extract and transform real-time click streams.

• ASDF is fast. It can be really helpful if you have gigabytes of JSON line separated values.
• ASDF is simple. It uses D's modelling power to make you write less boilerplate code.
• ASDF is tested and used in production for real World JSON generated by millions of web clients (we call it the great fuzzer).

see also github.com/tamediadigital/je a tool for fast extraction of json properties into a csv/tsv.

Simple Example

1. define your struct
2. call serializeToJson ( or serializeToJsonPretty for pretty printing! )
3. profit!

/+dub.sdl:
dependency "asdf" version="~>0.2.5"

#turns on SSE4.2 optimizations when compiled with LDC
dflags "-mattr=+sse4.2" platform="ldc"
+/
import asdf;

struct Simple
{
    string name;
    ulong level;
}

void main()
{
    auto o = Simple("asdf", 42);
    string data = `{"name":"asdf","level":42}`;
    assert(o.serializeToJson() == data);
    assert(data.deserialize!Simple == o);
}

Documentation

See ASDF API and Specification.

I/O Speed

• Reading JSON line separated values and parsing them to ASDF - 300+ MB per second (SSD).
• Writing ASDF range to JSON line separated values - 300+ MB per second (SSD).

Fast setup with the dub package manager

Dub is D's package manager. You can create a new project with:

dub init <project-name>

Now you need to edit the dub.json add asdf as dependency and set its targetType to executable.

(dub.json)

{
    ...
    "dependencies": {
        "asdf": "~><current-version>"
    },
    "targetType": "executable",
    "dflags-ldc": ["-mcpu=native"]
}

(dub.sdl)

dependency "asdf" version="~><current-version>"
targetType "executable"
dflags "-mcpu=native" platform="ldc"

Now you can create a main file in the source and run your code with

dub

Flags --build=release and --compiler=ldmd2 can be added for a performance boost:

dub --build=release --compiler=ldmd2

ldmd2 is a shell on top of LDC (LLVM D Compiler). "dflags-ldc": ["-mcpu=native"] allows LDC to optimize ASDF for your CPU.

Instead of using -mcpu=native, you may specify an additional instruction set for a target with -mattr. For example, -mattr=+sse4.2. ASDF has specialized code for [SSE4.2](https://en.wikipedia.org/wiki/SSE4#SSE4.2 instruction set).

Main transformation functions

Please also look into the Docs or Unittest for concrete examples!

ASDF Example (incomplete)

import std.algorithm;
import std.stdio;
import asdf;

void main()
{
    auto target = Asdf("red");
    File("input.jsonl")
        // Use at least 4096 bytes for real world apps
        .byChunk(4096)
        // 32 is minimum size for internal buffer. Buffer can be reallocated to get more memory.
        .parseJsonByLine(4096)
        .filter!(object => object
            // opIndex accepts array of keys: {"key0": {"key1": { ... {"keyN-1": <value>}... }}}
            ["colors"]
            // iterates over an array
            .byElement
            // Comparison with ASDF is little bit faster
            //   than comparison with a string.
            .canFind(target))
            //.canFind("red"))
        // Formatting uses internal buffer to reduce system delegate and system function calls
        .each!writeln;
}

Input

Single object per line: 4th and 5th lines are broken.

null
{"colors": ["red"]}
{"a":"b", "colors": [4, "red", "string"]}
{"colors":["red"],
    "comment" : "this is broken (multiline) object"}
{"colors": "green"}
{"colors": "red"]}}
[]

Output

{"colors":["red"]}
{"a":"b","colors":[4,"red","string"]}

JSON and ASDF Serialization Examples

Simple struct or object

struct S
{
    string a;
    long b;
    private int c; // private fields are ignored
    package int d; // package fields are ignored
    // all other fields in JSON are ignored
}

Selection

struct S
{
    // ignored
    @serdeIgnore int temp;

    // can be formatted to json
    @serdeIgnoreIn int a;

    //can be parsed from json
    @serdeIgnoreOut int b;

    // ignored if negative
    @serdeIgnoreOutIf!`a < 0` int c;
}

Key overriding

struct S
{
    // key is overrided to "aaa"
    @serdeKeys("aaa") int a;

    // overloads multiple keys for parsing
    @serdeKeysIn("b", "_b")
    // overloads key for generation
    @serdeKeyOut("_b_")
    int b;
}

User-Defined Serialization

struct DateTimeProxy
{
    DateTime datetime;
    alias datetime this;

    SerdeException deserializeFromAsdf(Asdf data)
    {
        string val;
        if (auto exc = deserializeScopedString(data, val))
            return exc;
        this = DateTimeProxy(DateTime.fromISOString(val));
        return null;
    }

    void serialize(S)(ref S serializer)
    {
        serializer.putValue(datetime.toISOString);
    }
}

//serialize a Doubly Linked list into an Array
struct SomeDoublyLinkedList
{
    @serdeIgnore DList!(SomeArr[]) myDll;
    alias myDll this;

    //no template but a function this time!
    void serialize(ref AsdfSerializer serializer)
    {
        auto state = serializer.listBegin();
        foreach (ref elem; myDll)
        {
            serializer.elemBegin;
            serializer.serializeValue(elem);
        }
        serializer.listEnd(state);
    }   
}

Serialization Proxy

struct S
{
    @serdeProxy!DateTimeProxy DateTime time;
}

@serdeProxy!ProxyE
enum E
{
    none,
    bar,
}

// const(char)[] doesn't reallocate ASDF data.
@serdeProxy!(const(char)[])
struct ProxyE
{
    E e;

    this(E e)
    {
        this.e = e;
    }

    this(in char[] str)
    {
        switch(str)
        {
            case "NONE":
            case "NA":
            case "N/A":
                e = E.none;
                break;
            case "BAR":
            case "BR":
                e = E.bar;
                break;
            default:
                throw new Exception("Unknown: " ~ cast(string)str);
        }
    }

    string toString()
    {
        if (e == E.none)
            return "NONE";
        else
            return "BAR";
    }

    E opCast(T : E)()
    {
        return e;
    }
}

unittest
{
    assert(serializeToJson(E.bar) == `"BAR"`);
    assert(`"N/A"`.deserialize!E == E.none);
    assert(`"NA"`.deserialize!E == E.none);
}

Finalizer

If you need to do additional calculations or etl transformations that happen to depend on the deserialized data use the finalizeDeserialization method.

struct S
{
    string a;
    int b;

    @serdeIgnoreIn double sum;

    void finalizeDeserialization(Asdf data)
    {
        auto r = data["c", "d"];
        auto a = r["e"].get(0.0);
        auto b = r["g"].get(0.0);
        sum = a + b;
    }
}
assert(`{"a":"bar","b":3,"c":{"d":{"e":6,"g":7}}}`.deserialize!S == S("bar", 3, 13));