Poison is a new JSON library for Elixir focusing on wicked-fast speed without sacrificing simplicity, completeness, or correctness.
Poison takes several approaches to be the fastest JSON library for Elixir.
Poison uses extensive sub binary matching, a hand-rolled parser using several techniques that are known to benefit HiPE for native compilation, IO list encoding and single-pass decoding.
Poison benchmarks sometimes puts Poison's performance close to jiffy
and
usually faster than other Erlang/Elixir libraries.
Poison fully conforms to RFC 7159, ECMA 404, and the JSONTestSuite.
First, add Poison to your mix.exs
dependencies:
def deps do
[{:poison, "~> 3.1"}]
end
Then, update your dependencies:
$ mix deps.get
Poison.encode!(%{"age" => 27, "name" => "Devin Torres"})
#=> "{\"name\":\"Devin Torres\",\"age\":27}"
Poison.decode!(~s({"name": "Devin Torres", "age": 27}))
#=> %{"age" => 27, "name" => "Devin Torres"}
defmodule Person do
@derive [Poison.Encoder]
defstruct [:name, :age]
end
Poison.encode!(%Person{name: "Devin Torres", age: 27})
#=> "{\"name\":\"Devin Torres\",\"age\":27}"
Poison.decode!(~s({"name": "Devin Torres", "age": 27}), as: %Person{})
#=> %Person{name: "Devin Torres", age: 27}
Poison.decode!(~s({"people": [{"name": "Devin Torres", "age": 27}]}),
as: %{"people" => [%Person{}]})
#=> %{"people" => [%Person{age: 27, name: "Devin Torres"}]}
Every component of Poison (encoder, decoder, and parser) are all usable on
their own without buying into other functionality. For example, if you were
interested purely in the speed of parsing JSON without a decoding step, you
could simply call Poison.Parser.parse
.
iex> Poison.Parser.parse!(~s({"name": "Devin Torres", "age": 27}), %{})
%{"name" => "Devin Torres", "age" => 27}
iex> Poison.Parser.parse!(~s({"name": "Devin Torres", "age": 27}), %{keys: :atoms!})
%{name: "Devin Torres", age: 27}
Note that keys: :atoms!
reuses existing atoms, i.e. if :name
was not
allocated before the call, you will encounter an argument error
message.
You can use the keys: :atoms
variant to make sure all atoms are created as
needed. However, unless you absolutely know what you're doing, do not do
it. Atoms are not garbage-collected, see
Erlang Efficiency Guide
for more info:
Atoms are not garbage-collected. Once an atom is created, it will never be removed. The emulator will terminate if the limit for the number of atoms (1048576 by default) is reached.
iex> Poison.Encoder.encode([1, 2, 3], %{}) |> IO.iodata_to_binary
"[1,2,3]"
Anything implementing the Encoder protocol is expected to return an IO list to be embedded within any other Encoder's implementation and passable to any IO subsystem without conversion.
defimpl Poison.Encoder, for: Person do
def encode(%{name: name, age: age}, options) do
Poison.Encoder.BitString.encode("#{name} (#{age})", options)
end
end
For maximum performance, make sure you @derive [Poison.Encoder]
for any
struct you plan on encoding.
When deriving structs for encoding, it is possible to select or exclude
specific attributes. This is achieved by deriving Poison.Encoder
with the
:only
or :except
options set:
defmodule PersonOnlyName do
@derive {Poison.Encoder, only: [:name]}
defstruct [:name, :age]
end
defmodule PersonWithoutName do
@derive {Poison.Encoder, except: [:name]}
defstruct [:name, :age]
end
In case both :only
and :except
keys are defined, the :except
option is
ignored.
According to RFC 7159 keys in a JSON object should be unique. This is enforced and resolved in different ways in other libraries. In the Ruby JSON library for example, the output generated from encoding a hash with a duplicate key (say one is a string, the other an atom) will include both keys. When parsing JSON of this type, Chromium will override all previous values with the final one.
Poison will generate JSON with duplicate keys if you attempt to encode a map
with atom and string keys whose encoded names would clash. If you'd like to
ensure that your generated JSON doesn't have this issue, you can pass the
strict_keys: true
option when encoding. This will force the encoding to fail.
Note: Validating keys can cause a small performance hit.
iex> Poison.encode!(%{:foo => "foo1", "foo" => "foo2"}, strict_keys: true)
** (Poison.EncodeError) duplicate key found: :foo
$ MIX_ENV=bench mix run bench/run.exs
As of 2017-05-15 on a 2.8 GHz Intel Core i7:
## EncoderBench
benchmark name iterations average time
maps (jiffy) 500000 7.88 µs/op
structs (Poison) 200000 9.46 µs/op
structs (Jazz) 100000 15.43 µs/op
structs (JSX) 100000 18.45 µs/op
maps (Poison) 100000 19.45 µs/op
maps (Jazz) 100000 21.61 µs/op
maps (JSX) 50000 31.76 µs/op
maps (JSON) 50000 34.08 µs/op
structs (JSON) 50000 47.56 µs/op
strings (jiffy) 10000 107.68 µs/op
lists (Poison) 10000 120.79 µs/op
string escaping (jiffy) 10000 139.92 µs/op
lists (jiffy) 10000 229.18 µs/op
lists (Jazz) 10000 236.86 µs/op
strings (JSON) 10000 237.97 µs/op
strings (JSX) 10000 283.87 µs/op
lists (JSX) 5000 336.96 µs/op
jiffy 5000 429.92 µs/op
strings (Jazz) 5000 430.78 µs/op
jiffy (pretty) 5000 431.55 µs/op
lists (JSON) 5000 559.31 µs/op
strings (Poison) 5000 574.26 µs/op
string escaping (Jazz) 1000 1313.51 µs/op
string escaping (JSX) 1000 1474.66 µs/op
Poison 1000 1546.53 µs/op
string escaping (Poison) 1000 1728.66 µs/op
Poison (pretty) 1000 1784.37 µs/op
Jazz 1000 2060.77 µs/op
JSON 1000 2250.89 µs/op
JSX 1000 2252.77 µs/op
Jazz (pretty) 1000 2317.55 µs/op
JSX (pretty) 500 5577.33 µs/op
## ParserBench
benchmark name iterations average time
UTF-8 unescaping (jiffy) 50000 60.05 µs/op
UTF-8 unescaping (Poison) 10000 112.53 µs/op
UTF-8 unescaping (JSX) 10000 282.83 µs/op
UTF-8 unescaping (JSON) 5000 469.26 µs/op
jiffy 5000 479.07 µs/op
Poison 5000 730.85 µs/op
JSX 1000 1947.77 µs/op
JSON 500 5175.11 µs/op
Issue 90 (jiffy) 100 18864.70 µs/op
Issue 90 (Poison) 50 50091.16 µs/op
Issue 90 (JSX) 10 155975.20 µs/op
Issue 90 (JSON) 1 1964860.00 µs/op
Poison is released under CC0-1.0.