Optimization for plain objects

[lucagez]

There's currently no support for serializing plain objects with unknown shape. A temporary workaround for this is to use a custom serializer, specifying null as a target type. Naively:

attr('null', (value) => JSON.stringify(value))

It could be possible to optimize this kind of scenario with a jit like approach:

1. A custom helper can handle the serialization of Objects with unknown shape
2. During the first serialization, JSON.stringify(value, <custom-replacer>) can be used to traverse all the properties of value and building a schema
3. The JSON.stringify serializer can then be replaced with the schema built in step 2

Caveats

Shape consistency: This jit like behavior would have an appreciable performance benefit if the unknown shape stays consistent for enough time to justify the performance penalty of building the schema.

Schema invalidation: A check should be performed on the incoming object to make sure its shape matches the one of the generated schema. Otherwise, the schema should be invalidated and recreated. This step could have a non trivial performance impact

Any idea/suggestion welcome

[dalisoft]

@lucagez Thanks. if you want run benchmark on your machine, look at https://github.com/nanoexpress/json-serialize/blob/master/bench-suite.js.

On main benchmark your library wins, but in Suite on gets a little less performance score