Skip to content

Fast JSON serializer for golang.

License

Notifications You must be signed in to change notification settings

apexlang/tinyjson

 
 

Folders and files

NameName
Last commit message
Last commit date

Latest commit

 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 

Repository files navigation

tinyjson

Build Status | Go Report Card

Forked from easyjson, which developed all the amazing codegen technology and optimizations. This fork modifies the repo to target TinyGo execution (no encoding/json or net or reflect), and particularly CosmWasm contracts (which have no floats).

Package tinyjson provides a fast and easy way to marshal/unmarshal Go structs to/from JSON without the use of reflection. In performance tests, tinyjson outperforms the standard encoding/json package by a factor of 4-5x, and other JSON encoding packages by a factor of 2-3x.

tinyjson aims to keep generated Go code simple enough so that it can be easily optimized or fixed. Another goal is to provide users with the ability to customize the generated code by providing options not available with the standard encoding/json package, such as generating "snake_case" names or enabling omitempty behavior by default.

Usage

# install
go get -u github.com/CosmWasm/tinyjson/...

# run
tinyjson -all <file>.go

The above will generate <file>_tinyjson.go containing the appropriate marshaler and unmarshaler funcs for all structs contained in <file>.go.

Please note that tinyjson requires a full Go build environment and the GOPATH environment variable to be set. This is because tinyjson code generation invokes go run on a temporary file (an approach to code generation borrowed from ffjson).

Options

Usage of tinyjson:
  -all
    	generate marshaler/unmarshalers for all structs in a file
  -build_tags string
        build tags to add to generated file
  -gen_build_flags string
        build flags when running the generator while bootstrapping
  -byte
        use simple bytes instead of Base64Bytes for slice of bytes
  -leave_temps
    	do not delete temporary files
  -no_std_marshalers
    	don't generate MarshalJSON/UnmarshalJSON funcs
  -noformat
    	do not run 'gofmt -w' on output file
  -omit_empty
    	omit empty fields by default
  -output_filename string
    	specify the filename of the output
  -pkg
    	process the whole package instead of just the given file
  -snake_case
    	use snake_case names instead of CamelCase by default
  -lower_camel_case
        use lowerCamelCase instead of CamelCase by default
  -stubs
    	only generate stubs for marshaler/unmarshaler funcs
  -disallow_unknown_fields
        return error if some unknown field in json appeared
  -disable_members_unescape
        disable unescaping of \uXXXX string sequences in member names

Using -all will generate marshalers/unmarshalers for all Go structs in the file excluding those structs whose preceding comment starts with tinyjson:skip. For example:

//tinyjson:skip
type A struct {}

If -all is not provided, then only those structs whose preceding comment starts with tinyjson:json will have marshalers/unmarshalers generated. For example:

//tinyjson:json
type A struct {}

Additional option notes:

  • -snake_case tells tinyjson to generate snake_case field names by default (unless overridden by a field tag). The CamelCase to snake_case conversion algorithm should work in most cases (ie, HTTPVersion will be converted to "http_version").

  • -build_tags will add the specified build tags to generated Go sources.

  • -gen_build_flags will execute the tinyjson bootstapping code to launch the actual generator command with provided flags. Multiple arguments should be separated by space e.g. -gen_build_flags="-mod=mod -x".

Structure json tag options

Besides standart json tag options like 'omitempty' the following are supported:

  • 'nocopy' - disables allocation and copying of string values, making them refer to original json buffer memory. This works great for short lived objects which are not hold in memory after decoding and immediate usage. Note if string requires unescaping it will be processed as normally.
  • 'intern' - string "interning" (deduplication) to save memory when the very same string dictionary values are often met all over the structure. See below for more details.

Generated Marshaler/Unmarshaler Funcs

For Go struct types, tinyjson generates the funcs MarshalTinyJSON / UnmarshalTinyJSON for marshaling/unmarshaling JSON. In turn, these satisfy the tinyjson.Marshaler and tinyjson.Unmarshaler interfaces and when used in conjunction with tinyjson.Marshal / tinyjson.Unmarshal avoid unnecessary reflection / type assertions during marshaling/unmarshaling to/from JSON for Go structs.

tinyjson also generates MarshalJSON and UnmarshalJSON funcs for Go struct types compatible with the standard json.Marshaler and json.Unmarshaler interfaces. Please be aware that using the standard json.Marshal / json.Unmarshal for marshaling/unmarshaling will incur a significant performance penalty when compared to using tinyjson.Marshal / tinyjson.Unmarshal.

Additionally, tinyjson exposes utility funcs that use the MarshalTinyJSON and UnmarshalTinyJSON for marshaling/unmarshaling to and from standard readers and writers. For example, tinyjson provides tinyjson.MarshalToHTTPResponseWriter which marshals to the standard http.ResponseWriter. Please see the GoDoc listing for the full listing of utility funcs that are available.

Controlling tinyjson Marshaling and Unmarshaling Behavior

Go types can provide their own MarshalTinyJSON and UnmarshalTinyJSON funcs that satisfy the tinyjson.Marshaler / tinyjson.Unmarshaler interfaces. These will be used by tinyjson.Marshal and tinyjson.Unmarshal when defined for a Go type.

Go types can also satisfy the tinyjson.Optional interface, which allows the type to define its own omitempty logic.

Type Wrappers

tinyjson provides additional type wrappers defined in the tinyjson/opt package. These wrap the standard Go primitives and in turn satisfy the tinyjson interfaces.

The tinyjson/opt type wrappers are useful when needing to distinguish between a missing value and/or when needing to specifying a default value. Type wrappers allow tinyjson to avoid additional pointers and heap allocations and can significantly increase performance when used properly.

Memory Pooling

tinyjson uses a buffer pool that allocates data in increasing chunks from 128 to 32768 bytes. Chunks of 512 bytes and larger will be reused with the help of sync.Pool. The maximum size of a chunk is bounded to reduce redundant memory allocation and to allow larger reusable buffers.

tinyjson's custom allocation buffer pool is defined in the tinyjson/buffer package, and the default behavior pool behavior can be modified (if necessary) through a call to buffer.Init() prior to any marshaling or unmarshaling. Please see the GoDoc listing for more information.

String interning

During unmarshaling, string field values can be optionally interned to reduce memory allocations and usage by deduplicating strings in memory, at the expense of slightly increased CPU usage.

This will work effectively only for string fields being decoded that have frequently the same value (e.g. if you have a string field that can only assume a small number of possible values).

To enable string interning, add the intern keyword tag to your json tag on string fields, e.g.:

type Foo struct {
  UUID  string `json:"uuid"`         // will not be interned during unmarshaling
  State string `json:"state,intern"` // will be interned during unmarshaling
}

Issues, Notes, and Limitations

  • tinyjson is still early in its development. As such, there are likely to be bugs and missing features when compared to encoding/json. In the case of a missing feature or bug, please create a GitHub issue. Pull requests are welcome!

  • Unlike encoding/json, object keys are case-sensitive. Case-insensitive matching is not currently provided due to the significant performance hit when doing case-insensitive key matching. In the future, case-insensitive object key matching may be provided via an option to the generator.

  • tinyjson makes use of unsafe, which simplifies the code and provides significant performance benefits by allowing no-copy conversion from []byte to string. That said, unsafe is used only when unmarshaling and parsing JSON, and any unsafe operations / memory allocations done will be safely deallocated by tinyjson. Set the build tag tinyjson_nounsafe to compile it without unsafe.

  • tinyjson is compatible with Google App Engine. The appengine build tag (set by App Engine's environment) will automatically disable the use of unsafe, which is not allowed in App Engine's Standard Environment. Note that the use with App Engine is still experimental.

  • Floats are formatted using the default precision from Go's strconv package. As such, tinyjson will not correctly handle high precision floats when marshaling/unmarshaling JSON. Note, however, that there are very few/limited uses where this behavior is not sufficient for general use. That said, a different package may be needed if precise marshaling/unmarshaling of high precision floats to/from JSON is required.

  • While unmarshaling, the JSON parser does the minimal amount of work needed to skip over unmatching parens, and as such full validation is not done for the entire JSON value being unmarshaled/parsed.

  • Currently there is no true streaming support for encoding/decoding as typically for many uses/protocols the final, marshaled length of the JSON needs to be known prior to sending the data. Currently this is not possible with tinyjson's architecture.

  • tinyjson parser and codegen based on reflection, so it won't work on package main files, because they cant be imported by parser.

Benchmarks

Most benchmarks were done using the example 13kB example JSON (9k after eliminating whitespace). This example is similar to real-world data, is well-structured, and contains a healthy variety of different types, making it ideal for JSON serialization benchmarks.

Note:

  • For small request benchmarks, an 80 byte portion of the above example was used.

  • For large request marshaling benchmarks, a struct containing 50 regular samples was used, making a ~500kB output JSON.

  • Benchmarks are showing the results of tinyjson's default behaviour, which makes use of unsafe.

Benchmarks are available in the repository and can be run by invoking make.

tinyjson vs. encoding/json

tinyjson is roughly 5-6 times faster than the standard encoding/json for unmarshaling, and 3-4 times faster for non-concurrent marshaling. Concurrent marshaling is 6-7x faster if marshaling to a writer.

tinyjson vs. ffjson

tinyjson uses the same approach for JSON marshaling as ffjson, but takes a significantly different approach to lexing and parsing JSON during unmarshaling. This means tinyjson is roughly 2-3x faster for unmarshaling and 1.5-2x faster for non-concurrent unmarshaling.

As of this writing, ffjson seems to have issues when used concurrently: specifically, large request pooling hurts ffjson's performance and causes scalability issues. These issues with ffjson can likely be fixed, but as of writing remain outstanding/known issues with ffjson.

tinyjson and ffjson have similar performance for small requests, however tinyjson outperforms ffjson by roughly 2-5x times for large requests when used with a writer.

tinyjson vs. go/codec

go/codec provides compile-time helpers for JSON generation. In this case, helpers do not work like marshalers as they are encoding-independent.

tinyjson is generally 2x faster than go/codec for non-concurrent benchmarks and about 3x faster for concurrent encoding (without marshaling to a writer).

In an attempt to measure marshaling performance of go/codec (as opposed to allocations/memcpy/writer interface invocations), a benchmark was done with resetting length of a byte slice rather than resetting the whole slice to nil. However, the optimization in this exact form may not be applicable in practice, since the memory is not freed between marshaling operations.

tinyjson vs 'ujson' python module

ujson is using C code for parsing, so it is interesting to see how plain golang compares to that. It is important to note that the resulting object for python is slower to access, since the library parses JSON object into dictionaries.

tinyjson is slightly faster for unmarshaling and 2-3x faster than ujson for marshaling.

Benchmark Results

ffjson results are from February 4th, 2016, using the latest ffjson and go1.6. go/codec results are from March 4th, 2016, using the latest go/codec and go1.6.

Note these benchmarks are outdated and currently don't run (I broke them removing the floats). We should get them back to compare with the current minimal state of tinyjson.

Unmarshaling

lib json size MB/s allocs/op B/op
standard regular 22 218 10229
standard small 9.7 14 720
tinyjson regular 125 128 9794
tinyjson small 67 3 128
ffjson regular 66 141 9985
ffjson small 17.6 10 488
codec regular 55 434 19299
codec small 29 7 336
ujson regular 103 N/A N/A

Marshaling, one goroutine.

lib json size MB/s allocs/op B/op
standard regular 75 9 23256
standard small 32 3 328
standard large 80 17 1.2M
tinyjson regular 213 9 10260
tinyjson* regular 263 8 742
tinyjson small 125 1 128
tinyjson large 212 33 490k
tinyjson* large 262 25 2879
ffjson regular 122 153 21340
ffjson** regular 146 152 4897
ffjson small 36 5 384
ffjson** small 64 4 128
ffjson large 134 7317 818k
ffjson** large 125 7320 827k
codec regular 80 17 33601
codec*** regular 108 9 1153
codec small 42 3 304
codec*** small 56 1 48
codec large 73 483 2.5M
codec*** large 103 451 66007
ujson regular 92 N/A N/A

* marshaling to a writer, ** using ffjson.Pool(), *** reusing output slice instead of resetting it to nil

Marshaling, concurrent.

lib json size MB/s allocs/op B/op
standard regular 252 9 23257
standard small 124 3 328
standard large 289 17 1.2M
tinyjson regular 792 9 10597
tinyjson* regular 1748 8 779
tinyjson small 333 1 128
tinyjson large 718 36 548k
tinyjson* large 2134 25 4957
ffjson regular 301 153 21629
ffjson** regular 707 152 5148
ffjson small 62 5 384
ffjson** small 282 4 128
ffjson large 438 7330 1.0M
ffjson** large 131 7319 820k
codec regular 183 17 33603
codec*** regular 671 9 1157
codec small 147 3 304
codec*** small 299 1 48
codec large 190 483 2.5M
codec*** large 752 451 77574

* marshaling to a writer, ** using ffjson.Pool(), *** reusing output slice instead of resetting it to nil

About

Fast JSON serializer for golang.

Resources

License

Stars

Watchers

Forks

Packages

No packages published

Languages

  • Go 99.0%
  • Other 1.0%