commit | 2a8ea663c65d95eb7219dd65fe699b519944ffb5 | [log] [tgz] |
---|---|---|
author | Michael Bolin <mbolin@fb.com> | Thu Jan 24 07:48:09 2019 |
committer | Michael Bolin <mbolin@fb.com> | Thu Jan 24 18:09:52 2019 |
tree | 43f059a34327c74776fd68a54c62d7ac29eff331 | |
parent | 493bad102fa42fea6f1365fc5809bacfbd423adb [diff] |
[RFC] Add an option to allow C99-like comments. This would introduce a `serde_json::Options` type that could be passed to `from_str`, `from_bytes, or `from_reader`. Initially, the only option would be `allow_comments`, though I would also like to add support for `allow_trailing_commas` going forward. I read through https://github.com/serde-rs/json/issues/168 and saw that this was rejected in the past, but I was curious if we could revisit this issue. The suggestion on the issue was to use [Hjson](https://github.com/hjson/hjson-rust), which I think is "too loose" for most peoples' taste. Also it has [not been updated to target serde 1.0.x yet](https://github.com/hjson/hjson-rust/issues/6). Allowing C99-like comments and trailing commas results in an input language that is still valid ECMAScript (it's just version 5.1 instead of 3), which cannot be said for things like Hjson. Because it is familiar to JavaScript developers, I have seen it used as an input language in a number of domains. For example, all settings files in VS Code appear to allow this: it even has a special language mode to recognize this named "JSON with Comments (jsonc)." It is fairly common for JSON parsers to allow options like this. For example, here is the full set of options supported by Folly's (C++ library) JSON parser: https://github.com/facebook/folly/blob/74ea53886c3333eda4eaf457d538a678ceaa5add/folly/json.h#L56-L72 Google's Gson parser has a `setLenient()` option: https://github.com/google/gson/blob/9d44cbc19a73b45971c4ecb33c8d34d673afa210/gson/src/main/java/com/google/gson/stream/JsonReader.java#L296-L324 There's also https://json5.org/, which tries to formalize the leniency. If `serde_json` is not on board with this direction, I guess I should fork this repo and create `serde_jsonc` or `serde_lenient_json`? I would prefer not to do that, but if you want to keep `serde_json` so that it only supports RFC 8259 or whatever it is, then I suppose forking is the best way forward. One final note: although https://github.com/serde-rs/json/issues/168 notes that comments were "explicitly excluded from JSON," it doesn't mean it was the right choice. I've been arguing against it for years now: http://bolinfest.com/essays/json.html.
Serde is a framework for serializing and deserializing Rust data structures efficiently and generically.
[dependencies] serde_json = "1.0"
You may be looking for:
#[derive(Serialize, Deserialize)]
JSON is a ubiquitous open-standard format that uses human-readable text to transmit data objects consisting of key-value pairs.
{ "name": "John Doe", "age": 43, "address": { "street": "10 Downing Street", "city": "London" }, "phones": [ "+44 1234567", "+44 2345678" ] }
There are three common ways that you might find yourself needing to work with JSON data in Rust.
Serde JSON provides efficient, flexible, safe ways of converting data between each of these representations.
Any valid JSON data can be manipulated in the following recursive enum representation. This data structure is serde_json::Value
.
enum Value { Null, Bool(bool), Number(Number), String(String), Array(Vec<Value>), Object(Map<String, Value>), }
A string of JSON data can be parsed into a serde_json::Value
by the serde_json::from_str
function. There is also from_slice
for parsing from a byte slice &[u8] and from_reader
for parsing from any io::Read
like a File or a TCP stream.
use serde_json::{Result, Value}; fn untyped_example() -> Result<()> { // Some JSON input data as a &str. Maybe this comes from the user. let data = r#" { "name": "John Doe", "age": 43, "phones": [ "+44 1234567", "+44 2345678" ] }"#; // Parse the string of data into serde_json::Value. let v: Value = serde_json::from_str(data)?; // Access parts of the data by indexing with square brackets. println!("Please call {} at the number {}", v["name"], v["phones"][0]); Ok(()) }
The result of square bracket indexing like v["name"]
is a borrow of the data at that index, so the type is &Value
. A JSON map can be indexed with string keys, while a JSON array can be indexed with integer keys. If the type of the data is not right for the type with which it is being indexed, or if a map does not contain the key being indexed, or if the index into a vector is out of bounds, the returned element is Value::Null
.
When a Value
is printed, it is printed as a JSON string. So in the code above, the output looks like Please call "John Doe" at the number "+44 1234567"
. The quotation marks appear because v["name"]
is a &Value
containing a JSON string and its JSON representation is "John Doe"
. Printing as a plain string without quotation marks involves converting from a JSON string to a Rust string with as_str()
or avoiding the use of Value
as described in the following section.
The Value
representation is sufficient for very basic tasks but can be tedious to work with for anything more significant. Error handling is verbose to implement correctly, for example imagine trying to detect the presence of unrecognized fields in the input data. The compiler is powerless to help you when you make a mistake, for example imagine typoing v["name"]
as v["nmae"]
in one of the dozens of places it is used in your code.
Serde provides a powerful way of mapping JSON data into Rust data structures largely automatically.
use serde::{Deserialize, Serialize}; use serde_json::Result; #[derive(Serialize, Deserialize)] struct Person { name: String, age: u8, phones: Vec<String>, } fn typed_example() -> Result<()> { // Some JSON input data as a &str. Maybe this comes from the user. let data = r#" { "name": "John Doe", "age": 43, "phones": [ "+44 1234567", "+44 2345678" ] }"#; // Parse the string of data into a Person object. This is exactly the // same function as the one that produced serde_json::Value above, but // now we are asking it for a Person as output. let p: Person = serde_json::from_str(data)?; // Do things just like with any other Rust data structure. println!("Please call {} at the number {}", p.name, p.phones[0]); Ok(()) }
This is the same serde_json::from_str
function as before, but this time we assign the return value to a variable of type Person
so Serde will automatically interpret the input data as a Person
and produce informative error messages if the layout does not conform to what a Person
is expected to look like.
Any type that implements Serde's Deserialize
trait can be deserialized this way. This includes built-in Rust standard library types like Vec<T>
and HashMap<K, V>
, as well as any structs or enums annotated with #[derive(Deserialize)]
.
Once we have p
of type Person
, our IDE and the Rust compiler can help us use it correctly like they do for any other Rust code. The IDE can autocomplete field names to prevent typos, which was impossible in the serde_json::Value
representation. And the Rust compiler can check that when we write p.phones[0]
, then p.phones
is guaranteed to be a Vec<String>
so indexing into it makes sense and produces a String
.
Serde JSON provides a json!
macro to build serde_json::Value
objects with very natural JSON syntax. In order to use this macro, serde_json
needs to be imported with the #[macro_use]
attribute.
use serde_json::json; fn main() { // The type of `john` is `serde_json::Value` let john = json!({ "name": "John Doe", "age": 43, "phones": [ "+44 1234567", "+44 2345678" ] }); println!("first phone number: {}", john["phones"][0]); // Convert to a string of JSON and print it out println!("{}", john.to_string()); }
The Value::to_string()
function converts a serde_json::Value
into a String
of JSON text.
One neat thing about the json!
macro is that variables and expressions can be interpolated directly into the JSON value as you are building it. Serde will check at compile time that the value you are interpolating is able to be represented as JSON.
let full_name = "John Doe"; let age_last_year = 42; // The type of `john` is `serde_json::Value` let john = json!({ "name": full_name, "age": age_last_year + 1, "phones": [ format!("+44 {}", random_phone()) ] });
This is amazingly convenient but we have the problem we had before with Value
which is that the IDE and Rust compiler cannot help us if we get it wrong. Serde JSON provides a better way of serializing strongly-typed data structures into JSON text.
A data structure can be converted to a JSON string by serde_json::to_string
. There is also serde_json::to_vec
which serializes to a Vec<u8>
and serde_json::to_writer
which serializes to any io::Write
such as a File or a TCP stream.
use serde::{Deserialize, Serialize}; use serde_json::Result; #[derive(Serialize, Deserialize)] struct Address { street: String, city: String, } fn print_an_address() -> Result<()> { // Some data structure. let address = Address { street: "10 Downing Street".to_owned(), city: "London".to_owned(), }; // Serialize it to a JSON string. let j = serde_json::to_string(&address)?; // Print, write to a file, or send to an HTTP server. println!("{}", j); Ok(()) }
Any type that implements Serde's Serialize
trait can be serialized this way. This includes built-in Rust standard library types like Vec<T>
and HashMap<K, V>
, as well as any structs or enums annotated with #[derive(Serialize)]
.
It is fast. You should expect in the ballpark of 500 to 1000 megabytes per second deserialization and 600 to 900 megabytes per second serialization, depending on the characteristics of your data. This is competitive with the fastest C and C++ JSON libraries or even 30% faster for many use cases. Benchmarks live in the serde-rs/json-benchmark repo.
Serde developers live in the #serde channel on irc.mozilla.org
. The #rust channel is also a good resource with generally faster response time but less specific knowledge about Serde. If IRC is not your thing, we are happy to respond to GitHub issues as well.
This crate currently requires the Rust standard library. For JSON support in Serde without a standard library, please see the serde-json-core
crate.
Serde JSON is licensed under either of
at your option.
Unless you explicitly state otherwise, any contribution intentionally submitted for inclusion in Serde JSON by you, as defined in the Apache-2.0 license, shall be dual licensed as above, without any additional terms or conditions.