Implement RawValue type #480
Conversation
|
Awesome! Thank you. 👍 |
| serde_json::from_str::<Wrapper>(r#"{"a": 1, "b": {"foo": 2}, "c": 3}"#).unwrap(); | ||
| assert_eq!(r#"{"foo": 2}"#, wrapper_from_str.b.as_ref()); | ||
|
|
||
| let wrapper_from_reader = serde_json::from_reader::<_, Wrapper<'static>>( |
There was a problem hiding this comment.
This is a compiler bug, this line shouldn't compile. I filed rust-lang/rust#54302 to follow up. Still thinking about what that means for how we want to expose RawValue.
There was a problem hiding this comment.
Hah, of course, not sure what I thought this would do!
Does that mean we might want to have an owned and a borrowed variant of RawValue?
There was a problem hiding this comment.
I believe so. I opened #485 to experiment with that design. Let me know what you think.
The way |
|
Ah, I was looking for an explanation like that, might be worth adding a code comment to make this clear to anyone coming across this in the future. As I said on the other PR (#485), do let me know if I can help move this forward in any way! |
Implements the
RawValuetype described in #355. On this benchmark, using theRawValuetype instead of deserializing/serializing viaValueresults in a 3-4x speed-up.I did not adhere fully to the way that the existing arbitrary precision feature is implemented, because I found it easier to use newtypes rather than structs to pass the raw data back and forth. If you want, I'm happy to either change it so that it uses the same approach as the arbitrary precision feature, or to change the arbitrary precision feature to use this (imo) simpler approach.
In the documentation for
RawValueI mentioned two important caveats which apply for this implementation:RawValuewill retain its original formatting and will not be minified or pretty-printed.RawValuecan not be used with the#[serde(flatten)]attribute, as it relies on the original input buffer.What you described in #323 might be a cleaner solution to these problems, but it also sounds more complex and resource hungry (using 4x-8x the amount of memory).
Let me know what you think!