System.Text.Json asynchronous streaming deserialization support

[mqudsi]

Hello team,

We've been eager to try the new System.Text.Json APIs for their native UTF8 support, but it seems based off the preview release announcement and the API exposed by the current preview release of the System.Text.Json that there's a lack of streaming support, i.e. reading individual JSON tokens without first having the entire document loaded into memory (and then parsed as a JsonDocument).

The Newtonsoft JSON library had a JsonTextReader that could lazily consume an underlying Stream and return a token at a time via JsonTextReader.ReadAsync(..), which moved the document cursor/pointer to the next node in the stream. It had some shortcomings (it couldn't skip past a nested node, you had to step through it IIRC) but it served the purpose of being able to navigate asynchronously to a particular node in a JSON document and then selectively parse its value without having to first buffer the entire stream beforehand or to parse the entirety of the document in order to enumerate its nodes.

I assumed that Utf8JsonReader would for sure do the same thing, but it seems that a Utf8JsonReader can only be created on top of an available ReadOnlySequence<T> and all its methods are blocking. JsonDocument can be created on top of a Stream and does not appear to consume the entire stream upon initialization, but only offers a ParseAsync() method that asynchronously reads from the underlying stream; except it consumes the entirety of the stream and loads the entire document into memory.

I'm probably missing something, but is there no way to asynchronously and lazily navigate a UTF8 JSON stream, without buffering the entire document in memory and without blocking unnecessarily for IO?

[mqudsi]

I see now that Utf8JsonReader is re-entrant and so an asynchronous wrapper around Utf8JsonReader that contains additional state can shell out to a fully synchronous helper function that is able to create the needed types and perform the streaming read.

[robertharvey416]

@mqudsi: Can you help this fellow here? What might such an arrangement look like in code?
https://softwareengineering.stackexchange.com/questions/409479

[mqudsi]

@robertharvey416 replied with my (extremely dirty and not cleaned-up) code and an explanation. Thanks for the bat signal.