Retrieving raw text from subobjects

[raphgianotti]

Description

For some of the code I am using nlohmann for I need a way to retrieve the raw/original text of a nested object within the json object I am parsing. This can be useful, for example, in cases where one may have a json that includes a nested object and its hash, where having a way to retrieve the raw text of that nested object would prevent having to set a strict format for the text that needs to be hashed (considering setting such strictness on the formatting can make things less usable).

Example

Given the following json:

{
    "name": "foo",
    "data":
    {
        "type": "typeA",
        "value": 1,
        "details": {
            "nested_type": "nested_typeA",
            "nested_value": 2
        }
    },
    "data_hash": "hashA"
}

I would like to be able to retrieve the exact text of "details" as it is (considering whitespaces, order, line breaks, etc.), so I can verify its hash matches "data_hash".

How this could be done

I have looked at the code and couldn't find a way to do this.

I believe it would be possible to add a new parser callback or extend the signature of the parser callback to include information about the character position in the original text for where an object starts/ends based on the type of parsing event (e.g., at object_start and object_end).

The SAX interface also seems like a possible candidate to be extended to provide this information while parsing an object.

For the internal implementation, I believe this could use information already held by the lexer to return the location in the original text of the elements being parsed.

I wanted to know if this is something that is possible and I am missing something, or whether it would be a contribution I could work on making to the project. And for the latter, whether the above suggestions seem like reasonable approaches to making this change.

[xkommando]

second to this. my cases:

1. i need to parse numbers safely, ie check for integer&float overflow, right now I'm writing boilerplate code to check for overflows but i'd like to use std charconv for this
2. parse number in scientific notations, eg. { "f64": "3.40282e+38" }

[nlohmann]

Some thoughts:

• Whitespace is currently discarded right in the lexer and is not preserved for following steps.
• For floating-point numbers (or numbers with scientific notation), we pass the raw string to the SAX callback.
• Everything else would not benefit from requesting the raw text, because the values would be implicit (true, false, null, integers) or already explicitly provided (strings).

[raphgianotti]

Replied in the PR thread, linking it here: #4517 (comment)

[nlohmann]

FYI: #4517 was merged and this feature is now available.