ValidPathInfo field does not adhere to the JSON guidlines

[Ericson2314]

ca field

Strings like fixed:r:sha256:adsfjlawefjliajsdl need to be parsed, and that is no good. We should use an object instead, something like

{
    "method": "nar",
    "hash": "sha256-adsfjlawefjliajsdl"
}

This should be implemented as a canonical nlohmann JSON "instance" for ContentAddress (and friends), with unit test characterization tests checking round trips. The ValidPathInfo JSON methods can then just use that nlohmann JSON instance.

Noticed while working on #10310 -- comparing the ca field in the tests was way more hackish than it should be.

signatures field

The <name>:<pub-key> format should be replaced with a two-field JSON object.

Other info

See also #9866 for something similar

CC @roberth

[roberth]

I don't think structured strings (oxymoron?) technically violate the JSON guidelines, but otherwise it seems that what you're proposing is an improvement.

Should the hash be parsed into a record object, or should we use SRI hash strings, or both?

[Ericson2314]

Should the hash be parsed into a record object, or should we use SRI hash strings, or both?

Good question. I am fine either way. At least the SRI string is not our own homegrown convention.