Should the json and jsonb types be treated more like bytea under pg_enable_utf8?

[rafl]

DBD::Pg provides a pg_enable_utf8 option to automatically decode textual values from bytes to character strings when your database encoding is utf8. This is great!

DBD::Pg also goes out of its way to treat values of type bytea specially when pg_enable_utf8 is enabled in order to not attempt to decode arbitrary binary data as if it was utf8 encoded data. This too is really helpful and what I'd expect for bytea values.

As per https://datatracker.ietf.org/doc/html/rfc8259#section-8.1, I'd expect JSON data to be a sequence of bytes in utf8 encoding and would've expected any jsonb data I retrieve using DBD::Pg to be valid JSON which I can pass to the standard decode_json functions from modules like JSON::XS, Cpanel::JSON::XS, Mojo::JSON, etc, independently of whether or not I chose to automatically decode textual data to character strings with pg_enable_utf8. Having to use different JSON encoding/decoding functions depending on the pg_enable_utf8 setting doesn't seem ideal.

I'm wondering if the notion of byte-like types to not be decoded under pg_enable_utf8, which currently only includes bytea, should perhaps be (optionally?) extended to also include jsonb and/or json and would like to hear others' thoughts on that.

I'd be happy to provide a patch to that effect, but I thought it'd be better to reach out first to discuss what such a change would look like, as it has potential implications on backwards compatibility and might need to be something to explicitly opt into by users of the library.

Thanks!

[jonjensen]

I see your point, and it makes sense, but I expect DBD::Pg to return a native Perl hash or array from a jsonb or json type, to do the work or unmarshaling as it does with arrays, not for me to have to manually decode it.

[rafl]

@jonjensen I think what you're asking for is more related to #32 than the issue I'm trying to address, though I can certainly see how allowing json/jsonb types to not be treated as character strings could be a first helpful step in the direction of that larger goal.

[jonjensen]

@rafl Yes, I think you're right. We want to hand raw bytes to any JSON decoder anyway, whether we manually decode or have DBD::Pg do it automatically.