Support parsing application/json-seq

[spbnick]

Support parsing RS-separated streams, as per RFC 7464.

Hi Michael, here are some (possibly) useful changes separated from my stream parsing support, freshly rebased.

Would be glad to hear your comments and fix it up as necessary!

[spbnick]

CI failed seemingly due to pypi.org certificate verification issues.

[spbnick]

Looks like the only broken test job is Build / tests (ubuntu-20.04, 3.5, false), and that's due to SSL.

[mwilliamson]

Just to check I've understood, it seems like there are two main changes here:

1. Using a JSON text sequence as an input
2. Outputting a JSON text sequence

Specifically, when passing in seq=True to the various input methods, any input text will be treated as a JSON text sequence, and any text output from .text() will be formatted as a JSON text sequence.

Have I understood that correctly?

[spbnick]

Just to check I've understood, it seems like there are two main changes here:

1. Using a JSON text sequence as an input
2. Outputting a JSON text sequence

Specifically, when passing in seq=True to the various input methods, any input text will be treated as a JSON text sequence, and any text output from .text() will be formatted as a JSON text sequence.

Have I understood that correctly?

Yes. I did it this way, because the jq tool does it this way, when you supply the --seq option. In my own code I separate these. I'm up for redoing this the way you'd prefer it.

Thanks for the review!

[mwilliamson]

I think you could make an argument for a seq argument applying to both the input and output, since, as you say, that matches jq, but given the Python bindings have a explicit input methods, you might reasonably expect that only to apply to the input.

Given the existing API, I think the least ambiguous approach might be to leave the existing methods as they are, and instead add two new methods, input_text_sequence() (for the input) and text_sequence() (for the output). What do you think?

[spbnick]

Given the existing API, I think the least ambiguous approach might be to leave the existing methods as they are, and instead add two new methods, input_text_sequence() (for the input) and text_sequence() (for the output). What do you think?

That works for me, updated!

[spbnick]

There's still SSL certificate verification problem in Build / tests (ubuntu-20.04, 3.5, false) (pull_request). I just removed the 3.5 from the matrix in my fork and the rest worked.

[mwilliamson]

I've just dropped support for Python 3.5, seeing as we don't build wheels for it anyway and usage seems low.

[mwilliamson]

I think this is good to go with the new methods. I'd prefer not to add slurp to the other input methods though: I think they make less sense outside of the context of text input (since you can just wrap the value in a list), and I think are unrelated to parsing text sequences.

There's also missing tests for outputting text sequences, but I'm happy to add those tests once this is merged.

[spbnick]

Sorry, I'm on a conference trip and don't have time to continue this right now. I'll get back to this the week after next one.