Modify code to allow :initform to be used for unspecified fields.

[smcallis]

When specifying a class slot, we're allowed to pass an :initform
specifying a form to use as the initial value of the slot when it's
not passed to the constructor.

Currently, if we we specify a slot with jeison-defclass, the behavior
is to require that field be present in the decoded json. This is a
problem for situations where fields may have a default value, but that
default isn't necessarily encoded in the payload (eg: gRPC). Thus we
change the default behavior to allow fields to not be present.

This is a two part fix:

1. modify jeison--read-internal to allow null fields to skip type check
2. modify jeison--read-slot to return nil when a slot value is null

Then, when jeison--read-class applies the constructor to initialization
values, the unspecified fields will be absent and the constructor will
use their :initform instead.

[smcallis]

Any thoughts on merging this?

[patrickt]

I ran into this and would love to see it fixed.

[smcallis]

Sadly this repo's had no commits for three years so I fear the worst :(

[bscottm]

@patrickt, @smcallis: It does look like this particular repo has hit a dead end. I'm going to fork it -- I added support for a vector-of type specification (yeah, that's actually useful) and I've integrated this merge request. I'm not ready to publish my fork (and I haven't pushed my updates either). Prospectively, it'll be on bscottm/decljson in the next day or two.

Are there any other enhancements you'd like to see?

[smcallis]

Nope mostly compatibility with gRPC was what I was after, so default values is all I need.

[bscottm]

@smcallis, @patrickt: Pushed updates/enhancements to my fork (bscottm/decljson). In addition to the :type (vector-of type) constraint, there's also :type (hashtab-of type) constraint, support both native and json.el parsers.

[SavchenkoValeriy]

Hi there folks, sorry I got out of my hibernation :)
@smcallis would you be able to add a couple of tests and a short description for the README?

As part of the consideration, maybe we can make this behaviour optional? Like mark a slot that it might be absent in the actual JSON and that some other value should be used in that situation.
It might come as a surprise for users who rely on stricter checks right?

[SavchenkoValeriy]

@bscottm Hey, I see you worked a bit on your own fork. Sorry for being so unresponsive.
Would you mind creating PRs with your changes?

[bscottm]

@SavchenkoValeriy: I incorporated this pull request into my fork, so you can effectively ignore this one.