Object of type 'AtomicInterval' is not JSON serializable

[gfitas]

Hi

I was wondering if it will be possible to make Interval and AtomicInterval JSON serializable ? 🙏 😄

Thank you in advance

[AlexandreDecan]

Hi!

I will have a look at what is required to support json serialization. That shouldn't be difficult to implement, but maybe there's a more generic way to support different (other) kinds of serialization.

I'll keep you in touch.

[AlexandreDecan]

It seems there is no "hook" that can be written on Interval or AtomicInterval to seamlessly support (de)serialization. However, I already see two workarounds:

• The first one is to implement new methods/functions that convert Interval instances from/to a Python built-in data type (e.g. tuples). The question then is: how to convert the bounds of an atomic interval, as they can be any custom objects;
• The second one, which is already available, is to rely on the from_string and to_string functions so that you can convert Interval instances from/to strings, and store them as strings in JSON. In that case, custom objects are explicitly supported through the conv attribute. However, this approach could be less straightforward for custom types (but very convenient, e.g., for integers).

Could you give a try to the latter solution?

[gfitas]

In my case, I use your library in an python Api that communicate in json. So if I use your last solution, I would have to parse the string intervals from JavaScript. Maybe you could generate array of custom object like this :
[{lowerV: 2, lowerT:'close',upperV: 4, upperT:'open'},
{lowerV: 5, lowerT:'open',upperV: 6, upperT:'close'},...]

Easy to read, easy to parse, easy to apply from any languages

[AlexandreDecan]

The problem is to encode/decode arbitrary objects that could be used as bounds. Integers are supported out of the box by json, but not, e.g., Timestamp.

Notice that your example is already "supported" with a bit of Python:

>>> i = I.closed(3, 4) | I.closed(7, 9)
>>> [(x.left, x.lower, x.upper, x.right) for x in i]
[(True, 3, 4, True), (True, 7, 9, True)]

>>> v = [(True, 3, 4, True), (True, 7, 9, True)]
>>> I.Interval(*[I.AtomicInterval(*x) for x in v])
[3,4] | [7,9]

I can add functions that do this job. They should also accept a parameter to specify how the bounds (arbitrary objects) can be converted from/to built-in data types. For example:

>>> i = I.closed(Timestamp(3), Timestamp(4)) | I.closed(Timestamp(7), Timestamp(9))
>>> I.to_data(i, conv=lambda t: t.as_int())
[(True, 3, 4, True), (True, 7, 9, True)]

>>> v = [(True, 3, 4, True), (True, 7, 9, True)]
>>> I.from_data(v, conv=Timestamp)
>>> [Timestamp(3),Timestamp(4)] | [Timestamp(7),Timestamp(9)]

[gfitas]

Yes that will be super great

[AlexandreDecan]

I opened a PR (#7). I still need to find why doctest is failing in Python 2.7.
I will probably bundle another new feature with release 1.7.0, I hope to find some time to work on it today. If so, you can expect 1.7.0 to be released by the end of the day.

[AlexandreDecan]

Closed by #7

[AlexandreDecan]

1.7.0 will be released in a few minutes, with from_data and to_data.