TypeError when formatting interval

[1ec5]

The format() function throws a TypeError when passing in an interval. For example:

edtf.format(edtf.default('2000/2001'))

throws:

Uncaught TypeError: Cannot read properties of undefined (reading 'value')

It might be possible to format some intervals natively using Intl.DateTimeFormat.prototype.formatRange(). However, this API doesn’t accept other kinds of intervals, such as /2001 and ../2001. Ultimately, I’d like to be able to rely on this library to output something like “before 2001”.

[inukshuk]

Oh, I missed formatRange, that's a good point! Since the formatting is not part of the EDTF spec the current implementation was just a quick stab at getting some reasonable outputs for most cases, with the idea to stay as close as possible to Intl.DateTimeFormat. We should definitely use formatRange for Intervals and then add some special handling for cases which can't be covered this way, like the two examples above.

[1ec5]

Another complication is when an interval is nested inside another interval. For example, a family tree might list someone’s birth and death as “1900 – 1975”, but if the birth date is only known from the person’s age at death, it might need to say something like “1900/1901 – 1975”. (I think that’s the convention in English, anyway.) I assume the format() function would only be responsible for one side of the range at a time, with the application joining it with a hyphen.

[inukshuk]

Intervals can't be nested, but you could use an interval with an uncertain or approximate start. For example, 1900?/1975 -- of course this is less precise than limiting the uncertainty to one of 1901 or 1900 and it sort of depends on the interpretation of uncertain or approximate. I've been considering adding a configuration option for the time computations, so that you can specify that by uncertain you mean +/- 2 and by approximate +/- 1 or something like that. But even that would be less specific than if you want to limit it to exactly two years. For such cases the spec supports Sets, e.g. [1900,1901] would mean either 1900 or 1901 but I don't think you can combine sets and intervals.

[1ec5]

Thinking about it some more, nesting probably wouldn’t be a problem after all.

In the family tree use case I described, the developer probably would know not to use this library to format the overall range, since a person wouldn’t exist in anything other than a range that could be formatted with custom heuristics. So it could similarly add a heuristic to convert the start date’s EDTF interval to a set.

(The hypothetical family tree use case is similar to an actual use case I have in OpenHistoricalMap/issues#981.)