Formatter throws variable length error

[chrisjenx]

Process: horse.haynet.debug, PID: 10367
java.lang.IllegalArgumentException: At most one variable-length numeric field in a row is allowed, but got several: [year, hourOfAmPm]. Parsing is undefined: for example, with variable-length month number and variable-length day of month, '111' can be parsed as Jan 11th or Nov 1st.
	at kotlinx.datetime.internal.format.parser.NumberSpanParserOperation.<init>(ParserOperation.kt:56)
	at kotlinx.datetime.internal.format.parser.ParserKt.concat$simplify(Parser.kt:69)
	at kotlinx.datetime.internal.format.parser.ParserKt.concat$simplify(Parser.kt:76)
	at kotlinx.datetime.internal.format.parser.ParserKt.concat(Parser.kt:125)
	at kotlinx.datetime.internal.format.ConcatenatedFormatStructure.parser(FormatStructure.kt:233)
	at kotlinx.datetime.internal.format.CachedFormatStructure.<init>(FormatStructure.kt:248)
	at kotlinx.datetime.format.AbstractDateTimeFormatBuilder.build(DateTimeFormatBuilder.kt:476)
	at kotlinx.datetime.format.DateTimeComponents$Companion.Format(DateTimeComponents.kt:59)

Example formatter:

format(
        format = DateTimeComponents.Format {
            monthName(MonthNames.ENGLISH_ABBREVIATED)
            chars(" ")
            day(Padding.NONE)
            chars(" ")
            year(Padding.ZERO)
            amPmHour(Padding.NONE)
            minute();
            amPmMarker("am", "pm")
        },

AFAIK, this is incorrect, or the error is wrong at minimum? As there are two variable lengths, day of month and amPmHour (as they are padding none), but year (should) expects 4 digits not 2-4.

Also in a formatting situation, I don't care about parsing, I'm formatting, so it would be really nice to skip the parser check when I'm only outputting. If parsing that date, 100% agree could potentially be wrong.

[chrisjenx]

Update: I see it's because they are back to back with each other, adding chars(" ") or some other spacer fixes it.

I see what the issue is highlighting tho, i.e. 202111 is correct, but if someone typo'd and put 20211 you get 2021 1pm vs 2021 11pm

This isn't a parsing issue in this case as you would always strip the first 4 digits for the year and be left with the time next, the formatter needs to be clearer what year() is or give us an option on if it can be variable length...

[dkhalanskyjb]

year (should) expects 4 digits not 2-4.

This isn't a parsing issue in this case as you would always strip the first 4 digits for the year and be left with the time next

year(Padding.ZERO) can be 4+ characters. For example, +1500011: is it +150001 with 1, or is it +15000 with 11? The logic of taking the first 4 characters doesn't help here.

Also in a formatting situation, I don't care about parsing, I'm formatting

The logic behind throwing this error when creating the format instead of just when parsing is that formatting something in an ambiguous format that can not be reasonably parsed is probably an error, too.

the formatter needs to be clearer what year() is

Here's the documentation for year(): https://kotlinlang.org/api/kotlinx-datetime/kotlinx-datetime/kotlinx.datetime.format/-date-time-format-builder/-with-year-month/year.html Is something missing, or can the clarity be improved?

give us an option on if it can be variable length...

We have carefully evaluated this option when designing the API and decided against it.

Currently, the kotlinx-datetime formatting API guarantees that if all the required fields are present, formatting should not fail. Formatting is used in places with high reliability requirements (like logging), and if some broken data enters the system, we want to at least see what data that was, instead of the data also crashing the formatter. Therefore, having formatters throw an exception for LocalDate(year = 15000, month = 1, day = 1) is something that we would like to avoid.

Other options include not throwing and instead outputting something, but in a way, that's even worse, as then, broken data is indistinguishable visually from normal data, complicating the diagnostics.

[chrisjenx]

Interesting,
I think my previous two thoughts remain then;

• Allow formatter only, would avoid this where I only care about output so there isn't really potential conflict
• Or allow stronger definitions for year length, you've already deviated from unicode at this point might as well make the API more useful/strict/defined.

[dkhalanskyjb]

Sure, we can consider different approaches, but could you please first describe the problem that we are solving? The format you've provided in the initial message looks artificial to me (I can't imagine someone wanting to print the year and the minute-of-hour side-by-side). Is there an actual format that you have where year() neighbors another variable-length numeric value without any separator between them?