UTC Plugin calculates incorrect results when using diff() in months (starting in v1.8.33)

[KosnIre]

Describe the bug
Using the diff() method, in months, returns incorrect results when using the UTC plugin starting in v1.8.33. It is also returning different results when using the plugin vs without.

The result appears to be off by one month in some cases.

NOTE: At first I thought it was related to #1000 but I think this is a different issue. For one, it affects a different version (1.8.33 here vs 1.8.32 there) and also that issue does not reference the UTC plugin. I also note that there were a few UTC related changes in 1.8.33.

Expected result: The difference between May and December should be calculated as 7 months.

Actual result: The difference between May and December is calculated as 6 months.

Information

• Affected Day.js Versions: 1.8.33, 1.8.34
• Unaffected Day.js Versions: 1.8.32 and below
• OS: Ubuntu 18.04
• Node v12.18.2
• Time zone: GMT-04:00 EDT (Eastern Time)

[KosnIre]

Test code

const dayjs = require('dayjs');
const utc = require('dayjs/plugin/utc');

dayjs.extend(utc);


/**
 * calcDateDiff - Test method that finds the difference between 2 ISO dates using dayjs.
 */ 
function calcDateDiff({ startDateISO, endDateISO, timeUnits }, utc=false) {
    let startDate, endDate;
    
    if (utc) {
        startDate = dayjs.utc(startDateISO); // convert to dayjs
        endDate = dayjs.utc(endDateISO);
    } else {
        startDate = dayjs(startDateISO); // convert to dayjs
        endDate = dayjs(endDateISO);
    }

    const dateDiff = endDate.diff(startDate, timeUnits);
    
    return dateDiff;
}

// Create test data

const info = {
    endDateISO: '2019-12-15T11:00:00',
    timeUnits: 'month',
};

const data = [];
let startDateISO;

// Add 12 start dates to our test data, going further into the past so we can compare results. 
for (let i=12; i >=1; i--) {
    const leadingZero = i < 10 ? '0' : '';
    startDateISO = `2019-${leadingZero}${i}-15T11:00:00`;
    data.push({
        ...info,
        startDateISO,
    });
}

// Run the test and ouput results to the console
for (const d of data) {
    const dateDiff = calcDateDiff(d);
    const dateDiffUTC = calcDateDiff(d, true);
    
    console.log('Start date:', d.startDateISO, 'End date:', d.endDateISO, 'diff', dateDiff, 'diff-UTC', dateDiffUTC);
}

Test results
Test results shown from v1.8.33. Note the series labeled "diff" (0, 1, 2, 3) vs the series labeled "diff-UTC" (0, 1, 1, 2). Oddly the sequences re-align at 10 and 11.

Start date: 2019-12-15T11:00:00 End date: 2019-12-15T11:00:00 diff 0 diff-UTC 0
Start date: 2019-11-15T11:00:00 End date: 2019-12-15T11:00:00 diff 1 diff-UTC 1
Start date: 2019-10-15T11:00:00 End date: 2019-12-15T11:00:00 diff 2 diff-UTC 1
Start date: 2019-09-15T11:00:00 End date: 2019-12-15T11:00:00 diff 3 diff-UTC 2
Start date: 2019-08-15T11:00:00 End date: 2019-12-15T11:00:00 diff 4 diff-UTC 3
Start date: 2019-07-15T11:00:00 End date: 2019-12-15T11:00:00 diff 5 diff-UTC 4
Start date: 2019-06-15T11:00:00 End date: 2019-12-15T11:00:00 diff 6 diff-UTC 5
Start date: 2019-05-15T11:00:00 End date: 2019-12-15T11:00:00 diff 7 diff-UTC 6
Start date: 2019-04-15T11:00:00 End date: 2019-12-15T11:00:00 diff 8 diff-UTC 7
Start date: 2019-03-15T11:00:00 End date: 2019-12-15T11:00:00 diff 9 diff-UTC 8
Start date: 2019-02-15T11:00:00 End date: 2019-12-15T11:00:00 diff 10 diff-UTC 10
Start date: 2019-01-15T11:00:00 End date: 2019-12-15T11:00:00 diff 11 diff-UTC 11

[iamkun]

Seems ok here ? https://runkit.com/embed/7xucw8zhictx

[KosnIre]

Hmm, yes that runkit runs ok for me too. That's interesting 🤔

I see the node version is v10.22.0 on runkit, so I tried switching to that locally, but I still get incorrect results.

Would you please try running this fiddle? It gives me incorrect results in both Firefox and Chrome.

https://jsfiddle.net/1o4apxgv/

[iamkun]

Screenshot from Chrome. Interesting 🤔

[KosnIre]

Well this is odd. Clearly there is some difference here, so I'm trying to figure out what that is.

For me, the fiddle outputs...

I asked a few otheres to run this test and here are some additional results for reference:

User 1: Windows, Chrome, UTC-04:00 (DST, yes). Result: Fail.
User 2: macOS, Chrome, UTC-04:00 (DST, yes). Result: Fail.
User 3: macOS, Chrome, UTC+2:00 (DST, yes). Result: Fail.
User 4: Windows, Chrome, UTC+5:30 (DST, no). Result: Pass.

My initial thought was this could have something to do with local time zone, but now I'm curious if it's related to Daylight Savings Time observance. I think I'll try to write some tests that parse the date in different timezones to see if I can find a stronger correlation.

One question: how does dayjs() internally diff() months? Is 30 days equal to one month? Is it subtracting milliseconds and converting back to months? Or is it using the native getMonth() API and subtracting the result?

[iamkun]

You can check the month diff function here

https://github.com/iamkun/dayjs/blob/dev/src/utils.js#L17

[ygj6]

My initial thought was this could have something to do with local time zone, but now I'm curious if it's related to Daylight Savings Time observance.

@KosnIre It seems that this difference is caused by Daylight Savings Time observance.

[iamkun]

🎉 This issue has been resolved in version 1.9.5 🎉

The release is available on:

• npm package (@latest dist-tag)
• GitHub release

Your semantic-release bot 📦🚀