This article lists several of the absurdities of the Date constructor, but only barely touches on the most unforgivable one. The example from the article is:
// Unless, of course, you separate the year, month, and date with hyphens.
// Then it gets the _day_ wrong.
console.log( new Date('2026-01-02') );
// Result: Date Thu Jan 01 2026 19:00:00 GMT-0500 (Eastern Standard Time)
In this example, the day is "wrong" because the constructor input is being interpreted as midnight UTC on January 2nd, and at that instantaneous point in time, it is 7pm on January 1st in Eastern Standard Time (which is the author's local time zone).
What's actually happening here is a comedy of errors. JavaScript is interpreting that particular string format ("YYYY-MM-DD") as an ISO 8601 date-only form. ISO 8601 specifies that if no time zone designator is provided, the time is assumed to be in local time. The ES5 spec authors intended to match ISO 8601 behavior, but somehow accidentally changed this to 'The value of an absent time zone offset is “Z”' (UTC).
Years later, they had realized their mistakes, and attempted to correct it in ES2015. And you can probably predict what happened. When browsers shipped the correct behavior, they got too many reports about websites which were relying on the previous incorrect behavior. So it got completely rolled back, sacrificed to the altar of "web compatibility."
For more info, see the "Broken Parser" section towards the bottom of this article:
>So it got completely rolled back, sacrificed to the altar of "web compatibility."
This is why I don't understand the lack of directives.
'use strict'; at the top of a file was ubiquitous for a long time and it worked. It didn't force rolling back incompatibilities, it let you opt into a stricter parsing of JavaScript.
It would have been nice for other wide changes like this to have like a 'strict datetime'; directive which would opt you into using this corrected behavior.
They couldn't and shouldn't do this sort of thing for all changes, but for really major changes to the platform this would be an improvement.
Or they could go all in on internal modules, like how you can import `node:fs` now. They could include corrected versions of globals like
To be fair, the new opt-in "use strict" here is "switch to Temporal". It's a new, stricter namespace object. Old Date code gets the old Date code quirks, new code gets the nice new Temporal API.
Internal modules would be handy in theory to maybe keep from having to dig through a thesaurus every time browsers decide to add a new, stricter version of an older API. Internal modules have even been proposed to TC-39 as a recommended way to continue to expand the JS API. Last I checked on that proposal it was stuck behind several concerns including:
1. Feature detection: detecting if Temporal available is as easy as `if ('Temporal' in globalThis) {}`, but detecting if a module import is missing is a bit harder. Right now the standard is that loading a module fails with an Error if one of its imports fails. You can work around that by doing a dynamic import inside a try/catch, but that's a lot of extra boilerplate compared to `const thingINeed = 'someApi' in globalThis ? someApi() : someApiPolyfill()`. I've seen multiple proposals on that front from extensions to import maps and `with { }` options on the import itself.
2. Bikeshedding (and lots of it): defining a URI scheme like `browser:` or `standard:` takes a bunch of thought on how you expand it. If it is just `browser:some-api` you run the risk of eventually polluting all the easy names in the exact way people worry about the risk of over-polluting `globalThis` (and in the way that it can be weirdly hard to find an available one-word name on npm), you've just moved the naming problem from one place to the other. On the other side, if you go down the road of something like `es-standard:https://tc39.es/ecma262/2025/v1/final-draft/Temporal`, even (especially) assuming users would mostly importmap that to something shorter you've recreated XMLNS URIs in a funny new hat and people who use JS all certainly have plenty of opinions on XMLNS URIs, many are very vocal in their hatred of it, but also they came out of a strong backwards incompatibility fixing desire exactly like this. (As they say time is a flat circle.)
> To be fair, the new opt-in "use strict" here is "switch to Temporal".
This. Don't break old code, just provide new best practices.
Update linters (or ideally first class language rules, like in Rust's "edition"s), to gradually kill off old behavior. Without having to do a decade long Python 2 -> 3 migration.
Temporal is nice. It learned from the many failures and dead bodies that came before it. And it had lots of good implementations to look at: Joda Time, Chrono, etc.
PHP suffers from this too. By too strict BC PHP has become a really mess of a language. IIRC there still is the ad-hoc parameter order convention and lack of any namespacing for builtins. Everything is global.
With JS i kind of get it as you cant control the env. Bit PHP does not have this limitation, and they still cant fix the mess that is PHP.
> To be fair, the new opt-in "use strict" here is "switch to Temporal"
Yes, but adding an entire new API that solves way more date-related problems is obviously much more difficult than fixing a very clear and well-understood bug. Temporal is only just now starting to ship, over 15 years after this bug was introduced and over 10 years since they decided to never fix the bug.
nuget has a convention of system packages that are empty if the target platform implements functionality natively and provides independent implementations for platforms that don't support it, as a result you can unconditionally import that package on all platforms.
> It would have been nice for other wide changes like this to have like a 'strict datetime'; directive which would opt you into using this corrected behavior.
That would be ugly, because you'd want some parts of your program (eg libraries) to use the old behaviour, and other parts might want the new behaviour. How would you minify multiple modules together if they all expect different behaviour from the standard library?
In my opinion the right way to do this is to have multiple constructors (as Obj-C, swift, C and rust all do). Eg:
let d = new Date(...) // old behaviour (not recommended for new code)
let d = Date.fromISOString(...) // fixed behaviour
The big downside of this is that its tricky to keep track of which fields and functions people should really stop using in modern javascript. It'd be nice if there was a way to enable more shouty warnings during development for deprecated JS features.
I find it very unfortunate that browsers (or rather, the spec) do not support some kind of versioning. If we could declare which version of HTML, JS and CSS to use, it would allow for breaking changes without breaking the entire web.
There are so many (in hindsight bad) design choices and implementation accidents that currently exist in perpetuity because of backwards compatibility; the web would really benefit if every now and then we could shed old baggage.
It would also force browsers to implement multiple slightly different engine modes, vastly complicating the browser code.
There are already a few cases, eg quirks mode vs standards mode and “use strict” mode, which was considered necessary for moving forward, but clearly it also complicates things for browsers. We dont want more modes than what is necessary.
This was the approach Perl took and much as I love(d) that language, it do get pretty out of hand after a while if you wanted to adopt any newer or stricter language features.
directives sort of kinda work if you squint the eyes, but only as a crutch and only if you can't/don't want to change the API.
> Or they could go all in on internal modules, like how you can import `node:fs` now. They could include corrected versions of globals like
`import Date from 'browser:date';`
This is what happened here, only the API changed as well
I very much remember coding a function that split the string on their components and then rebuild them to ensure the date was created without time zone.
Sometimes a date is just a date. Your birthday is on a date, it doesn't shift by x hours because you moved to another state.
The old Outlook marked birthdays as all-day events, but stored the value with time-zone, meaning all birthdays of people whose birthday I stored in Belgium were now shifted as I moved to California...
I always found it weird when systems code dates as DateTime strings. There needs to be a different primitive for Date, which is inherently timezone-less, and DateTime, which does require a timezone.
After having a bunch of problems with dealing with Dates coded as DateTime, I've begun coding dates as a Date primitive, and wrote functions for calculation between dates ensuring that timezone never creeps its way into it. If there is ever a DateTime string in a Date column in the database, it's impossible to know what the date was supposed to be unless you know you normalized it at some point on the way up.
Then I found that a lot of DatePicker libraries, despite being in "DATE" picker mode, will still append a local timezone to its value. So I had to write a sanitizer for stripping out the TZ before sending up to the server.
That said, I am pretty excited about Temporal, it'll still make other things easier.
The BCL-provided DateTime was really confusing, especially when you just needed a Date. They eventually got around to including a DateOnly, but before that happened I switched to a library called "Noda" (or Joda in Java) and after a bit of a learning curve, it made everything a lot easier to reason about.
It has LocalDates and LocalDateTimes, as well as Instants to store UTC times. It also offers ZonedDateTimes, but I don't use those as much. I work in healthcare. And so many regulations involve strictly dates. Like, "You have 5 days to do X", not "You have 120 hours to do X", and as such, storing the time with a lot of this data can add more complexity.
Wow! And yeah I'd imagine Healthcare requires a bit more strictness, you can't really afford an off by one day error, or sometimes even an off by one hour error.
The calculations themselves are usually pretty easy with a few exceptions, it's just that there are TONS of them. And they all slightly depend on each other. And they change. Often.
There needs to be a difference between an Instant, an Instant at an Observed Location, and a 'Specification for constructing a date that might or might not have already passed or pass in the future'.
E.G. in a conversation "Lets open the shop at 9am every day that it isn't closed." Is a fairly simple recurrence, with some exceptions*. If timezones change the scheduled time remains evaluated again on each day.
Yeah that's a good point, and also takes into account the dreaded DST (what are this business's operating hours for example, which remains the same locally but would change in UTC)
I mean... That's kinda how it works? More than once I've halfway forgotten birthdays of friends who live in timezones to my east, and then sent them a message saying "Happy birthday! (It still is where I am, lol)".
I'm not necessarily defending the implementation, just pointing out another way in which time is irreducibly ambiguous and cursed.
A reminder associated with the birthday can and should be changed if I change time zones. But the birthday date didn’t change so it shouldn’t move to a different day.
> But the birthday date didn’t change so it shouldn’t move to a different day.
But it does. My brother moved to the US for a few years. So we’d send him birthday wishes on the day of his birthday (Australia time), and he’d get them the day before his birthday (his time). Now he’s moved back to Australia, the same thing happens in reverse-he gets birthday wishes from his American friends the day after his birthday.
My wife has lots of American friends on Facebook (none of whom she knows personally, all people she used to play Farmville with)-and she has them wishing her a happy birthday the day after her birthday too. Maybe she’s doing the same to them in reverse.
You reminded me of some riddle I had once read that was about trying to figure out how someone could be born one year later but still be older than someone born in previous year. The answer to the riddle also relies on timezones. For sure, birthdates involve time zones.
The riddle explanation was something like: A baby is born in New York City at 12:15 AM on January 1. Thirty minutes later, another baby is born in Los Angeles, where the local time is 9:45 PM on December 31. Although the New York baby is actually older by 30 minutes, the calendar dates make it appear as though the Los Angeles baby was born first.
The other biggest fun trick of timezone math to a riddle like that would be the International Date line where a baby born on one side of it can be born on the "day before" by calendar reckoning despite being born 30 minutes after the other side of the line.
Fraternal (not identical) twins, born aboard a ship traveling west to east across the Pacific. One of them officially born January 1st, 2016. The younger-by-30-minutes twin officially born December 31st, 2015. They'll have the hardest time persuading people that they're really twins once they're grown up.
Yes, and if you ask any midwife, OB/GYN, or other person who routinely delivers babies, I'm sure you'll hear about plenty of born-on-different-days twins. One of my in-laws is a doctor who delivers babies; she has lots of stories, some of which she's restricted by HIPAA from sharing. But once a baby is born, the baby's birth date is public knowledge so she can share that info. So she often will tell her husband, "My patient is going into labor, I have to go to the hospital" without naming the patient. Then after the baby is born she'll say "Mrs. Smith's baby was born at 11 PM last night" because now it's a matter of public record who the mother was, whereas before it was protected by HIPAA. Next time I talk to her, I'll ask her if she's ever personally delivered any twins with different birthdays.
The timezones thing, of course, is just a way to have the younger twin be born "a year ahead" of the older twin by having their births be in two different timezones. Only practical way that would happen would be aboard a ship, because 1) babies born aboard an airplane would probably end up using the time zone of departure or of destination for their birth, and so twins would not be counted as being born in different time zones. And 2) any land-based transportation such as a car or a train would likely pull over (or in the case of a train, let the pregnant woman off at the nearest station) so that the woman giving birth doesn't have to do so in a moving vehicle. So a ship is the only moving vehicle where this kind of thing could likely happen, as there's no option of getting off in the middle of the ocean. It could happen while crossing time zones east-to-west, but crossing the International Date Line west-to-east makes more of an interesting thought experiment.
Yes, I've given this silly joke scenario way more thought than it really deserves. :-)
Yes, it's the same, the IDL just makes it easier for it to work, as otherwise the babies have to be born on either side of midnight while crossing the time zone line. With the IDL the birth time could be almost any time of day except for crossing over midnight and the joke would work.
I agree with the "don't break the web" design principle, but I sometimes disagree with precisely where TC39 draws the line. There is obviously a cost to breaking old, unchanging websites. But there's also a cost to allowing old, unchanging websites to hold the entire web hostage. Balancing those costs is a subjective matter.
As far as I know, TC39 doesn't have any clear guidelines about how many websites or how many users must be affected in order to reject a proposed change to JavaScript behavior. Clearly there are breaking changes that are so insignificant that TC39 should ignore them (imagine a website with some JavaScript that simply iterates over every built-in API and crashes if any of them ever change).
The decided not to have it there because they didn't like the idea of maintaining version 4.0 forever in their engines.
That's basically why they never did anything like "use strict" again.
IMO, that's a bad choice. Giving yourself the ability to have new behavior and features based on a version is pretty natural and how most programming languages evolve. Having perpetual backwards and fowards compatibility at all times is both hard to maintain and makes it really hard to fix old mistakes.
The only other reason they might have chosen this route is because it's pretty hard to integrate the notion of compatibility levels into minifiers.
Have an optional parameter to opt in to the old behaviour and keep the new correct behaviour the default (without the parameter) seems like a decent choice.
To preserve backwards compatibility and not require all those old sites to update, the legacy behavior would have to be the default, with opt-in for the new behavior.
That is the opposite approach. Also an option. One could also deprecate the call without parameter and force always a parameter with which behaviour. The deprecation could last enough time that those websites would have been rewritten multiple times ;)
The control interface burned into your hardware device will not have been rewritten. And it's not like you can have a flag day where everyone switches over, so the lifespan of those hardware devices isn't that relevant.
Backwards compatibility is a large part of the point of the Web.
You could version everything at whatever granularity you like, but over time that accumulates ("bug 3718938: JS gen24 code incorrectly handles Date math as if it were gen25-34", not to mention libraries that handle some versions but not others and then implicitly pass their expectations around via the objects they create so your dependency resolver has to look at the cross product of versions from all your depencies...)
There is no free lunch. A deprecation warning lasting a decade before erroring will break less that some css boxing models and strict mode in many browsers.
Nah, that's not a "sacrifice", but the only sane way. In the ideal case, clearly document the constructor with a warning that it's not ISO conformant and offer a ISO conformant alternative.
In my (unfortunate) experience, DateTime/Timezone handling is one of the things most prone to introduce sneaky, but far-reaching bugs as it is. Introducing such a behaviour change that (usually) won't fail-fast, will often seemingly continue working as before until it doesn't and is deceptively tricky to debug/pinpoint/fix ist just asking for a fast lane into chaos.
And even with JS going the extra mile on backwards compatibility, I don't think most other languages would introduce that kind of breaking change in that way either.
I think you got it backwards. The only way is guessing, it's all random.
It's the internal inconsistencies that get me. Like, OK, I understand that there might be some quirks, maybe due to some weird backwards compatibility or technical limitation, but there are multiple incompatible quirks _inside_ this single interface! It's terrible, and things like this are a huge part of the reason JS was long considered (and sometimes still is) a Not So Good language.
the new Date() constructor is an amalgamation of like 5 different specs, and unless the input matches one of them, which one kicks in is up to the implementer's choice
I think NaN itself is a bit of an error object, especially in how it's passed through subsequent math functions, which is a different choice than throwing up.
But besides that I think you're right, Invalid Date is pretty weird and I somehow never ran into it.
One consequence is you can still call Date methods on the invalid date object and then you get NaN from the numeric results.
Personally, I like that UTC is the default time zone. Processing of dates should happen in a standardized time zone. It’s only when you want to display it that the date should become local.
UTC is a fine default time zone, but that doesn't matter here.
A datetime with a timezone and a datetime without one are two different things, both of them useful. My birthday does not have a time zone. My deadline does.
The company deadline for getting some document returned? It might or might not, that's policy.
Poetically: we are born free of time zones. We die bound to one.
It doesn't to me. It should be obvious that there are plenty of valid uses of dates and times which implicitly refer to either an exact instant in time, or the expression of a time in a certain reckoning.
A birthday doesn't have a time zone because the concept of a birthday is more about the date on the calendar on the wall, not any universally understood singular instant in time; and so what matters most when it comes to your birthday is where you are. Your birthday doesn't move to the day before or after just because you travel to the other side of the globe.
A deadline has a time zone because when your boss says he wants the project done by 4PM, he means 4PM wherever you both currently are -- and the specific instant in time he referred to doesn't change if you get on a train and move a time zone over before that 4PM occurs.
And it may in fact be time zone and not just UTC with an offset; because if your boss tells you he wants a certain report on his desk by 4PM every day; when your local time zone goes into daylight saving time, it doesn't suddenly mean the report needs to be filed by 5PM instead.
In the first of these cases, the date itself has no time zone and is valid in whatever context its being read from. In the second, the instant in time might be expressed in UTC time with or without a specific offset. In the third, each of the successive instants in time may shift around with respect to UTC even while it continues to be referred to with one constant expression.
None of these are subjective interpretations. They're a consequence of the fact that as humans we've overloaded our representation of date/time with multiple meanings.
idk about you but I can get a happy birthday on the hour of my actual birth from people in the know but I never literally prepare things for the exact hour of the deadline. It's more like a day sort of thing
It does not. I'm Australian and our timezones are ahead of the US (NSW time is about 15-17 hours ahead of US Eastern time). If I took a flight from Sydney to New York (22~ hours) on my birthday, the US custom's officer would wish me happy birthday when I landed the next day.
Therefore, birthdays are not bound by timezone at all.
This will result in incorrect behavior when, between converting to UTC and back to the original timezone, the timezone database has changed, which happens more often than you think.
Depends what you're actually storing. There are plenty of cases where the timezone is not metadata; it defines how the datetime should be interpreted.
For example: your local mom and pop corner store's daily opening and closing times. Storing those in UTC is not correct because mom and pop don't open and close their store based on UTC time. They open and close it based on the local time zone.
You conflate different concepts here. The actual moment of opening and closing can be stored in UTC, because it's proper time. Scheduling algorithm is an algorithm, not time. You can use DSL similar to time to code this algorithm, but being DSL it can go only so far to implement it.
You don't need to store the timezone anywhere, you just need to know the current local timezone when the stored UTC time is used. And that's why storing in UTC is better, because it only takes one conversion to represent it in some arbitrary local time.
If you stored it as a local time (ie: with TZ), then if it's ever later translated to a different local time (different TZ), you're now dealing with all the quirks of 2 different timezones. It's great way to be off by some multiple of 15 minutes, or even a day or two!
Heck, even if it's the same exact location, storing in local time can still require conversion if that location uses daylight savings! You're never safe from needing to adapt to timezones, so storing datetimes in the most universal format is pretty much always the best thing to do.
It's a common source of off-by-one date formatting bugs in client-rendered web apps, particularly ones that pass around "YYYY-MM-DD" date strings (common for OpenAPI JSON APIs).
const dateStringFromApiResponse = "2026-01-12";
const date = new Date(dateStringFromApiResponse);
const formatter = new Intl.DateTimeFormat('en-US', { dateStyle: 'long' });
formatter.format(new Date("2026-01-12"));
// 'January 11, 2026'
Local time is unparsable, and this case is only human readable, because humans can handle ambiguity ad hoc. Parsing it as UTC is a reasonable default for a machine parser, at least the only workable one.
There's a lot wrong with Javascript's Date, but the fact that it's an object is is not really in the top 10.
Would it have been nice if the Date object had been immutable? Sure, but the fact that changing the mutable object does indeed change the object shouldn't be a shock
What happened to me is I passed a date to an external library, and then after that library did its work, that date was changed. Super annoying even if you know that it's a mutable object.
It's definitely a shock when something else changes the date object you've been holding on to. The problem with mutable values has never been when you (that is, the local context) change them. It's always that you can't trust that nothing else (some very non-local code) does.
That's a fair observation, and yet anecdotally I agree with the parent comment. In my career I fixed quite a few bugs about dates being passed around and unexpectedly modified, while I struggle to remember the same problem with objects in general (could be a case of selective memory).
If I had the guess, I'd say it's a combination of:
- the difference between the mental model and the implementation. Dates are objects but "feel" like values: dates are parsed from a single value, and when stored/printed they collapse back to a single value (as opposed to custom objects which are generally a bag of properties, and when printed/stored they still look like an object)
- most common date operations causes the original date object to be mutated, which implicitly causes developers to mutate the passed value even if that's not what they explicitly meant
So the default combination is a calling code that expects date to be treated as a value, and the called code accidentally mutating the data because it's convenient to do so. If anything then causes the original value to be saved back in the db, the data gets corrupted.
Most experienced developers will remember to make a copy of the date object both in the calling code and in the receiving code, but the default remains dangerously easy to get wrong.
What parent commenter meant was language level support of immutable objects. There is const, in for example JavaScript, but it supports only immutability of variables, not objects themselves. That later one is possible in C++ for example, where const can be used for both. Of course, it’s still possible to fake immutability with interfaces in some languages, or have a real one forced by implementation (like with Temporals), but it’s much nicer to have an indicator forcing that.
I would like to add, that both variable and object immutability should be the default, and mutability should have a keyword, not other way around how in C++ and Java.
This is a skill issue imo. Yes, if you change the referenced object you get a different value. Just because you are not paying attention to the change does not a problem of the language make.
There are million other things legitimately wrong wit JS, developers being bad at understanding referenced objects is not one of them.
I do find it annoying how the Temporal API, just like nearly all other datetime APIs, has 0 support for querying leap-second information in any shape or form. Suggested workarounds like temporal-tai all require plugging in a leap-second file and keeping it updated, which is especially painful for client-side JS, where you can't just download a leap-second file from someone else's site thanks to the SOP. Meanwhile, browsers update on a cadence more than sufficient to keep an up-to-date copy, but the datetime APIs refuse to expose leap-second info because they're too committed to "only UTC is in-scope for this project".
(The context is that I want to write some JS tools for astronomical calculations, but UTC conversions need leap-second info, so this trend makes it impossible to write something that Just Works™.)
>only UTC is in-scope for this project
>tools for astronomical calculations
Pity, since UTC is objectively the wrong time for astronomical calculations. Among other problems, UTC runs slightly slower or faster depending on how far the Earth is from the Sun. UTC does not run uniformly (outside of Earth-at-sealevel), instead the length of 1 second will slightly grow or shrink depending on the current configuration of the Solar system.
As you allude to, the correct time scale for this purpose would be TBD (aka Barycentric Dynamical Time), which applies relativistic corrections to act like the atomic clock is fixed at the barycentre of the Solar system. This is the only clock that actually runs "smoothly" for the purposes of astronomical calculations.
> Among other problems, UTC runs slightly slower or faster depending on how far the Earth is from the Sun. UTC does not run uniformly (apart from Earth-at-sealevel), instead the length of 1 second will slightly grow or shrink depending on the current configuration of the Solar system.
That is completely wrong. UTC seconds are exactly SI seconds, which are all the same uniform length (defined by quorum of atomic clocks).
At sea level on Earth, UTC seconds are all exactly the same, yes. That's the definition of UTC.
The trick is, when you're working with things on the scale of the Solar system and larger, it no longer makes sense to assume your frame is "at sea level on Earth." Your relativistic reference frame has shifted, so (thanks Einstein!) time literally changes underneath your feet.
The primary mechanism (but not the only one) is that a clock on Earth will tick slower when Earth is closer to the Sun, due to the effects of gravitational time dilation.[0]
So yes, a clock on Earth always runs at a uniform rate. But because the universe is fundamentally Einsteinian, that still means that eg if you're working with the orbit of Jupiter or a millisecond pulsar, you will see small introduced timing errors if you try to use UTC (or even UT1 which is UTC without leap seconds) instead of TBD.
But it's all relative, all reference frames are different and relative from each other and there is no one reference frame that is somehow special. TBD runs unevenly relative to UTC as much as UTC runs unevenly relative to TBD.
> where you can't just download a leap-second file from someone else's site thanks to the SOP
WDYM by this? Why does the SOP prevent a website from hosting a leap seconds file? All they need to do is set Access-Control-Allow-Origin to allow websites to access it. Or provide it as a JS file—in which case no headers are necessary at all. All the SOP prevents is you hotlinking someone else's leap-seconds file and using their bandwidth without their opt-in.
> Meanwhile, browsers update on a cadence more than sufficient to keep an up-to-date copy
Is this true? I don't know any browser right now that ships with a copy of a leapseconds data file. Adding such a data file and keeping it up to date would probably be a pretty non-trivial task for new browser developers—just for something the browser will never end up using itself. It's not like the ICU/CLDR files where browsers are going to need them anyway for rendering their own user-interface components.
> All they need to do is set Access-Control-Allow-Origin to allow websites to access it. All the SOP prevents is you hotlinking someone else's leap-seconds file and using their bandwidth without their opt-in.
They can, but the major providers (read: the ones I would trust to update it) don't. The IERS doesn't [0], the USNO doesn't [1], IANA doesn't [2], and NIST uses FTP [3]. Keep in mind that these files are constantly being downloaded by various clients for NTP and whatnot, it's not like these providers want to restrict public access, they just don't bother to set the header that would allow JS requests.
> Is this true? I don't know any browser right now that ships with a copy of a leapseconds data file.
From ECMA-262:
> It is required for time zone aware implementations (and recommended for all others) to use the time zone information of the IANA Time Zone Database https://www.iana.org/time-zones/.
Any browser that ships with a copy of tzdb, or knows where to find a copy from the OS, should have access to its leapseconds file. Unless you mean that all of them go solely through ICU and its data files? Which I suppose could be an obstacle unless ICU were to start exposing them.
> but the datetime APIs refuse to expose leap-second info because they're too committed to "only UTC is in-scope for this project".
This doesn't make sense on at least two different levels.
First, pedantically, the definition of UTC as a time scale is that it includes leap seconds. So if you're committed to UTC, then you're supporting leap seconds.
Second, and to more broadly address your point, you should say, "they're too committed to 'only the POSIX time scale is in-scope for this project.'" That more accurately captures the status quo and also intimates the problem: aside from specialty applications, basically everything is built on POSIX time, which specifically ignores the existence of leap seconds.
Sure, but my gripe isn't even that we ought to change the "POSIX time by default" status quo (the ship has long sailed that everyone counts durations by 'calendar seconds'), it's that the underlying libraries don't even provide enough information for "specialty applications" to reliably correct for it, short of perpetually updating it themselves.
Why should they though? To me it seems like a niche of a niche. I think there is plenty of room for scientific datetime libraries to service this need in a way that is likely better than what a general purpose datetime library would provide. (And indeed, there are many of them.)
I say this as someone who had leap second support working in a pre-release version of Jiff[1] (including reading from leapsecond tzdb data) but ripped it out for reasons.[2]
It’s not “niche” if you do things synchronized to GPS timestamps. (i.e. a significant portion of telecom, a bunch of electrical grid stuff, etc).
Anything using GPS as lock references to synchronize stuff that needs to be aligned to the millisecond absolutely cannot tolerate stuff like “the leap second smear”.
The vast majority of people using date time libraries don’t need to represent dates before 1980 or dates after 2036. It doesn’t mean you just bury your head in the sand and ignore a pretty critical part of how time representation works.
That's part of it: If I were writing a standalone program that could extract info from tzdb or whatever, I'd happily jump through those hoops, and not bother anyone else's libraries. I don't really care about the ergonomics. But for JS scripts in particular, there is no information available that is not provided by either the browser APIs or someone's server. And such servers are not in great supply.
> I do find it annoying how the Temporal API, just like nearly all other datetime APIs, has 0 support for querying leap-second information in any shape or form.
That’s because human time keeping doesn’t use leap seconds.
> like nearly all other datetime APIs, has 0 support for querying leap-second information
That's probably because you only need leap second accuracy in niche use cases, like astronomy or GPS. In JavaScript specifically, that kind of accuracy isn't needed for 99% of client-side use cases. Most date-time libraries work with POSIX time which assumes 86,400 seconds each day.
We’ve been loving using it in our Deno servers since last year. It’s been frustrating that we haven’t been able to upgrade our web client date logic yet, since even though Firefox has supported Temporal for a while, Chrome have really dragged their feet
I don't know if that's totally fair to Chrome. The specification was undergoing a lot of changes at the time, and V8 decided to wait for the specification to stabilize; meanwhile, Anba kept working on the implementation for FireFox. Additionally, the version of Temporal that Deno exposed last year was the heavily out of date to the most recent specification and had a large portion of the specification that was not even implemented.
It is a lot more complex than moment, but only because there's a lot of inherent complexity to dates and times that moment just doesn't deal with. So you need to be explicit about whether you're dealing with dates, times, or datetime objects, whether or not the object you're working with has a timezone, etc. Where moment is generally designed to have a convenient API, Temporal is designed to have a correct API.
The main selling point for me is that it has proper data times for dates, times, etc.
Most date/time libraries that I've seen have a only single "date/time" or "timestamp" type, then they have to do things like representing "January 13 2026" as "January 13 2026 at midnight local time" or "January 13 2026 at midnight UTC."
Temporal has full-fledged data types representing the different concepts: an Instant is a point in time. A PlainDate is just a date. A PlainTime is just a time. ("We eat lunch at 11am each day.") A ZonedDateTime is an Instant in a known time zone. Etc.
Temporal draws a lot of inspiration from Java's Joda-Time (which also went on to inspire .NET's Noda Time, Java's official java.time API, and JavaScript's js-joda). This is helpful; it means that some concepts can transfer if you're working in other languages. And, more importantly, it means that it benefits from a lot of careful thought on how to ergonomically and effectively represent date/time complexities.
Late to the party, I really wish everyone would copy Rails + Ruby, but specifically it's Rails additions.
2 things it got right:
1. Like the article a great API - Time.current.in_time_zone('America/Los_Angeles') + 3.days - 4.months + 1.hour
2. Rails overloads Ruby's core library Time. You're in 1 object the whole time no swap/wondering.
In the py world, pendulum is close but just like the article, it's cumbersome as it's still a separate obj (i.e. Temporal vs Date) and so you need to "figure out" what you have to manipulate or need to cast it first.
Overloading the core libs is dangerous for a whole host of reasons but for the end developer it's a pleasure to use.
If we could just do `new Date().add({ days: 1})` it would be so easier.
Is this what it looks like? A specific concept like time units being defined as members of more general types like numbers? I.e. if I type `1.` to get auto-complete, am I going to see days, and all the rest, as options?? That API design pattern would be a nightmare!
In Ruby, I assume this is done by monkey patching, so yes, it would have all the issues you mention and fear.
In more modern languages like Kotlin, there is a notion of extension methods and properties, where you would be able to write a library that allows this syntax, but the .days property would only be accessible in files where you have explicitly imported it (so basically a synthetic sugar for a static function call).
Here is the neat part about Ruby, your autocomplete barely works and your IDE can only guess what you want, instead of relying on a good language service…
Why write in this dismissive tone? Regardless of your personal situation, it would be hard to believe you don't share the common knowledge that virtually everybody uses auto-complete to some degree, e.g. to remember, discover, or abbreviate type/instance methods, argument lists, etc. And why is "API" in quotes? It's very normal to refer to interfaces in languages/platforms as APIs (e.g. "JavaScript's Date API").
Regardless, whether or not a person uses autocomplete for this API is irrelevant - in this case it would be anybody using numbers for things outside this API, and maximally it would be the whole platform if this design pattern is not unique to this API. I.e. the simplicity of this one API has no bearing on the question.
And yet people have been using it for decades without much trouble. I can't remember ever seeing any complaints about API explosion from adding methods to numbers in ruby, and rails does a fair bit of it.
Possibly one reason is with a fluent API that's pretty understandable, you don't need to rely on autocomplete. (And indeed, all this happened before that was a big thing in tooling.)
> When an immutable value is assigned to a variable, the JavaScript engine creates a copy of that value and stores the copy in memory
Not exactly. The language doesn't specify whether the value is copied or not and, precisely because values are immutable, there's no way for a user to tell if it was or wasn't.
For example, strings are also immutable value types, but you can be certain that no JS engine is fully copying the entire string every time you assign one to a variable or pass it to a parameter.
// A numeric string between 32 and 49 is assumed to be in the 2000s:
console.log( new Date( "49" ) );
// Result: Date Fri Jan 01 2049 00:00:00 GMT-0500 (Eastern Standard Time)
// A numeric string between 33 and 99 is assumed to be in the 1900s:
console.log( new Date( "99" ) );
// Result: Date Fri Jan 01 1999 00:00:00 GMT-0500 (Eastern Standard Time)
Good article, but “Java deprecated their Date way back in 1997” is not exactly true. They deprecated a lot of methods and constructors in JDK1.1 when Calendar was introduced, but the class itself was never deprecated and it was the preferred way to represent a point in time until the “modern” approach was provided in java.time in JDK8 (c2014)
(I can't find the 1.1 docs, but they were the same)
It's one of my favourite examples of how languages pretty much always get date and time hopelessly wrong initially. Java now has one of the best temporal APIs.
Yeah, it effectively became a typed wrapper of a long epoch millis value. Generally treated as immutable by convention in my experience, although of course it technically wasn't as the setters were never removed.
It was hopelessly wrong initially, and got even worse when they added the horrible sql Date/Timestamp/etc classes.
With java.time though, it is the gold standard as far as I've seen.
The timezone handling alone makes Temporal worth the wait. I've lost count of how many bugs I've shipped because Date silently converts everything to local time when you least expect it.
The ZonedDateTime type is the real win here - finally a way to say "this is 3pm in New York" and have it stay 3pm in New York when you serialize and deserialize it. With Date you'd have to store the timezone separately and reconstruct it yourself, which everyone gets wrong eventually.
Only downside I can see is the learning curve. Date was bad but it was consistently bad in ways we all memorized. Temporal is better but also much larger - lots of types to choose between.
Built a scheduler with pretty much all my moment/moment-tz questions answered through ChatGPT. One of the things it excels at, crawling long lived API documentation, answers, etc.
The Date API is fine and relatively normalized. Once you understand it it's very easy to work with. It's biggest problem is that it simply does not support timezones, which is the main reason to use Temporal.
Yea, that's understood to be the opinion, blandly repeating it adds little to the discussion.
It's simple. In it's simplicity it left many features on the floor. I just can't connect with the idea that someone would need to constantly be on MDN in order to work with it. It's not so horrible that it defies logic.
It’s not simple, though. Simple would be something like an object wrapping YYYY-MM-DD, like a COBOL programmer in the 1950s would’ve used. Instead, people have made thousands of variations of bugs around the complexity even basic usage forces you to internalize like the month number being zero-based while the year is 1900-based but the day of the month is 1-based following standard usage.
Checking its API, I'm surprised that Temporal.Duration has a constructor with many parameters for years, months, days, ... all the way to nanoseconds, while Temporal.Instant has no way at all to create it given a current year/month/day, only from unix timestamp equivalents (or strings)
That seems to be functionality you'd want to have? Or is the intention you convert your numbers to string first and then back to a Temporal.Instant?
It's perfectly reasonable to default seconds, minutes, hours, etc. to zero in the Duration constructor. But for Instant, it doesn't make sense to default those to zero unless you specify time zone offset.
And indeed, the static method Instant.from does accept an RFC 9557 string, which requires a 2-digit hour and a time zone offset, but can omit minutes and seconds:
I guess they don’t want people getting confused between local and UTC values for the fields in the constructor (especially as if it took local values it would need to handle DST transitions)
it's because a year/month/date isn't enough to uniquely identify an instant in time. you can create a Temporal.PlainDate(Time) with those values though, then convert to the other types depending on what you need and it needs (e.g. time zone)
Late by a decade or more (JSR310 was released in 2014), but still a good development. I've tried convincing colleagues to use js-joda in the past but they thought they were keeping it simple by sticking to moment.js. They weren't.
If you ever find yourself in the situation where you have to decide whether 99 parses to year 1999 and what to do with 100 and where to draw the line between parsing the 90s and current years, you should probably pause, reath in andaout and take a long walk instead. And after you return you should scrap the whole thing and go back to the drawing board.
Implementing such a feature has not only no value, it has negative value. When you program libraries or interfaces you ahould think aboit how people will use it 95% of the time and mane that usecase as simple, predictable and free of potential footguns as possible. This is the opposite of that. It feels like something 15 year old me would have programmed after reading the first book on PHP and not something anybosy with any experience could have thought to be a good thing.
The current global availability of native Temporal is 1.81%. For context, IE11(!) has a higher global usage than Temporal has native support. For my organization, this likely means we're years from being able to use Temporal in production, because getting the polyfills approved is such a hassle.
Keep in mind that even as of December last year, Chrome didn't ship with it yet (i.e. support is less than one month old). Safari still does not.
Please never do version checks. Test for the existence of the exact features/methods you need instead - this is trivial in JS: if(Temporal)
Checking against version numbers helps cement existing browser monopolies, makes it difficult for new browsers to view websites (even if the browser correctly implements every feature), and encourages everyone to spoof version numbers / browser names which leads to them becoming a less and less useful signal. See any browser’s User-Agent string for an example of this
The article is super weird. It never mentions Date.now(). It dances around the subject and exhaustively mentions the equivalent convention for Temporal.
If you want Date to act like Temporal then only use Date.now() as your starting point. It generates the number of milliseconds since 1 Jan 1970. That means the starting output is a number type in integer form. It does not represent a static value, but rather the distance between now and some universal point in the past, a relationship. Yes, Temporal is a more friendly API, but the primary design goal is to represent time as a relational factor.
Then you can format the Date.now() number it into whatever other format you want.
the article has examples of unexpected behavior with timestamps too, so... How do you covert to your desired format without going through Date? Please don't say date-fns
I do not believe you understood the comment you replied to. A date number, whether from Date.now() or the Temporal equivalent, is not a time stamp value. It’s just a number.
Then I do not believe you understood the article you've commented on, because it essentially about the thing you handwaved away via
> Then you can format the Date.now() number it into whatever other format you want.
The while thing is about how to transition from a date, wherever it's string or timestamp to a date object.
The idea is that you'll either format it to a string, do equality checks, calculate durations etc after after. Your idea of using timestamps doesn't address anything the article was about
I think the workflow your OP here and I use treats the timestamps (numbers) as the only value that matters. The "converted to" format is always just a temporary client side rendering, so the meticulous attention to conversions seems overkill.
But I do see that there is tons of stuff about Date that is best ignored
So, hold on--the author's soul-breaking complaint isn't all of the "quirks" and inconsistencies with the Date functions, but rather the fact that it's an object? Specifically, an object with mutable properties in a language when all objects have mutable properties?
I mean, the author's conclusion is correct. But I disagree with the rationale. It's like hating an evil dictatorship because they use the wrong font in their propaganda.
The old JS Date API is far from perfect and I'm happy it being replaced, but part of the problem is various string-based formats and people being sloppy using them. Not to mention general complexity in time/date concept with timezones, summer time, leap seconds, etc.
For string format, just stick with ISO 8601. If you need to parse less-standard formats, use a robust library of your choise. The standard library should not try to support parsing zillion obscure formats. Outputting localized / human-readable format should be a responsibility of localization API anyway.
I also think that many libraries/APIs involving formatting things have some US centric design limitations, i.e. tendency to treat US formats as native and international support is often a bit after-thought. Especially with older stuff like the JS Date API.
The problem with the date format is that the US one absolutely totally insane. Whenever you use something ordered you have to choose ordering. For date US choose the absurd kind. Y-d-m should never have been used. Remove that and around 90% of the string based format problems disappear.
I dislike the "Plain" prefix to the temporal objects, e.g. PlainDateTime. The prefix says nothing about the behavior of the class. Plain as opposed to what; ZonedDateTime. I would prefer "Local" over "Plain"; LocalDateTime.
Local would imply the date is in the current machine timezone, while PlainDateTime is zoneless. It may be in the server timezone, or anything else. The main difference is that it does not make sense to convert it to Instant or ZonedDateTime without specifying the timezone or offset
Is Temporal even in though? Last I checked (last year or so), I had to use some bleeding edge version of Firefox to get it, and absolutely nothing else had it. I do agree though it's lovely, and I'd love to see native support for it.
Nope. Only Firefox and Chrome have it, in their latest versions. No Safari or Edge support yet. So this article is a bit premature (unless you use the polyfill.)
If you do `new Date('2025-11-31T00:00:00Z')` you get `2025-12-01T00:00:00.000Z`, which is weird and potential cause for bugs when parsing date string from input. Right because of these inconsistency I created a small package backed as Regex to validate Date string input before throwing it on `new Date(stringDate)`.
> My complaint is about more than parsing or syntax or “developer ergonomics” ... My problem with Date is that using it means deviating from the fundamental nature of time itself.
I don't really have a problem with the substance of this blog post. I have a problem with this exaggerated writing style. It means deviating from the fundamental purpose of writing itself!
I had to scroll all the way to the end to find the actual point, and it was underwhelming.
> Unlike Date, the methods we use to interact with a Temporal object result in new Temporal objects, rather than requiring us to use them in the context of a new instance
Bro, just be honest. This entire blog post was totally about developer ergonomics and that's okay. We all hate the way Date works in javascript.
It's not an exaggeration - you're used to dramatic phrases that use similar wording ("fundamental nature of time itself"), but in this case it's a regular old literally-true statement. Date is used to represent two things: Timestamps and human times (date, time, tz). But it only actually represents the former. Using it to represent the latter is a hack we simply put up with.
Pedantically, Temporal also deviates from the fundamental nature of time itself. Temporal.Instant? In which accelerating frame of reference? It supports equality, which is a nonsense concept.
I hope that we can move to using DurationLike objects in place of ms (or sometimes seconds?) as plain numbers like in `setInterval`. `sleep({seconds 5})` is so much better than `sleep(5000)`.
related: i had to jump through the Date hoops recently (again) when rewriting uPlot's DST and timezone handling, but i'm pretty happy with the result that avoided much complexity, performance loss, and adding a dependency: https://github.com/leeoniya/uPlot/pull/1072
At one point we realy need to make web2. With a blank slate and maybe a much more sane implementation of Websites. Maybe just allow CSS Html and all functions need to run over Wasm. Wasm would need to get some more features for that but at least, you would avoid allot of bullshit.
What’s really sad is that this was obvious in 1995. If they’d taken a slightly longer time back then, millions of developers would have avoided tripping over those unnecessary quirks.
Some, yes, but if you looked at what was common outside of C programs this was antiquated even then. A SQL developer didn’t have to apply offsets and had better primitives for date math.
The JSC people have been relatively involved as the spec was developed, and they have it behind a flag in their TP builds . It’ll probably ship in Safari once it ships in Chrome so that their users don’t complain about broken websites
Temporal objects do not store formatting information. Unless you mean e.g. dropping the time, using a different time zone, etc - but those aren't formatting changes, they logically change the semantics of the data. Just like `myInt += 1` is not changing the "formatting" of `myInt`.
Remember: Date and Temporal objects are logically different things. A Date represents an absolute point in time (a timestamp), while a Temporal object represents a human time (calendar date, clock time, time zone). The fact that Dates are used to represent human time for lack of a better structure is the entire problem statement - the hole that all these other APIs like Temporal try to fill in.
I just groked it before and thought "plain" in `Temporal.PlainDateTime` indicates some specific format instead of just being zoneless. It is actually always using ISO8601 fur build in toString conversions, so I don't really have anything to complain about.
This article lists several of the absurdities of the Date constructor, but only barely touches on the most unforgivable one. The example from the article is:
In this example, the day is "wrong" because the constructor input is being interpreted as midnight UTC on January 2nd, and at that instantaneous point in time, it is 7pm on January 1st in Eastern Standard Time (which is the author's local time zone).What's actually happening here is a comedy of errors. JavaScript is interpreting that particular string format ("YYYY-MM-DD") as an ISO 8601 date-only form. ISO 8601 specifies that if no time zone designator is provided, the time is assumed to be in local time. The ES5 spec authors intended to match ISO 8601 behavior, but somehow accidentally changed this to 'The value of an absent time zone offset is “Z”' (UTC).
Years later, they had realized their mistakes, and attempted to correct it in ES2015. And you can probably predict what happened. When browsers shipped the correct behavior, they got too many reports about websites which were relying on the previous incorrect behavior. So it got completely rolled back, sacrificed to the altar of "web compatibility."
For more info, see the "Broken Parser" section towards the bottom of this article:
https://maggiepint.com/2017/04/11/fixing-javascript-date-web...
>So it got completely rolled back, sacrificed to the altar of "web compatibility."
This is why I don't understand the lack of directives.
'use strict'; at the top of a file was ubiquitous for a long time and it worked. It didn't force rolling back incompatibilities, it let you opt into a stricter parsing of JavaScript.
It would have been nice for other wide changes like this to have like a 'strict datetime'; directive which would opt you into using this corrected behavior.
They couldn't and shouldn't do this sort of thing for all changes, but for really major changes to the platform this would be an improvement.
Or they could go all in on internal modules, like how you can import `node:fs` now. They could include corrected versions of globals like
`import Date from 'browser:date';`
has corrected behavior, for example
To be fair, the new opt-in "use strict" here is "switch to Temporal". It's a new, stricter namespace object. Old Date code gets the old Date code quirks, new code gets the nice new Temporal API.
Internal modules would be handy in theory to maybe keep from having to dig through a thesaurus every time browsers decide to add a new, stricter version of an older API. Internal modules have even been proposed to TC-39 as a recommended way to continue to expand the JS API. Last I checked on that proposal it was stuck behind several concerns including:
1. Feature detection: detecting if Temporal available is as easy as `if ('Temporal' in globalThis) {}`, but detecting if a module import is missing is a bit harder. Right now the standard is that loading a module fails with an Error if one of its imports fails. You can work around that by doing a dynamic import inside a try/catch, but that's a lot of extra boilerplate compared to `const thingINeed = 'someApi' in globalThis ? someApi() : someApiPolyfill()`. I've seen multiple proposals on that front from extensions to import maps and `with { }` options on the import itself.
2. Bikeshedding (and lots of it): defining a URI scheme like `browser:` or `standard:` takes a bunch of thought on how you expand it. If it is just `browser:some-api` you run the risk of eventually polluting all the easy names in the exact way people worry about the risk of over-polluting `globalThis` (and in the way that it can be weirdly hard to find an available one-word name on npm), you've just moved the naming problem from one place to the other. On the other side, if you go down the road of something like `es-standard:https://tc39.es/ecma262/2025/v1/final-draft/Temporal`, even (especially) assuming users would mostly importmap that to something shorter you've recreated XMLNS URIs in a funny new hat and people who use JS all certainly have plenty of opinions on XMLNS URIs, many are very vocal in their hatred of it, but also they came out of a strong backwards incompatibility fixing desire exactly like this. (As they say time is a flat circle.)
> To be fair, the new opt-in "use strict" here is "switch to Temporal".
This. Don't break old code, just provide new best practices.
Update linters (or ideally first class language rules, like in Rust's "edition"s), to gradually kill off old behavior. Without having to do a decade long Python 2 -> 3 migration.
Temporal is nice. It learned from the many failures and dead bodies that came before it. And it had lots of good implementations to look at: Joda Time, Chrono, etc.
PHP suffers from this too. By too strict BC PHP has become a really mess of a language. IIRC there still is the ad-hoc parameter order convention and lack of any namespacing for builtins. Everything is global.
With JS i kind of get it as you cant control the env. Bit PHP does not have this limitation, and they still cant fix the mess that is PHP.
> To be fair, the new opt-in "use strict" here is "switch to Temporal"
Yes, but adding an entire new API that solves way more date-related problems is obviously much more difficult than fixing a very clear and well-understood bug. Temporal is only just now starting to ship, over 15 years after this bug was introduced and over 10 years since they decided to never fix the bug.
nuget has a convention of system packages that are empty if the target platform implements functionality natively and provides independent implementations for platforms that don't support it, as a result you can unconditionally import that package on all platforms.
> It would have been nice for other wide changes like this to have like a 'strict datetime'; directive which would opt you into using this corrected behavior.
That would be ugly, because you'd want some parts of your program (eg libraries) to use the old behaviour, and other parts might want the new behaviour. How would you minify multiple modules together if they all expect different behaviour from the standard library?
In my opinion the right way to do this is to have multiple constructors (as Obj-C, swift, C and rust all do). Eg:
The big downside of this is that its tricky to keep track of which fields and functions people should really stop using in modern javascript. It'd be nice if there was a way to enable more shouty warnings during development for deprecated JS features.I find it very unfortunate that browsers (or rather, the spec) do not support some kind of versioning. If we could declare which version of HTML, JS and CSS to use, it would allow for breaking changes without breaking the entire web.
There are so many (in hindsight bad) design choices and implementation accidents that currently exist in perpetuity because of backwards compatibility; the web would really benefit if every now and then we could shed old baggage.
It would also force browsers to implement multiple slightly different engine modes, vastly complicating the browser code.
There are already a few cases, eg quirks mode vs standards mode and “use strict” mode, which was considered necessary for moving forward, but clearly it also complicates things for browsers. We dont want more modes than what is necessary.
This was the approach Perl took and much as I love(d) that language, it do get pretty out of hand after a while if you wanted to adopt any newer or stricter language features.
> it do get pretty out of hand after a while if you wanted to adopt any newer or stricter language features.
How does it get out of hand?
FWIW, I just do `use v5.32;` or similar to opt-in to everything from that version.
https://perldoc.perl.org/functions/use#use-VERSION
Of course, if you instead want to pick-and-choose features, then I can see the list growing large.
Maybe something like rust's editions, where you can opt into a set of breaking changes made at a certain time.
directives sort of kinda work if you squint the eyes, but only as a crutch and only if you can't/don't want to change the API.
> Or they could go all in on internal modules, like how you can import `node:fs` now. They could include corrected versions of globals like `import Date from 'browser:date';`
This is what happened here, only the API changed as well
I very much remember coding a function that split the string on their components and then rebuild them to ensure the date was created without time zone.
Sometimes a date is just a date. Your birthday is on a date, it doesn't shift by x hours because you moved to another state.
The old Outlook marked birthdays as all-day events, but stored the value with time-zone, meaning all birthdays of people whose birthday I stored in Belgium were now shifted as I moved to California...
I always found it weird when systems code dates as DateTime strings. There needs to be a different primitive for Date, which is inherently timezone-less, and DateTime, which does require a timezone.
After having a bunch of problems with dealing with Dates coded as DateTime, I've begun coding dates as a Date primitive, and wrote functions for calculation between dates ensuring that timezone never creeps its way into it. If there is ever a DateTime string in a Date column in the database, it's impossible to know what the date was supposed to be unless you know you normalized it at some point on the way up.
Then I found that a lot of DatePicker libraries, despite being in "DATE" picker mode, will still append a local timezone to its value. So I had to write a sanitizer for stripping out the TZ before sending up to the server.
That said, I am pretty excited about Temporal, it'll still make other things easier.
Temporal does have PlainDate, which is the Date primitive you're describing (by a different name, presumably to not collide with the old Date type).
https://developer.mozilla.org/en-US/docs/Web/JavaScript/Refe...
This is great! Thanks for sharing
This has been a huge source of frustration in C#.
The BCL-provided DateTime was really confusing, especially when you just needed a Date. They eventually got around to including a DateOnly, but before that happened I switched to a library called "Noda" (or Joda in Java) and after a bit of a learning curve, it made everything a lot easier to reason about.
It has LocalDates and LocalDateTimes, as well as Instants to store UTC times. It also offers ZonedDateTimes, but I don't use those as much. I work in healthcare. And so many regulations involve strictly dates. Like, "You have 5 days to do X", not "You have 120 hours to do X", and as such, storing the time with a lot of this data can add more complexity.
Wow! And yeah I'd imagine Healthcare requires a bit more strictness, you can't really afford an off by one day error, or sometimes even an off by one hour error.
You're absolutely right.
The calculations themselves are usually pretty easy with a few exceptions, it's just that there are TONS of them. And they all slightly depend on each other. And they change. Often.
There needs to be a difference between an Instant, an Instant at an Observed Location, and a 'Specification for constructing a date that might or might not have already passed or pass in the future'.
E.G. in a conversation "Lets open the shop at 9am every day that it isn't closed." Is a fairly simple recurrence, with some exceptions*. If timezones change the scheduled time remains evaluated again on each day.
Yeah that's a good point, and also takes into account the dreaded DST (what are this business's operating hours for example, which remains the same locally but would change in UTC)
I mean... That's kinda how it works? More than once I've halfway forgotten birthdays of friends who live in timezones to my east, and then sent them a message saying "Happy birthday! (It still is where I am, lol)".
I'm not necessarily defending the implementation, just pointing out another way in which time is irreducibly ambiguous and cursed.
A reminder associated with the birthday can and should be changed if I change time zones. But the birthday date didn’t change so it shouldn’t move to a different day.
> But the birthday date didn’t change so it shouldn’t move to a different day.
But it does. My brother moved to the US for a few years. So we’d send him birthday wishes on the day of his birthday (Australia time), and he’d get them the day before his birthday (his time). Now he’s moved back to Australia, the same thing happens in reverse-he gets birthday wishes from his American friends the day after his birthday.
My wife has lots of American friends on Facebook (none of whom she knows personally, all people she used to play Farmville with)-and she has them wishing her a happy birthday the day after her birthday too. Maybe she’s doing the same to them in reverse.
But using UTC doesn't solve that, unless the recipient of the birthday wishes is close to the prime meridian.
You reminded me of some riddle I had once read that was about trying to figure out how someone could be born one year later but still be older than someone born in previous year. The answer to the riddle also relies on timezones. For sure, birthdates involve time zones.
The riddle explanation was something like: A baby is born in New York City at 12:15 AM on January 1. Thirty minutes later, another baby is born in Los Angeles, where the local time is 9:45 PM on December 31. Although the New York baby is actually older by 30 minutes, the calendar dates make it appear as though the Los Angeles baby was born first.
The other biggest fun trick of timezone math to a riddle like that would be the International Date line where a baby born on one side of it can be born on the "day before" by calendar reckoning despite being born 30 minutes after the other side of the line.
Fraternal (not identical) twins, born aboard a ship traveling west to east across the Pacific. One of them officially born January 1st, 2016. The younger-by-30-minutes twin officially born December 31st, 2015. They'll have the hardest time persuading people that they're really twins once they're grown up.
This works even without timezones. If they're born even a second apart, it can so happen on different days (if they're born around midnight)
Yes, and if you ask any midwife, OB/GYN, or other person who routinely delivers babies, I'm sure you'll hear about plenty of born-on-different-days twins. One of my in-laws is a doctor who delivers babies; she has lots of stories, some of which she's restricted by HIPAA from sharing. But once a baby is born, the baby's birth date is public knowledge so she can share that info. So she often will tell her husband, "My patient is going into labor, I have to go to the hospital" without naming the patient. Then after the baby is born she'll say "Mrs. Smith's baby was born at 11 PM last night" because now it's a matter of public record who the mother was, whereas before it was protected by HIPAA. Next time I talk to her, I'll ask her if she's ever personally delivered any twins with different birthdays.
The timezones thing, of course, is just a way to have the younger twin be born "a year ahead" of the older twin by having their births be in two different timezones. Only practical way that would happen would be aboard a ship, because 1) babies born aboard an airplane would probably end up using the time zone of departure or of destination for their birth, and so twins would not be counted as being born in different time zones. And 2) any land-based transportation such as a car or a train would likely pull over (or in the case of a train, let the pregnant woman off at the nearest station) so that the woman giving birth doesn't have to do so in a moving vehicle. So a ship is the only moving vehicle where this kind of thing could likely happen, as there's no option of getting off in the middle of the ocean. It could happen while crossing time zones east-to-west, but crossing the International Date Line west-to-east makes more of an interesting thought experiment.
Yes, I've given this silly joke scenario way more thought than it really deserves. :-)
Isn't that precisely the same?
Doesn't even have to be the International Date line, any two timezones work.
Yes, it's the same, the IDL just makes it easier for it to work, as otherwise the babies have to be born on either side of midnight while crossing the time zone line. With the IDL the birth time could be almost any time of day except for crossing over midnight and the joke would work.
> sacrificed to the altar of "web compatibility."
What should they have done instead? Force everybody to detect browser versions and branch based on that, like in the olden days of IE5?
(Serious question, maybe I'm overlooking some smart trick.)
I agree with the "don't break the web" design principle, but I sometimes disagree with precisely where TC39 draws the line. There is obviously a cost to breaking old, unchanging websites. But there's also a cost to allowing old, unchanging websites to hold the entire web hostage. Balancing those costs is a subjective matter.
As far as I know, TC39 doesn't have any clear guidelines about how many websites or how many users must be affected in order to reject a proposed change to JavaScript behavior. Clearly there are breaking changes that are so insignificant that TC39 should ignore them (imagine a website with some JavaScript that simply iterates over every built-in API and crashes if any of them ever change).
Browsers should version their languages. They should say "if you use <html version="5.2"> or bigger, this is the behavior".
Somehow, the standard groups decided to remove the versioning that was there.
The decided not to have it there because they didn't like the idea of maintaining version 4.0 forever in their engines.
That's basically why they never did anything like "use strict" again.
IMO, that's a bad choice. Giving yourself the ability to have new behavior and features based on a version is pretty natural and how most programming languages evolve. Having perpetual backwards and fowards compatibility at all times is both hard to maintain and makes it really hard to fix old mistakes.
The only other reason they might have chosen this route is because it's pretty hard to integrate the notion of compatibility levels into minifiers.
Have an optional parameter to opt in to the old behaviour and keep the new correct behaviour the default (without the parameter) seems like a decent choice.
To preserve backwards compatibility and not require all those old sites to update, the legacy behavior would have to be the default, with opt-in for the new behavior.
That is the opposite approach. Also an option. One could also deprecate the call without parameter and force always a parameter with which behaviour. The deprecation could last enough time that those websites would have been rewritten multiple times ;)
The control interface burned into your hardware device will not have been rewritten. And it's not like you can have a flag day where everyone switches over, so the lifespan of those hardware devices isn't that relevant.
Backwards compatibility is a large part of the point of the Web.
You could version everything at whatever granularity you like, but over time that accumulates ("bug 3718938: JS gen24 code incorrectly handles Date math as if it were gen25-34", not to mention libraries that handle some versions but not others and then implicitly pass their expectations around via the objects they create so your dependency resolver has to look at the cross product of versions from all your depencies...)
There is no free lunch. A deprecation warning lasting a decade before erroring will break less that some css boxing models and strict mode in many browsers.
Nah, that's not a "sacrifice", but the only sane way. In the ideal case, clearly document the constructor with a warning that it's not ISO conformant and offer a ISO conformant alternative.
In my (unfortunate) experience, DateTime/Timezone handling is one of the things most prone to introduce sneaky, but far-reaching bugs as it is. Introducing such a behaviour change that (usually) won't fail-fast, will often seemingly continue working as before until it doesn't and is deceptively tricky to debug/pinpoint/fix ist just asking for a fast lane into chaos.
And even with JS going the extra mile on backwards compatibility, I don't think most other languages would introduce that kind of breaking change in that way either.
You might want to play with https://jsdate.wtf/
One can't fathom how weird JS Date can be.
Guessed 2 of the first 3 questions.
Got to question 4 and gave up:
There's literally no way of guessing this crap. It's all random.I think you got it backwards. The only way is guessing, it's all random.
It's the internal inconsistencies that get me. Like, OK, I understand that there might be some quirks, maybe due to some weird backwards compatibility or technical limitation, but there are multiple incompatible quirks _inside_ this single interface! It's terrible, and things like this are a huge part of the reason JS was long considered (and sometimes still is) a Not So Good language.
I had no idea we even had an `Invalid Date` object, that's legitimately insane. Some other fun ones:
are both valid dates lol.the new Date() constructor is an amalgamation of like 5 different specs, and unless the input matches one of them, which one kicks in is up to the implementer's choice
The choice here is really surprising. I was half-expecting NaN, that you omitted.
Is there any other instance of the standard JS library returning an error object instead of throwing one? I can't think of any.
I think NaN itself is a bit of an error object, especially in how it's passed through subsequent math functions, which is a different choice than throwing up.
But besides that I think you're right, Invalid Date is pretty weird and I somehow never ran into it.
One consequence is you can still call Date methods on the invalid date object and then you get NaN from the numeric results.
The fun trick is that Invalid Date is still a Date:
You were half-correct on expecting NaN, it's the low level storage of Invalid Date: Invalid Date is just a Date with the "Unix epoch timestamp" of NaN. It also follows NaN comparison logic: It's an interesting curio directly related to NaN.> Invalid Date is just a Date with the "Unix epoch timestamp" of NaN. It also follows NaN comparison logic: > > > invalid === new Date(NaN) > false
This is just because a JS Date is an object and have nothing to do with the inner representation.
Personally, I like that UTC is the default time zone. Processing of dates should happen in a standardized time zone. It’s only when you want to display it that the date should become local.
UTC is a fine default time zone, but that doesn't matter here.
A datetime with a timezone and a datetime without one are two different things, both of them useful. My birthday does not have a time zone. My deadline does.
The company deadline for getting some document returned? It might or might not, that's policy.
Poetically: we are born free of time zones. We die bound to one.
> My birthday does not have a time zone. My deadline does.
That seems subjective
It doesn't to me. It should be obvious that there are plenty of valid uses of dates and times which implicitly refer to either an exact instant in time, or the expression of a time in a certain reckoning.
A birthday doesn't have a time zone because the concept of a birthday is more about the date on the calendar on the wall, not any universally understood singular instant in time; and so what matters most when it comes to your birthday is where you are. Your birthday doesn't move to the day before or after just because you travel to the other side of the globe.
A deadline has a time zone because when your boss says he wants the project done by 4PM, he means 4PM wherever you both currently are -- and the specific instant in time he referred to doesn't change if you get on a train and move a time zone over before that 4PM occurs.
And it may in fact be time zone and not just UTC with an offset; because if your boss tells you he wants a certain report on his desk by 4PM every day; when your local time zone goes into daylight saving time, it doesn't suddenly mean the report needs to be filed by 5PM instead.
In the first of these cases, the date itself has no time zone and is valid in whatever context its being read from. In the second, the instant in time might be expressed in UTC time with or without a specific offset. In the third, each of the successive instants in time may shift around with respect to UTC even while it continues to be referred to with one constant expression.
None of these are subjective interpretations. They're a consequence of the fact that as humans we've overloaded our representation of date/time with multiple meanings.
idk about you but I can get a happy birthday on the hour of my actual birth from people in the know but I never literally prepare things for the exact hour of the deadline. It's more like a day sort of thing
It does not. I'm Australian and our timezones are ahead of the US (NSW time is about 15-17 hours ahead of US Eastern time). If I took a flight from Sydney to New York (22~ hours) on my birthday, the US custom's officer would wish me happy birthday when I landed the next day.
Therefore, birthdays are not bound by timezone at all.
it's a cool thing to wish happy birthday on the hour they were born (if you know it) but I can't say the same about preparing for deadline.
I don't get a happy birthday from customs though, cool!
This will result in incorrect behavior when, between converting to UTC and back to the original timezone, the timezone database has changed, which happens more often than you think.
If time is stored in UTC, the result is correct even if the timezone database is corrupted, because timezone is only metadata and doesn't affect time.
Depends what you're actually storing. There are plenty of cases where the timezone is not metadata; it defines how the datetime should be interpreted.
For example: your local mom and pop corner store's daily opening and closing times. Storing those in UTC is not correct because mom and pop don't open and close their store based on UTC time. They open and close it based on the local time zone.
You conflate different concepts here. The actual moment of opening and closing can be stored in UTC, because it's proper time. Scheduling algorithm is an algorithm, not time. You can use DSL similar to time to code this algorithm, but being DSL it can go only so far to implement it.
Storing them in UTC is valid here also, but their IANA time zone string should also be stored ‘somewhere’.
You don't need to store the timezone anywhere, you just need to know the current local timezone when the stored UTC time is used. And that's why storing in UTC is better, because it only takes one conversion to represent it in some arbitrary local time.
If you stored it as a local time (ie: with TZ), then if it's ever later translated to a different local time (different TZ), you're now dealing with all the quirks of 2 different timezones. It's great way to be off by some multiple of 15 minutes, or even a day or two!
Heck, even if it's the same exact location, storing in local time can still require conversion if that location uses daylight savings! You're never safe from needing to adapt to timezones, so storing datetimes in the most universal format is pretty much always the best thing to do.
If this is comedy, sign me up for tragedy.
This feels like something that must be the root of innumerable small and easily overlooked bugs out there.
It's a common source of off-by-one date formatting bugs in client-rendered web apps, particularly ones that pass around "YYYY-MM-DD" date strings (common for OpenAPI JSON APIs).
I'm having flashbacks to writing Power Automate expressions to reconcile Dates passed from Outlook metadata to Excel
Basically just extracting numbers from string index or regex and rearranging them to a string Excel would recognize
Local time is unparsable, and this case is only human readable, because humans can handle ambiguity ad hoc. Parsing it as UTC is a reasonable default for a machine parser, at least the only workable one.
Maggie is a killer dev. Momentjs probably saved humanity millions of hours of collective coding and debugging
There's a lot wrong with Javascript's Date, but the fact that it's an object is is not really in the top 10.
Would it have been nice if the Date object had been immutable? Sure, but the fact that changing the mutable object does indeed change the object shouldn't be a shock
What happened to me is I passed a date to an external library, and then after that library did its work, that date was changed. Super annoying even if you know that it's a mutable object.
It's definitely a shock when something else changes the date object you've been holding on to. The problem with mutable values has never been when you (that is, the local context) change them. It's always that you can't trust that nothing else (some very non-local code) does.
also: there is a certain popular library for web app development that is based on diffing state between renders based on object equality
That’s how every other object works, why would that be surprising?
That's a fair observation, and yet anecdotally I agree with the parent comment. In my career I fixed quite a few bugs about dates being passed around and unexpectedly modified, while I struggle to remember the same problem with objects in general (could be a case of selective memory).
If I had the guess, I'd say it's a combination of:
- the difference between the mental model and the implementation. Dates are objects but "feel" like values: dates are parsed from a single value, and when stored/printed they collapse back to a single value (as opposed to custom objects which are generally a bag of properties, and when printed/stored they still look like an object)
- most common date operations causes the original date object to be mutated, which implicitly causes developers to mutate the passed value even if that's not what they explicitly meant
So the default combination is a calling code that expects date to be treated as a value, and the called code accidentally mutating the data because it's convenient to do so. If anything then causes the original value to be saved back in the db, the data gets corrupted.
Most experienced developers will remember to make a copy of the date object both in the calling code and in the receiving code, but the default remains dangerously easy to get wrong.
The lack of const means having objects as arguments is pretty dangerous
What parent commenter meant was language level support of immutable objects. There is const, in for example JavaScript, but it supports only immutability of variables, not objects themselves. That later one is possible in C++ for example, where const can be used for both. Of course, it’s still possible to fake immutability with interfaces in some languages, or have a real one forced by implementation (like with Temporals), but it’s much nicer to have an indicator forcing that.
I would like to add, that both variable and object immutability should be the default, and mutability should have a keyword, not other way around how in C++ and Java.
This is a skill issue imo. Yes, if you change the referenced object you get a different value. Just because you are not paying attention to the change does not a problem of the language make.
There are million other things legitimately wrong wit JS, developers being bad at understanding referenced objects is not one of them.
I do find it annoying how the Temporal API, just like nearly all other datetime APIs, has 0 support for querying leap-second information in any shape or form. Suggested workarounds like temporal-tai all require plugging in a leap-second file and keeping it updated, which is especially painful for client-side JS, where you can't just download a leap-second file from someone else's site thanks to the SOP. Meanwhile, browsers update on a cadence more than sufficient to keep an up-to-date copy, but the datetime APIs refuse to expose leap-second info because they're too committed to "only UTC is in-scope for this project".
(The context is that I want to write some JS tools for astronomical calculations, but UTC conversions need leap-second info, so this trend makes it impossible to write something that Just Works™.)
As you allude to, the correct time scale for this purpose would be TBD (aka Barycentric Dynamical Time), which applies relativistic corrections to act like the atomic clock is fixed at the barycentre of the Solar system. This is the only clock that actually runs "smoothly" for the purposes of astronomical calculations.
https://stjarnhimlen.se/comp/time.html
https://www2.mps.mpg.de/homes/fraenz/systems/systems2art/nod...
> Among other problems, UTC runs slightly slower or faster depending on how far the Earth is from the Sun. UTC does not run uniformly (apart from Earth-at-sealevel), instead the length of 1 second will slightly grow or shrink depending on the current configuration of the Solar system.
That is completely wrong. UTC seconds are exactly SI seconds, which are all the same uniform length (defined by quorum of atomic clocks).
At sea level on Earth, UTC seconds are all exactly the same, yes. That's the definition of UTC.
The trick is, when you're working with things on the scale of the Solar system and larger, it no longer makes sense to assume your frame is "at sea level on Earth." Your relativistic reference frame has shifted, so (thanks Einstein!) time literally changes underneath your feet.
The primary mechanism (but not the only one) is that a clock on Earth will tick slower when Earth is closer to the Sun, due to the effects of gravitational time dilation.[0]
So yes, a clock on Earth always runs at a uniform rate. But because the universe is fundamentally Einsteinian, that still means that eg if you're working with the orbit of Jupiter or a millisecond pulsar, you will see small introduced timing errors if you try to use UTC (or even UT1 which is UTC without leap seconds) instead of TBD.
[0] https://en.wikipedia.org/wiki/Gravitational_time_dilation
But it's all relative, all reference frames are different and relative from each other and there is no one reference frame that is somehow special. TBD runs unevenly relative to UTC as much as UTC runs unevenly relative to TBD.
Absolutely, I completely agree. Einstein has taught you well. ;) The only thing that matters is choosing the right reference frame for the job.
When high accuracy is required, UTC is not the right reference frame for the job of astronomical calculations. That's all.
> where you can't just download a leap-second file from someone else's site thanks to the SOP
WDYM by this? Why does the SOP prevent a website from hosting a leap seconds file? All they need to do is set Access-Control-Allow-Origin to allow websites to access it. Or provide it as a JS file—in which case no headers are necessary at all. All the SOP prevents is you hotlinking someone else's leap-seconds file and using their bandwidth without their opt-in.
> Meanwhile, browsers update on a cadence more than sufficient to keep an up-to-date copy
Is this true? I don't know any browser right now that ships with a copy of a leapseconds data file. Adding such a data file and keeping it up to date would probably be a pretty non-trivial task for new browser developers—just for something the browser will never end up using itself. It's not like the ICU/CLDR files where browsers are going to need them anyway for rendering their own user-interface components.
> All they need to do is set Access-Control-Allow-Origin to allow websites to access it. All the SOP prevents is you hotlinking someone else's leap-seconds file and using their bandwidth without their opt-in.
They can, but the major providers (read: the ones I would trust to update it) don't. The IERS doesn't [0], the USNO doesn't [1], IANA doesn't [2], and NIST uses FTP [3]. Keep in mind that these files are constantly being downloaded by various clients for NTP and whatnot, it's not like these providers want to restrict public access, they just don't bother to set the header that would allow JS requests.
> Is this true? I don't know any browser right now that ships with a copy of a leapseconds data file.
From ECMA-262:
> It is required for time zone aware implementations (and recommended for all others) to use the time zone information of the IANA Time Zone Database https://www.iana.org/time-zones/.
Any browser that ships with a copy of tzdb, or knows where to find a copy from the OS, should have access to its leapseconds file. Unless you mean that all of them go solely through ICU and its data files? Which I suppose could be an obstacle unless ICU were to start exposing them.
[0] https://hpiers.obspm.fr/iers/bul/bulc/ntp/leap-seconds.list
[1] https://maia.usno.navy.mil/ser7/tai-utc.dat
[2] https://data.iana.org/time-zones/tzdb/leap-seconds.list
[3] ftp://ftp.boulder.nist.gov/pub/time/leap-seconds.list
Could you put some kind of CORS proxy in front of those URLs? (I know it sucks that you have to do this at all, but c'est la vie.)
You could even write a Cloudflare Worker or a Val on val.town to do that, and add a bit of caching on top so you don't hit your providers too often.
> but the datetime APIs refuse to expose leap-second info because they're too committed to "only UTC is in-scope for this project".
This doesn't make sense on at least two different levels.
First, pedantically, the definition of UTC as a time scale is that it includes leap seconds. So if you're committed to UTC, then you're supporting leap seconds.
Second, and to more broadly address your point, you should say, "they're too committed to 'only the POSIX time scale is in-scope for this project.'" That more accurately captures the status quo and also intimates the problem: aside from specialty applications, basically everything is built on POSIX time, which specifically ignores the existence of leap seconds.
Sure, but my gripe isn't even that we ought to change the "POSIX time by default" status quo (the ship has long sailed that everyone counts durations by 'calendar seconds'), it's that the underlying libraries don't even provide enough information for "specialty applications" to reliably correct for it, short of perpetually updating it themselves.
Why should they though? To me it seems like a niche of a niche. I think there is plenty of room for scientific datetime libraries to service this need in a way that is likely better than what a general purpose datetime library would provide. (And indeed, there are many of them.)
I say this as someone who had leap second support working in a pre-release version of Jiff[1] (including reading from leapsecond tzdb data) but ripped it out for reasons.[2]
[1]: https://github.com/BurntSushi/jiff
[2]: https://github.com/BurntSushi/jiff/issues/7
It’s not “niche” if you do things synchronized to GPS timestamps. (i.e. a significant portion of telecom, a bunch of electrical grid stuff, etc).
Anything using GPS as lock references to synchronize stuff that needs to be aligned to the millisecond absolutely cannot tolerate stuff like “the leap second smear”.
"niche" doesn't mean "unimportant." It just means "uncommon." It's a relative term.
The vast vast majority of people using general purpose datetime libraries are not in need of GPS handling or high precision scientific calculations.
The vast majority of people using date time libraries don’t need to represent dates before 1980 or dates after 2036. It doesn’t mean you just bury your head in the sand and ignore a pretty critical part of how time representation works.
> (including reading from leapsecond tzdb data)
That's part of it: If I were writing a standalone program that could extract info from tzdb or whatever, I'd happily jump through those hoops, and not bother anyone else's libraries. I don't really care about the ergonomics. But for JS scripts in particular, there is no information available that is not provided by either the browser APIs or someone's server. And such servers are not in great supply.
> I do find it annoying how the Temporal API, just like nearly all other datetime APIs, has 0 support for querying leap-second information in any shape or form.
That’s because human time keeping doesn’t use leap seconds.
Can’t you just mirror the data? You could even embed it in the javascript file itself.
> like nearly all other datetime APIs, has 0 support for querying leap-second information
That's probably because you only need leap second accuracy in niche use cases, like astronomy or GPS. In JavaScript specifically, that kind of accuracy isn't needed for 99% of client-side use cases. Most date-time libraries work with POSIX time which assumes 86,400 seconds each day.
Just put it in an npm package.
I'm really surprised at how Temporal is only just rolling out in Chrome stable.
I would have hoped it'd be ready for wider use by now.
https://caniuse.com/temporal
We’ve been loving using it in our Deno servers since last year. It’s been frustrating that we haven’t been able to upgrade our web client date logic yet, since even though Firefox has supported Temporal for a while, Chrome have really dragged their feet
I don't know if that's totally fair to Chrome. The specification was undergoing a lot of changes at the time, and V8 decided to wait for the specification to stabilize; meanwhile, Anba kept working on the implementation for FireFox. Additionally, the version of Temporal that Deno exposed last year was the heavily out of date to the most recent specification and had a large portion of the specification that was not even implemented.
All the chromiums are scheduled for this week. Still TP for WebKit.
we've been using this Temporal polyfill and it's been awesome so far: https://github.com/js-temporal/temporal-polyfill
Note that this is 51kb, it's not exactly lightweight https://bundlephobia.com/package/@js-temporal/polyfill@0.5.1. Still good for forward compatibility or on the server, but for smaller apps it's significant.
Yep, I’ve been using this one which is lighter (20kB): https://github.com/fullcalendar/temporal-polyfill/
How is it compared to moment and especially luxon?
It is a lot more complex than moment, but only because there's a lot of inherent complexity to dates and times that moment just doesn't deal with. So you need to be explicit about whether you're dealing with dates, times, or datetime objects, whether or not the object you're working with has a timezone, etc. Where moment is generally designed to have a convenient API, Temporal is designed to have a correct API.
The main selling point for me is that it has proper data times for dates, times, etc.
Most date/time libraries that I've seen have a only single "date/time" or "timestamp" type, then they have to do things like representing "January 13 2026" as "January 13 2026 at midnight local time" or "January 13 2026 at midnight UTC."
Temporal has full-fledged data types representing the different concepts: an Instant is a point in time. A PlainDate is just a date. A PlainTime is just a time. ("We eat lunch at 11am each day.") A ZonedDateTime is an Instant in a known time zone. Etc.
Temporal draws a lot of inspiration from Java's Joda-Time (which also went on to inspire .NET's Noda Time, Java's official java.time API, and JavaScript's js-joda). This is helpful; it means that some concepts can transfer if you're working in other languages. And, more importantly, it means that it benefits from a lot of careful thought on how to ergonomically and effectively represent date/time complexities.
It’s a web standard api and very small weight. Highly recommend.
Didn't moment basically say in so many words, use Temporal?
Moment said many years ago "use Luxon" [1]. Luxon is not yet at the "use Temporal" stage [2].
[1] https://momentjs.com/docs/#/-project-status/
[2] https://github.com/moment/luxon/discussions/1742#discussionc...
https://momentjs.com/docs/#/-project-status/future/
Late to the party, I really wish everyone would copy Rails + Ruby, but specifically it's Rails additions.
2 things it got right:
1. Like the article a great API - Time.current.in_time_zone('America/Los_Angeles') + 3.days - 4.months + 1.hour
2. Rails overloads Ruby's core library Time. You're in 1 object the whole time no swap/wondering.
In the py world, pendulum is close but just like the article, it's cumbersome as it's still a separate obj (i.e. Temporal vs Date) and so you need to "figure out" what you have to manipulate or need to cast it first.
Overloading the core libs is dangerous for a whole host of reasons but for the end developer it's a pleasure to use.
If we could just do `new Date().add({ days: 1})` it would be so easier.
>3.days - 4.months + 1.hour
Is this what it looks like? A specific concept like time units being defined as members of more general types like numbers? I.e. if I type `1.` to get auto-complete, am I going to see days, and all the rest, as options?? That API design pattern would be a nightmare!
In Ruby, I assume this is done by monkey patching, so yes, it would have all the issues you mention and fear.
In more modern languages like Kotlin, there is a notion of extension methods and properties, where you would be able to write a library that allows this syntax, but the .days property would only be accessible in files where you have explicitly imported it (so basically a synthetic sugar for a static function call).
Here is the neat part about Ruby, your autocomplete barely works and your IDE can only guess what you want, instead of relying on a good language service…
I haven't used autocomplete in over 20 years. Why does this even need an autocomplete? The "API" is dead simple.
Why write in this dismissive tone? Regardless of your personal situation, it would be hard to believe you don't share the common knowledge that virtually everybody uses auto-complete to some degree, e.g. to remember, discover, or abbreviate type/instance methods, argument lists, etc. And why is "API" in quotes? It's very normal to refer to interfaces in languages/platforms as APIs (e.g. "JavaScript's Date API").
Regardless, whether or not a person uses autocomplete for this API is irrelevant - in this case it would be anybody using numbers for things outside this API, and maximally it would be the whole platform if this design pattern is not unique to this API. I.e. the simplicity of this one API has no bearing on the question.
> That API design pattern would be a nightmare!
And yet people have been using it for decades without much trouble. I can't remember ever seeing any complaints about API explosion from adding methods to numbers in ruby, and rails does a fair bit of it.
Possibly one reason is with a fluent API that's pretty understandable, you don't need to rely on autocomplete. (And indeed, all this happened before that was a big thing in tooling.)
> 3.days - 4.months + 1.hour
How is that a good thing?
> Rails overloads Ruby's core library Time. You're in 1 object the whole time regardless of what you do.
How is that a good thing?
When what you have is a pickaxe, everything looks like a rail...
Bit of a tangent, but I really wish JavaScript was versioned. Allowing MDN to clean up the syntax and APIs would be incredible.
> When an immutable value is assigned to a variable, the JavaScript engine creates a copy of that value and stores the copy in memory
Not exactly. The language doesn't specify whether the value is copied or not and, precisely because values are immutable, there's no way for a user to tell if it was or wasn't.
For example, strings are also immutable value types, but you can be certain that no JS engine is fully copying the entire string every time you assign one to a variable or pass it to a parameter.
typo:
the second interval should start at 50, not 33Good article, but “Java deprecated their Date way back in 1997” is not exactly true. They deprecated a lot of methods and constructors in JDK1.1 when Calendar was introduced, but the class itself was never deprecated and it was the preferred way to represent a point in time until the “modern” approach was provided in java.time in JDK8 (c2014)
Not exactly true, but they deprecated absolutely everything that made it a date. It expresses deep regret in the medium of annotations:
https://javaalmanac.io/jdk/1.2/api/java/util/Date.html
(I can't find the 1.1 docs, but they were the same)
It's one of my favourite examples of how languages pretty much always get date and time hopelessly wrong initially. Java now has one of the best temporal APIs.
Yeah, it effectively became a typed wrapper of a long epoch millis value. Generally treated as immutable by convention in my experience, although of course it technically wasn't as the setters were never removed.
It was hopelessly wrong initially, and got even worse when they added the horrible sql Date/Timestamp/etc classes.
With java.time though, it is the gold standard as far as I've seen.
Java's time and duration representations, heavily based on Joda, should be the standard that every language works towards.
It's just about perfect in every way. It makes it easy to do the right thing and it's very pleasant to read.
https://javaalmanac.io/jdk/1.1/api/java.util.Date.html
Curses, I was close. Also: yikes to the design.
The timezone handling alone makes Temporal worth the wait. I've lost count of how many bugs I've shipped because Date silently converts everything to local time when you least expect it.
The ZonedDateTime type is the real win here - finally a way to say "this is 3pm in New York" and have it stay 3pm in New York when you serialize and deserialize it. With Date you'd have to store the timezone separately and reconstruct it yourself, which everyone gets wrong eventually.
Only downside I can see is the learning curve. Date was bad but it was consistently bad in ways we all memorized. Temporal is better but also much larger - lots of types to choose between.
>I've lost count of how many bugs I've shipped because Date silently converts everything to local time when you least expect it.
You mean methods like getHours/getUTCHours?
I suspect that the arrival of ChatGPT caused traffic to the MDN Date API documentation to go down by at least half.
To be fair it's exactly the type of stuff I'd be glad if I never have to think about again.
Go down by half of user traffic, go up 10x of crawler traffic.
Built a scheduler with pretty much all my moment/moment-tz questions answered through ChatGPT. One of the things it excels at, crawling long lived API documentation, answers, etc.
The Date API is fine and relatively normalized. Once you understand it it's very easy to work with. It's biggest problem is that it simply does not support timezones, which is the main reason to use Temporal.
The Date API is horrible
Yea, that's understood to be the opinion, blandly repeating it adds little to the discussion.
It's simple. In it's simplicity it left many features on the floor. I just can't connect with the idea that someone would need to constantly be on MDN in order to work with it. It's not so horrible that it defies logic.
It’s not simple, though. Simple would be something like an object wrapping YYYY-MM-DD, like a COBOL programmer in the 1950s would’ve used. Instead, people have made thousands of variations of bugs around the complexity even basic usage forces you to internalize like the month number being zero-based while the year is 1900-based but the day of the month is 1-based following standard usage.
It's not simple, it has hidden pitfalls and footguns. Using it is the fastest way to blood on the floor (to quote a senior developer I worked with).
yep. i had a hell of a time building a scheduling system for a business that worked across timezones with the date API
Checking its API, I'm surprised that Temporal.Duration has a constructor with many parameters for years, months, days, ... all the way to nanoseconds, while Temporal.Instant has no way at all to create it given a current year/month/day, only from unix timestamp equivalents (or strings)
That seems to be functionality you'd want to have? Or is the intention you convert your numbers to string first and then back to a Temporal.Instant?
It's perfectly reasonable to default seconds, minutes, hours, etc. to zero in the Duration constructor. But for Instant, it doesn't make sense to default those to zero unless you specify time zone offset.
And indeed, the static method Instant.from does accept an RFC 9557 string, which requires a 2-digit hour and a time zone offset, but can omit minutes and seconds:
Temporal.Instant.from("2026-01-12T00+08:00")
I guess they don’t want people getting confused between local and UTC values for the fields in the constructor (especially as if it took local values it would need to handle DST transitions)
it's because a year/month/date isn't enough to uniquely identify an instant in time. you can create a Temporal.PlainDate(Time) with those values though, then convert to the other types depending on what you need and it needs (e.g. time zone)
Late by a decade or more (JSR310 was released in 2014), but still a good development. I've tried convincing colleagues to use js-joda in the past but they thought they were keeping it simple by sticking to moment.js. They weren't.
I remember running into the moment.js issue where my package size doubled by adding a date library: https://github.com/moment/moment/issues/3376
If you ever find yourself in the situation where you have to decide whether 99 parses to year 1999 and what to do with 100 and where to draw the line between parsing the 90s and current years, you should probably pause, reath in andaout and take a long walk instead. And after you return you should scrap the whole thing and go back to the drawing board.
Implementing such a feature has not only no value, it has negative value. When you program libraries or interfaces you ahould think aboit how people will use it 95% of the time and mane that usecase as simple, predictable and free of potential footguns as possible. This is the opposite of that. It feels like something 15 year old me would have programmed after reading the first book on PHP and not something anybosy with any experience could have thought to be a good thing.
Except Temporal is not, in fact, out.
https://caniuse.com/temporal
The current global availability of native Temporal is 1.81%. For context, IE11(!) has a higher global usage than Temporal has native support. For my organization, this likely means we're years from being able to use Temporal in production, because getting the polyfills approved is such a hassle.
Keep in mind that even as of December last year, Chrome didn't ship with it yet (i.e. support is less than one month old). Safari still does not.
Chrome will ship in 144. It's about to release.
Once Chrome does, for many devs, you can simply enforce a version check and say "please use latest Chrome" and be done with it.
Please never do version checks. Test for the existence of the exact features/methods you need instead - this is trivial in JS: if(Temporal)
Checking against version numbers helps cement existing browser monopolies, makes it difficult for new browsers to view websites (even if the browser correctly implements every feature), and encourages everyone to spoof version numbers / browser names which leads to them becoming a less and less useful signal. See any browser’s User-Agent string for an example of this
Thankfully temporal-polyfill only depends on temporal-spec. It was pretty easy to get that through.
Cheers, great read.
Eh... both are in common usage, including by institutions like the BBC.
https://www.bbc.co.uk/future/article/20240308-daylight-savin... (you can find both "saving" and "savings" here, I guess they couldn't decide which to use?)
The article is super weird. It never mentions Date.now(). It dances around the subject and exhaustively mentions the equivalent convention for Temporal.
If you want Date to act like Temporal then only use Date.now() as your starting point. It generates the number of milliseconds since 1 Jan 1970. That means the starting output is a number type in integer form. It does not represent a static value, but rather the distance between now and some universal point in the past, a relationship. Yes, Temporal is a more friendly API, but the primary design goal is to represent time as a relational factor.
Then you can format the Date.now() number it into whatever other format you want.
the article has examples of unexpected behavior with timestamps too, so... How do you covert to your desired format without going through Date? Please don't say date-fns
I do not believe you understood the comment you replied to. A date number, whether from Date.now() or the Temporal equivalent, is not a time stamp value. It’s just a number.
Then I do not believe you understood the article you've commented on, because it essentially about the thing you handwaved away via
> Then you can format the Date.now() number it into whatever other format you want.
The while thing is about how to transition from a date, wherever it's string or timestamp to a date object. The idea is that you'll either format it to a string, do equality checks, calculate durations etc after after. Your idea of using timestamps doesn't address anything the article was about
I think the workflow your OP here and I use treats the timestamps (numbers) as the only value that matters. The "converted to" format is always just a temporary client side rendering, so the meticulous attention to conversions seems overkill. But I do see that there is tons of stuff about Date that is best ignored
So, hold on--the author's soul-breaking complaint isn't all of the "quirks" and inconsistencies with the Date functions, but rather the fact that it's an object? Specifically, an object with mutable properties in a language when all objects have mutable properties?
I mean, the author's conclusion is correct. But I disagree with the rationale. It's like hating an evil dictatorship because they use the wrong font in their propaganda.
The old JS Date API is far from perfect and I'm happy it being replaced, but part of the problem is various string-based formats and people being sloppy using them. Not to mention general complexity in time/date concept with timezones, summer time, leap seconds, etc.
For string format, just stick with ISO 8601. If you need to parse less-standard formats, use a robust library of your choise. The standard library should not try to support parsing zillion obscure formats. Outputting localized / human-readable format should be a responsibility of localization API anyway.
I also think that many libraries/APIs involving formatting things have some US centric design limitations, i.e. tendency to treat US formats as native and international support is often a bit after-thought. Especially with older stuff like the JS Date API.
The problem with the date format is that the US one absolutely totally insane. Whenever you use something ordered you have to choose ordering. For date US choose the absurd kind. Y-d-m should never have been used. Remove that and around 90% of the string based format problems disappear.
I live in Australia and we have d-m-Y, and even I would say just use ISO 8601. Just use the standard, please, pretty please with a cherry on top.
I dislike the "Plain" prefix to the temporal objects, e.g. PlainDateTime. The prefix says nothing about the behavior of the class. Plain as opposed to what; ZonedDateTime. I would prefer "Local" over "Plain"; LocalDateTime.
Local would imply the date is in the current machine timezone, while PlainDateTime is zoneless. It may be in the server timezone, or anything else. The main difference is that it does not make sense to convert it to Instant or ZonedDateTime without specifying the timezone or offset
you have to consider the existence of PlainDate which is a Date without both a Time and a TimeZone
Is Temporal even in though? Last I checked (last year or so), I had to use some bleeding edge version of Firefox to get it, and absolutely nothing else had it. I do agree though it's lovely, and I'd love to see native support for it.
You can always check in "Can I use _?": https://caniuse.com/temporal
It's fully supported as of version 139[0], so since ~June of last year.
[0] https://developer.mozilla.org/en-US/docs/Web/JavaScript/Refe...
Seems like the next version(s) of Chrome will have it: https://caniuse.com/temporal
Nope. Only Firefox and Chrome have it, in their latest versions. No Safari or Edge support yet. So this article is a bit premature (unless you use the polyfill.)
If you do `new Date('2025-11-31T00:00:00Z')` you get `2025-12-01T00:00:00.000Z`, which is weird and potential cause for bugs when parsing date string from input. Right because of these inconsistency I created a small package backed as Regex to validate Date string input before throwing it on `new Date(stringDate)`.
https://www.npmjs.com/package/iso-8601-regex
Somewhat related: Jiff (https://github.com/BurntSushi/jiff) is a Rust library inspired by the Temporal api.
The library used for V8's temporal implementation is in Rust too ;)
> My complaint is about more than parsing or syntax or “developer ergonomics” ... My problem with Date is that using it means deviating from the fundamental nature of time itself.
I don't really have a problem with the substance of this blog post. I have a problem with this exaggerated writing style. It means deviating from the fundamental purpose of writing itself!
I had to scroll all the way to the end to find the actual point, and it was underwhelming.
> Unlike Date, the methods we use to interact with a Temporal object result in new Temporal objects, rather than requiring us to use them in the context of a new instance
Bro, just be honest. This entire blog post was totally about developer ergonomics and that's okay. We all hate the way Date works in javascript.
It's not an exaggeration - you're used to dramatic phrases that use similar wording ("fundamental nature of time itself"), but in this case it's a regular old literally-true statement. Date is used to represent two things: Timestamps and human times (date, time, tz). But it only actually represents the former. Using it to represent the latter is a hack we simply put up with.
Pedantically, Temporal also deviates from the fundamental nature of time itself. Temporal.Instant? In which accelerating frame of reference? It supports equality, which is a nonsense concept.
I hope that we can move to using DurationLike objects in place of ms (or sometimes seconds?) as plain numbers like in `setInterval`. `sleep({seconds 5})` is so much better than `sleep(5000)`.
related: i had to jump through the Date hoops recently (again) when rewriting uPlot's DST and timezone handling, but i'm pretty happy with the result that avoided much complexity, performance loss, and adding a dependency: https://github.com/leeoniya/uPlot/pull/1072
fun demos: https://leeoniya.github.io/uPlot/demos/timezones-dst.html
Temporal feels less like a replacement for Date and more like an admission that Date was trying to represent too many different concepts at once.
Not yet, adoption is kind of poor atm.
This is not to detract from the quality of the article which is a top-notch introduction to this new API.
Temporal being immutable alone is worth whatever migration pain we're about to endure.
At one point we realy need to make web2. With a blank slate and maybe a much more sane implementation of Websites. Maybe just allow CSS Html and all functions need to run over Wasm. Wasm would need to get some more features for that but at least, you would avoid allot of bullshit.
I use Moment.
And a few years later, I'm still using Moment.
And now I'm still using Moment.
And now...
I actually had no idea javascript Date was this bad until now.
What’s really sad is that this was obvious in 1995. If they’d taken a slightly longer time back then, millions of developers would have avoided tripping over those unnecessary quirks.
to be fair, a lot of lessons about date/time API came way after that year
Some, yes, but if you looked at what was common outside of C programs this was antiquated even then. A SQL developer didn’t have to apply offsets and had better primitives for date math.
If only Safari had support for Temporal... maybe next year...
There's a polyfill, I wonder if that works in Safari:
https://github.com/js-temporal/temporal-polyfill
Anyone have links that indicate when Safari might have it?
The JSC people have been relatively involved as the spec was developed, and they have it behind a flag in their TP builds . It’ll probably ship in Safari once it ships in Chrome so that their users don’t complain about broken websites
For now, there's a polyfill: https://github.com/js-temporal/temporal-polyfill
I am not sure if I like mixing value and formatting in single object. On other hand anything will be an improvement compared to that terrible old API.
Temporal objects do not store formatting information. Unless you mean e.g. dropping the time, using a different time zone, etc - but those aren't formatting changes, they logically change the semantics of the data. Just like `myInt += 1` is not changing the "formatting" of `myInt`.
Remember: Date and Temporal objects are logically different things. A Date represents an absolute point in time (a timestamp), while a Temporal object represents a human time (calendar date, clock time, time zone). The fact that Dates are used to represent human time for lack of a better structure is the entire problem statement - the hole that all these other APIs like Temporal try to fill in.
I just groked it before and thought "plain" in `Temporal.PlainDateTime` indicates some specific format instead of just being zoneless. It is actually always using ISO8601 fur build in toString conversions, so I don't really have anything to complain about.
[dead]
[flagged]
What’s the superior alternative and in which domains?
bro these ads I can't
Yes, you should use shiny new library instead of the glitchy javascript standard..
Unless you want your website to run in browsers older than a year
Maybe in 10 years we can start using the shiny new thing?
Polyfills are a thing, no?
Sure, but you are adding an attack vector or two, and a couple of kilobytes.