This does not trace things back far enough. The root is where IANA has long since segregated out a set of timezone file names into a "backward" collection:
If one traces references, one finds this connected bug on Launchpad. Amusingly to anyone who has ever seen these sweeping timezone database changes over the years, Launchpad marked it as "This bug affects 1 person.".
All of the US/* timezone names, such as US/Pacific here, have been backwards compatibility measures in place for the whole of the 21st century and some of the late 20th. The Olson database in the 1980s (mod.sources v08i085, comp.sources.unix v14i030) used these names. But the naming scheme changed somewhen in the 1990s to a continent/city and ocean/city form and backwards compatibility with the old names has been preserved ever since by the "backward" file.
Hang on a second, "(...) in 2023. US/* was moved to tzdata-legacy (...)"
US/* was moved to 'backward' (the file for backward compatibility) in the tz database in 1993(!) and as such was essentially marked as deprecated long enough.
https://data.iana.org/time-zones/tzdb/backward
You're telling me you didn't notice ? It was on display in the bottom of a locked filing cabinet stuck in a disused lavatory with a sign on the door saying ‘Beware of the Leopard.
Not only did I not notice, I have never known that country prefixes were a thing, having to deal with tzdata since 1999. I wonder if that timezone was typed in manually? I doubt Postgres 15 file contained it to begin with.
In a large fraction of cases in the FOSS world, it comes across that the developers really do want to communicate this sort of thing, but there's no clarity on where or how they should do so. See for example various deprecations in Python packaging tools (and standards).
In this case, they did communicate it, and the aforegiven Vogon reference is a mischaracterization. The naming convention is in the current IANA doco and Eggert copy.
Paul Eggert explained the continent/ocean plus largest city naming convention on a WWW page almost a quarter of a century ago. The WWW page was so well publicized that you can find its URL baked into at least four of the O'Reilly animal-cover books from the early 2000s.
You know, the tzdata people are quite haughty. They claim to store all that change, accurately, and yet here we are.
An example of this falsehood? Well, in the 70s my father convinced most of my hometown, at least the portions between Main St and Wharf, that DST was absurd.
For almost an entire year, this was observed.
Do you think they kept record in tzdata? I tried to convince them, but no! I still have some dateplanners my father had printed up, and even a picture of the sign that was out on the road (to alert visitors!).
But no!
Do not trust the tzdata people. As you can see, they are not so accurate.
Once upon a time I stumbled upon a small book of maps showing the time zones in Indiana each year in the mid-20th century. It was fascinating to see how various towns and cities would jump back and forth between Eastern and Central year to year, with little regard for the surrounding rural areas.
I frequently kick myself for losing track of that book.
You are just not providing sufficient proof. Think Wikipedia style proof. The fact that DST is or isn't observed should be published in a real book. A random dude printing some date planners or putting up a sign to be photographed isn't enough proof.
It seems like they're being pragmatic, not haughty:
> "The tz database is not authoritative, and it surely has errors. [...] Errors in the tz database arise from many sources: Sometimes, different people in the same city maintain clocks that differ significantly. [...] Even if the time is specified by law, locations sometimes deliberately flout the law. [...] Any attempt to pass the tz database off as the definition of time should be unacceptable to anybody who cares about the facts."
Then you get the reverse. I just upgraded to macOS Sonoma (yes I'm always one major version behind with Apple stuff...), and I was annoyed as heck when I had to click through "Look what's new in Calendar!", "Look what's new in Reminders!", "Look what's new in StripClubs!"... I need to use my software right now, I will not read this. Then I will forget it ever popped up, and will not read it in the future either.
There's a vast difference between Apple showing you all it's useless new AI integrations and developers telling system administrators what they need to know.
> I have not heard of a city moving between continents however.
Continents are conventional are there are multiple competing conventions. The same place absolutely can and will move continents if you decide to swap out one of those conventions for another.
Or consider Cyprus-traditional geography assigns it to Asia-it may be an island, but that’s the nearest continental landmass-and so IANA tzdata lists it as Asia/Nicosia-but since it (or at least the internationally recognised part of it) is in the EU, a lot of people view it as part of Europe. And the standard name for its time zone is “Eastern European Time”
> The worst part about this is that it didn't get so much as a mention in the Debian 13 release notes. I read through that document before going for it and never encountered it. Indeed, even now, you won't find "tzdata" or "zone" in it.
I had another issue during my upgrade to Debian 13 that also wasn't mentioned in the release notes. I filed a bug report, but I was told that the issue was not important enough to put in the release notes, and I should have instead more closely read NEWS.Debian of the package. So I don't really know anymore what the "Issues to be aware of" chapter in the release notes is for, because apparently it's only a small selection of issues you need to be aware of.
Sharing undocumented gotcha: nginx default config (/etc/nginx/nginx.conf) now has `server_tokens off` set with a comment that it's "common practice" (agreed). This is not in upstream's git version of the file[1], I therefore presume it's a Debian maintainer change.
The operational upgrade failure is if you have any existing drop-in config that sets `server_tokens off` already a hard fail Nginx error about duplicated keywords will occur, causing the apt process to exit with failure code(s) during the standard upgrade process.
The version of pdns-recursor in trixie uses a different configuration
file syntax then older versions.
(There is a tool to convert the old configuration syntax to the new, but it requires a working installation. There is a command line option to enable support for the old format, which if nothing else helps to be able to run the conversion tool, but there's no information how to enable that command line option in the way Debian starts pdns-recursor.)
As an Australian, I love that `Australia/Eucla` is there, with it's silly UTC+8.75.
Some of the other Australian entries are bizarre though - 'North' and 'South'? There is no standard timezone across the whole north or south of the country. And Broken Hill is in NSW, so why are they different time zones?
I ran into a similar bug in Git Lab where the v18 upgrade didn’t replace now-unsupported time zones so things like scheduled pipelines simply stopped running and were labeled inactive in the UI without explanation.
The fix is simply to change them to the non-backwards linked names but it caused some confusing API errors since data which would no longer pass validation was already in the database and didn’t look wrong.
I work with a development team who manages data integration and migration for massive datasets, and they have sensibly set a standard that every date/time value they store in their databases will be UTC.
But they explicitly or implicitly have also decided not to store the timezone in the strings, so every single value is technically ambiguous. Absolutely drives me crazy.
Update: since there have been questions, these are strings, not native datetime values.
Even if you stored the timezone, the values are technically ambiguous. Even when you are using globally unique timezone identifiers like 'America/New_York' instead of 'EST' and its several meanings. A timestamp such as '2025-09-13 13:00 America/New_York' could end up referring to a different instant next week due to a correction in the timezone database. Unlikely for this sort of problem to happen and need correcting in major timezones, but it has happened for less popular and historical timezones. The way for them to be non-ambiguous, representing an unchangable instant in time, is to store the timestamp converted to a timezone that will never have retroactive changes and has no daylight savings time transitions, such as UTC. At which point, storing the timezone identifier is redundant (and violates the principle of not allowing illegal values to be represented if you follow that).
I feel strongly that a string time value with no specified timezone is dangerous.
You have to make certain it is always handled as UTC, and if you export it to another system without making the timezone explicit, or anyone who is unfamiliar with the convention views it, you’ve lost the guarantee.
When you’re shuffling around petabytes of data, adding another few megabytes to include the “Z” at the end of every timestamp is hardly a major expense.
In many systems it is reasonable not to include a zone. Usually that goes hand-in-hand with the desire to use a more compact representation, such as storing a numerical timestamp with a customary interpretation (usually UTC or TAI). If you must store a string, you may as well include a zone. Using ambiguously-zoned timestamps is an invitation for bugs. I feel your pain.
I work for a company with servers, employees, and customers in basically every time zone around the world. And yet every server, internal tool, and workflow uses Pacific time. UTC is used precisely nowhere. Setting aside the issues of DST, I imagine it's convenient for the employees and managers in HQ but absolute madness for everyone else.
My non-Walmart startup was expanding from California to other regions many years ago, and there was some debate about whether we’d use UTC or America/Los_Angeles globally. I was part of Team Over My Dead Body.
It also makes sense. Timezone is user specific - if you have users from all over the globe, they will need different settings, so this should be set in frontend.
This is why Debian users should use apt-listchanges to display the latest NEWS.Debian items on upgrade.
I wouldn't expect to see all the important news of the tens of thousands of packages I don't have installed in the release notes.
I was hit by this using unstable (before they made a NEWS item), but when upgrading my stable machines to trixie I got a proper warning/reminder of this specific thing.
You’re not wrong in general terms, but tzdata is fairly well essential, a default component, and a dependency for unpteen things in the ecosystem. Of course a line must be drawn somewhere, but this really is a “major” caveat.
That’s one of the normal Continent/City time zones, not one of the weird Country/ZoneName ones that have been deprecated for ages. “America” here refers to the continent or continents[0], not the USA.
0: Note that different cultures disagree on whether there are two continents called North America and South America, or one continent called America.
> At the time, I went "WTF?" and just commented it out to get it running again. I had bigger fish to fry... and just kind of forgot about it. Everything seemed fine.
I get this, but now you get to laugh and dust yourself off. If it had silently started doing the wrong thing, that would be a worthy complaint.
I'm still not totally sure why these names were deprecated in the first place...
I mean, the folks who run the tz db definitely know what they're doing, it just never 100% clicked with my thinking.
I always prefer `US/Eastern` over `America/New_York` -- it seems more "canonical" to me. New York is _currently_ the anchor city for ET, but will it always be? The place I live (Boston) is currently on ET, but in the future it might be on Atlantic Time. If there was an `America/Boston`, I would use that to be safe, but since there isn't, it just seems better to be to be specific that I mean "Eastern Time" and not "whatever the time is in NYC"... At least then if Boston switches to a different tz, I could intentionally switch to "Atlantic Time" -- doesn't that make more sense? Versus I guess what I'd have to do, which is switch to `America/Puerto_Rico`? (I had to actually search that one, too bad there's no `US/Atlantic`...)
> the folks who run the tz db definitely know what they're doing
It's one guy. He demonstrably does NOT know what he's doing.
> I always prefer `US/Eastern`
As you should. It's the actual name of the timezone as published by the entity that defines it. Outside of the goofy definitions in the tz file it's what everyone living _inside_ of that timezone would call it and see it referenced as.
To call this "backwards" is an absolute insult to civil time keeping and drives me insane.
> doesn't that make more sense?
I always thought it should be served through DNS. Then each country can just define it's own TZ record type and embed it at the root of their country code domain and could expand on it however they like.
eastern.timezone.us
Also, since domain names have punycode for internationalization, you could actually call timezones in countries like Mexico what they're actually named for end users.
If New-York shifts its timezone, then it will still be New-York and America. And the new tzdata will reflect that shift so your data will always be correct (as in reflecting the time it is in new york). For local time, you will switch to a new city and the date will still be correct.
Exactly, but that is my point: usually my data is not intended to reflect the time it is in New York. Being tied to a (semi-)arbitrary city changes the actual meaning of the zone slightly. If every city was represented and I could choose "Boston" then that would make sense (for data is intended to reflect the time it is in Boston) but of course that's not entirely practical.
(I'll note that I agree with the general wisdom to store data in UTC; here when I talk about zones I'm either talking about user local machine time or display)
FWIW if Boston switches, it won’t be America/Puerto_Rico, it’ll get a new zone name (probably America/Boston). Tzdb zones express that everywhere in that zone has always been on the same time, since the advent of standard timekeeping, so they always fracture when some subset moves to a different zone.
Ooofff. The two difficult things in software engineering, naming, and timestamps.
This hit me in the early 2000s and now everything I do is in UTC. All dates, timestamps, everything, UTC. If you want to look at a local window in time, convert the window to a utc start and end date and search. When viewing, use a js function to translate the utc date to a local one to print. The mental gymnasium of local to utc to local again…
For example: “this meeting will take place at 10 AM on July 31st, 2026, US Pacific time” cannot be expressed in UTC. You can guess what time UTC that refers to, and you’ll probably be right, but you’ll be wrong if it turns out for example that the US abolishes DST before that date.
Yup, it's a function that leads to a time, depending on some things. And you can't express it in UTC, so you better store the whole function (as a string, for example). QED
> At the time, I went "WTF?" and just commented it out to get it running again. I had bigger fish to fry... and just kind of forgot about it. Everything seemed fine.
You're running a database system and you just casually comment out the configuration setting the timezone?
In what way did everything "seem fine"? SELECT 1 returned something? No further investigation required??
Is there a problem with ISO3166 denoted information in general or is there a specific US issue here? I would think ISO code denoted tzdata was a public good in some sense.
This has nothing to do with ISO 3166 at all. The tzdata database is the work of a single person, Arthur Olson, not a committee (or rather it was, from its birth in the 1980s until 2011 when a company decided to sue him for no reason).
And for most of its life it’s been an explicit policy that timezones are named by continent and representative population center, not by country, to avoid entangling it in territorial disputes and improve naming stability for historical data. The US/* (and Canada/*, etc.) names are deprecated and have been since 1995 (?), but apparently people were still using them because the deprecation wasn’t really apparent unless one was especially into reading release notes.
All the legacy time zones were moved out of the default zoneinfo install. It’s not a us-specific issue but the legacy US/ timezones remain in widespread use, and they stop working on Debian 13 ootb (possibly Ubuntu noble as well?).
> Does anyone know why they are still in widespread use?
Because of a lack of things compelling people to change them until it causes a breakage. And then when it does cause a breakage, most people would rather move heaven and earth to complain, research workarounds, etc. rather than just change it. (Institutional structures can also make "just" changing it far harder than that should be.)
There’s definitely inertia but I think it’s also that the US/ names match official usage: nobody, not even residents, says the time zone is New York because the official name is Eastern time.
Some of this is surely just muscle memory or intertia as well. I remember random config values from when I was trying out linux boxes back in high school that I replicated into files that just don't get touched for decades afterwards.
When was the last time you rebuilt your company's postgres config from scratch?
> When was the last time you rebuilt your company's postgres config from scratch?
Last year, when we upgrade to version 17.
I looked at the example/template configuration, diffed it with our configuration from PG15, and for every change decided whether to keep our version or the new setting.
I didn't use it, but Debian/APT has had a tool to do this sort of comparison for any software upgrade for as long as I can remember.
Do other people just copy the old config and shout "YOLO!"?
When your configuration value from PG15 was different from the old default and the new default on something like the statement timeout, and the statement timeout was previously working fine, and obstensibly set for a reason.... I dunno, are you going to question the value that much?
And you might say "well I know how a statement timeout works" and I agree! I would also generally agree that something like a timezone setting would generally be something I'd expect to be fairly stable.
That's what I meant about rebuilding the config from scratch. Rebuilding from scratch would almost involve _not even looking at the existing configuration_ and then doing first principles footwork to figure out what is needed.
I think the diffing flow you went through is the right way to do it, but I believe that flow might lead to some values getting less scrutiny than others. Still perfectly reasonable though
Still typing "nano -w filename" each and every time since back around y2k when I was working on Linux for the first time I was told that bad things could happen if I didn't...
I suspect some of it will be because the legacy form is a bit more intuitive than the standard form. You don’t really use continents and cities as a reference to time zones normally, countries and local subdivisions makes more sense, but as other people note, it brings up POLITICS.
You don't use them normally in the US, I've been referring to europe/amsterdam or europe/paris all my life in Linux installers and various equipment. I've never ever encountered netherlands/amsterdam or something like that.
It’s the same in Europe as it is in the US. Normal people refer to Europe/Paris as CET, just like normal people refer to America/New_York as Eastern Time.
Toronto's bigger. Same reason Boston and Washington are "New York" and Seatle/SanFrancisco are Los_Angeles
(Detroit gets its own zone because it was significantly different to new york before 1970, normally it's only difference post 1970 which merit a new zone)
The timezone names are defined by the Olson database, not Debian. It is the only sensible system in our ecosystem. It certainly beats Outlook which still wrongly insists that my timezone is GMT just because I live in the UK and still confuses everyone (it isn't during the summer; we use BST over the summer, not GMT).
> It is the only sensible system in our ecosystem. It certainly beats Outlook which still wrongly insists that my timezone is GMT just because I live in the UK and still confuses everyone (it isn't during the summer; we use BST over the summer, not GMT).
I went to both Outlook and Teams to check and I have the option to select both "(UTC) Universal Coordinated Time" and "(UTC+0) Dublin, Edinburgh, Lisbon, London", with the later adapting to changes in the summer; but I do agree it's clunkier than the Olson database, combining multiple regions in a single option while splitting regions with the same timezone rules into different ones.
> ...I have the option to select..."(UTC+0) Dublin, Edinburgh, Lisbon, London"
This is factually wrong already. In summer, London is not UTC+0. They mean "UTC+0 ignoring DST", but that is not useful. If they're going to be specific by specifying a UTC offset, what's the point if it doesn't include DST? How is that useful as an identifier when it's ambiguous? With their history of getting it wrong, this just introduces doubt about its correctness.
Further, if you ask Outlook to show you two timezones at once and do not override labels, it will label BST "UTC+0" (it isn't; it's UTC+1!) while also calling eg. India "UTC+5:30", implying a time difference of 5.5 hours when it is actually 4.5 hours. This isn't just a case of "ah - they actually mean ..."; it's most definitely wrong!
The problem is that it has a very US-centric view of what DST is. You can mostly ignore it in the US when calculating time zones because the entire country changes DST at the same time. This is not the case internationally.
> This is factually wrong already. In summer, London is not UTC+0.
Yeah, the "regular" time is UTC+0, with it changing in the summer. I'm aware it's a really poor implementation, but it is there as a separate option from "UTC" itself preserving the same offset (0) all year.
> The problem is that it has a very US-centric view of what DST is. You can mostly ignore it in the US when calculating time zones because the entire country changes DST at the same time. This is not the case internationally.
Probably the reason they, for some reason, split the setting for "Amsterdam, Bern, Berlin, Rome, Stockholm, Vienna" and "Brussels, Copenhaguen, Madrid, Paris" even though they all follow the same timezone and change simultaneously.
> Your TZ doesn't change between summer and winter. What changes is the shift
My TZ is GMT in winter and BST in summer. I am not in GMT in summer. GMT continues to exist in summer, doesn't shift but my clock doesn't follow it.
The UTC "shift" changes indeed. When I am DST-shifted, calling me "UTC" is absolutely wrong.
The practical issue is that people still use "UTC" and "GMT" interchangeably, which is roughly correct anyway since they remain the same in practice. But then during summer when someone says GMT I don't know if they actually mean BST (they mean my local time) or UTC (they mean the global point of reference). That ambiguity only arises because Outlook (and you, apparently) conflate GMT and BST. It's far more of a problem for those actually living in a UTC-adjacent time zone (do you?), especially because being only one hour off, usually both options seem equally likely in context.
Any moderately reasonable system would be backwards compatible and/or migrate existing values
> The worst part about this is that it didn't get so much as a mention in the Debian 13 release notes. I read through that document before going for it and never encountered it. Indeed, even now, you won't find "tzdata" or "zone" in it.
How and where, in principle, do you suppose it should (be able to) do so, given the functionality? It's just localizing timestamps, and the result might bubble through multiple layers of software before being presented to anyone. It's not as if C APIs get designed up front to return structures with a boolean "btw you called this in a nonstandard way that might fuck up in the future" flag (or, say, a pointer to a deprecation-warning structure; and then that interface has to be rock-solid stable to be of use) for every call. And if they did, nobody would ever write code that checks it.
> How and where, in principle, do you suppose it should (be able to) do so, given the functionality?
System logs?
> It's not as if C APIs get designed up front to return structures with a boolean "btw you called this in a nonstandard way that might fuck up in the future" flag (or, say, a pointer to a deprecation-warning structure; and then that interface has to be rock-solid stable to be of use) for every call. And if they did, nobody would ever write code that checks it.
Yeah, fair point, no-one should be using a C API at this point.
> “Yes,” said Arthur, “yes I did. It was on display in the bottom of a locked filing cabinet stuck in a disused lavatory with a sign on the door saying ‘Beware of the Leopard’.”
The timezones are very clearly marked as deprecated on the lists I looked at, but on the other hand I don't live in the USA so I've never had to look particularly hard. Everything I configure is either Etc/UTC or Europe/London which is nice and easy to remember...
love the takes on this one blog, but "I'm copying an old conf file for over 20yrs and now got this issue" is kinda weak.
American exceptionalism time zones aren't used since the 90s. even the cpus from that time are already dropped from kernel support. heck even the text encoding is gone.
That’s unkind. I want to do the right thing in such cases, but I’m also learning about this today for the very first time. I’ve never, not once, heard that US/Pacific was a bad idea until this post. If not for this, I still wouldn’t know. I thought it and America/Los_Angeles were semantically identically and just kind of symlinks to PST8PDT or whatever.
If anything, the city TZ always felt off, like I was opting in to that specific city’s strange legal decisions or something.
> If anything, the city TZ always felt off, like I was opting in to that specific city’s strange legal decisions or something.
that is exactly what time zones are for :) not being snarky (wasn't before either, i really love that blog!). but the whole reason for tz is to join the ever changing oddities of political bodies from one very specific region.
The geographic extent of legislative time zones can change, they can morph, split and merge, and also appear and vanish completely. In consequence, the only way to unambiguously specify which local time results when adding durations backwards and forwards in time, without being constrained to limited periods of legislative non-change, is to fix a specific geographical location, and to specify how local time has changed and is changing at that particular location. That is what the tz database does. In principle, you could define a separate time zone for each point on Earth, but that isn’t practical. So the compromise is to pick representative cities.
One important thing to understand is that the time zones of the tz database, and hence generally the time zones used in computing, are a slightly different concept than legislative time zones.
I get all that, I really do. And I know America/LA is a reasonable pick: it’s a large, well known city nearby that’s always going to be at the same time I am when I’m at home.
Still feels weird, though. What if LA specifically passes a time zone law so that now it’s sometimes wrong for everyone else in California. Do we add an America/Cali_except_LA zone?
That’s probably hypothetical. It seems unlikely. But what a major pain in the ass if it did happen?
Another city would be picked to represent the rest of California, and the point is that using that city as a time zone ID would then work not only for future times, but also backwards for any times in the past. If, instead, you had America/California, or US/Pacific, those would be ambiguous either forward or backwards in time.
So the trade-off is timezones being specific to a particular city but remaining unambiguous forward and backward in time.
You can’t avoid pain when there is a change in the geographic area of a legislative time zone. But you can avoid the case of a time zone ID becoming ambiguous in terms of the UTC<—>local time mapping it’s supposed to define. The latter is the aim of the present scheme.
That can happen and has. And the result is basically as you described it.
Yes, it's silly and inefficient, but so are time zones. It's not an easy problem to wrangle on a computer, for reasons that are exactly as you have described.
Yeah, it’s easy enough to pick at edge cases, but it’s amazing we have something that generally works this well at all. I don’t know if I have any better ideas, at least ones that the smart people maintaining the DB haven’t already considered and rejected.
Debian is known to have made similar monstrously stupid decisions.
For example, they patch OpenSSH source code in a way that makes defaults behave differently than upstream. In the name of backwards compatibility of course.
I assume this will continue until it doesn't anymore, and the only notification you shall receive from the ivory tower is a cryptic one-liner buried in a changelog somewhere.
> For example, they patch OpenSSH source code in a way that...
Isn't it the same thing with the RedHat downstreams ? (Not necessarily OpenSSH but other packages)
IIRC RedHat do all sorts of things to keep their gov / corp customers happy, also usually in the name of backwards compatability, all of which then end up in the downstreams.
Russ Allbery left over bureaucracy and systemd. It sounds like it's chocked full of people who want power and an excuse to patch downstream to create a cottage industry of quirks, busywork, and codependency.
I prefer real choice and light patches that try to upstream as much as possible, or workaround upstream obstinacy rather than create incompatible idiosyncrasies. One area that isn't well represented in barely a/no distro is init freedom neither married to nor completely divorced from the sprawling octopus.
You are confusing Russ Allbery with me, while at the same time making it sound like I have a problem with systemd, which is not the case. Russ remains a debian developer.
As an actual answer, it's not too bad on Debian; we only really use/need: systemd (system and user), -logind, -journald, -udevd. All in all, not too many tentacles but there are a few...
* https://data.iana.org/time-zones/tzdb/backward
If one traces references, one finds this connected bug on Launchpad. Amusingly to anyone who has ever seen these sweeping timezone database changes over the years, Launchpad marked it as "This bug affects 1 person.".
* https://bugs.launchpad.net/ubuntu/+source/tzdata/+bug/200807...
The rules for the "backward" file are here:
* https://data.iana.org/time-zones/theory.html#naming
All of the US/* timezone names, such as US/Pacific here, have been backwards compatibility measures in place for the whole of the 21st century and some of the late 20th. The Olson database in the 1980s (mod.sources v08i085, comp.sources.unix v14i030) used these names. But the naming scheme changed somewhen in the 1990s to a continent/city and ocean/city form and backwards compatibility with the old names has been preserved ever since by the "backward" file.
* https://groups.google.com/g/comp.unix.solaris/c/ON_MPZxVdv0/...
Debian has moved into a non-depended-upon package a backwards compatibility measure that is as old as Debian itself.