You shouldn't blame Ubuntu here. The Python development community did this to itself.
It offloaded a huge cost of porting literally everything in the universe mostly for syntactic niceties like print and parenthesis. That was an astronomical unfunded downstream mandate. Now, it's paying the consequences. There are codebases which couldn't afford to do that, who are suffering. For example, there's a lot of academic code which was grant-funded, which is widely-used, but with no more funding. Python is used a lot for numerics. That code had better keep working, if you want scientific replicability and apples-to-apples comparison of results.
There are ones who _can't_ do that, for a variety of reasons (many regulatory). Like FORTRAN, Python 2 needs to be around for a long, long time. If I have a system with a multimillion dollar audit / certification process, I can't just up-and-port to Python 3 like a web app.
Bless Ubuntu and other organizations who simply didn't say F-U to all legacy codebases.
I might not even mind if Python 3 was designed to do anything useful: supporting compiled code better, multicore, GPGPU, JIT, or whatever. However, it was mostly minor syntactic sugar. The only breaking change in Python 3 which made sense was probably unicode, and that one didn't have to be so breaking.
It was just so random, stupid, and arbitrary.
Yes, the Python community should deal with "receive random bug reports, random feature requests, random crash reports" and other crap. This should be a very visible failure, or we'll see this again with other projects (not to mention Python 4).
Yes the python 3 upgrade was a failure. The python community has learned from the mistakes. What do you want it to do now? This was long enough ago that some Python devs weren’t born yet.
And it wasn’t “all for a print statement” or something, that’s not the pain of the upgrade nor what’s blocking anyone. You would know that, if you had more than passing knowledge about the upgrade.
Pulling scientific reproducibility as an example is a funny one. Python is the least of the issues in that sector. Maybe let’s introduce Docker there, that would be a good, you know, FIRST step. And then python version is irrelevant anyway.
If you want something reproduced, the entire environment should be reproduced, including library versions, compiler versions, etc. Reproducing independently on newer systems will involve updates anyway.
What would I want the Python 3 community to do now? For a start, I would like the Python 3 community not to post attacks like yours. To quote you: "Wow, how reading this makes me hope Ubuntu Pro fails catastrophically."
This suggests the Python 3 community did *not* learn.
To your point about science: that's not quite how science works. I rarely want to run your code + data + system wholesale. You already ran your code on your data, and unless you fabricated something, I can see the results you got from your publication.
In most cases:
- I want to try your algorithm on new data.
- I want to improve upon your algorithm
- I want to try a new algorithm on your data
... and so on. This allows apples-to-apples comparisons.
I sometimes want to be able to replicate results from the 1900's, and commonly going back to the sixties and eighties. You're on a warpath against people using Python for anything not designed for tech-industry style ephemeral code.
Before Python 2 -> Python 3, this was could usually be done either without changes, or with sufficiently minimal changes that it wasn't outside of the realm of a grad student. The whole docker-style mess of dependencies isn't typically a problem; most code uses numpy+pandas, and occasionally, scipy+pylab.
We have decades of experience with this just working with C, C++, Java, and Fortran code, and it more-or-less works okay at this point.
To your conversation style: I'm very familiar with the differences between Python 2 and Python 3. Insulting people isn't a good substitute for an argument. Docker is less than a decade old. Even if it solved the problems you think it solves (it doesn't), and was usable by non-CS grad students (it isn't), this kind of technology churn is exactly what you want to *avoid*.
I hope Ubuntu Pro succeeds spectacularly, to the point where I'd be glad to pay Ubuntu for my *personal* desktop.
If the Python community accepted Python 2 needs to be maintained for a long time (decades, at least), that would go a long ways.
WinAmp 2 (1998) was widely liked. WinAmp 3 (2002) was considered bloated, and flopped.
So Nullsoft followed it up with WinAmp 5 – because 2+3=5 – in 2003, which was very broadly the codebase of WinAmp 2 (small and lean) plus the skin support from WinAmp 3 (the only part people liked).
This won people back, and WinAmp is still around and got an update this year, 20 years on.
I think it's too late for there to be a Python 5, but I did read a blog post long ago – which I can't find again, or I'd link to it – which proposed a similar compromise fix to Python, in considerable technical detail.
I am with @blagie on this: the Python world handled the 2→3 transition spectacularly badly. V3 didn't deliver enough, and strong-arming people by just end-of-lifing Python 2 and expecting the world to move on was foolhardy and short-sighted.
(And I don't even use the language myself. I'm just observing.)
It's a real shame Tauthon didn't get more traction and support.
And PHP 6 didn't really happen -- AFAICT as a total outsider, Unicode support proved too hard and it was never released; the community backported the important bits to PHP 5, and then a new PHP 7, more modest in scope, developed from PHP 5.
The Python world could have done the same, and Tauthon was an effort in that direction.
It's too late now. I suspect that, just as Perl has lost a massive amount of interest and use, partly from the nearly-two-decade-long effort to release Perl 5, Python has done the same -- sabotaged its own community with this high-handed "your leaders know best" approach.
I didn't know about tauthon. If I had, I would have thrown my support behind it. That makes a lot of sense.
With programming languages, I feel like there's so much dumb, blind luck involved in which ones win. I personally dislike Perl, but Larry Wall is so clearly and obviously a genius. Everything he does has good, logical reasons, and the Perl 6 mess was at least an understandable mistake and well-reasoned. I can see myself making similar mistakes.
The Python one was just dumb and obnoxious. So much of Python 2→3 feels like a random walk, with things changing, rather than improving.
Is this better? Sometimes. Certainly not when you're working with students, which is a major use-case for Python.
Changes since then have felt equally half-baked. async is important, but no one thought through how it might work in a terminal, compatibility with non-async code, and especially with coroutines using yield. I would have the rate of progress be half, but see things implemented correctly and intelligently.
> With programming languages, I feel like there's so much dumb, blind luck involved in which ones win
Very true.
Among my favourites:
Apple's Dylan was an amazing language with an amazing development environment. Ignored.
Oberon, the last scion of Pascal, is a great clean blindingly fast language, and simple too, and the OS of the same name is a thing of beauty.
Both ignored.
It's OSes too.
I mean, UNIX was good, for its time... but Plan 9 was better, in terms of network and GUI integration, but the world ignored it. Inferno is better than Plan 9, in terms of integrating the Taos virtual-processor CPU-independence (and a saner GUI than Plan 9, too) but it went nowhere either.
(Yes, licence agreements had something to do with this, but Linux was neither the first nor the last free UNIX. Before it was BSD, notably 386BSD, and before 386BSD was Coherent (not free, but cheap, and showed it was doable.) After Linux... there are dozens.
If people learned from the model, then even if Plan 9 was under some restrictive licence conditions, it could have been cloned, just as UNIX was, repeatedly.
Concurrent DOS was pretty amazing in its day.
GEM was amazing, but only Atari ST fans noticed.
EPOC, EPOC32 and Symbian were great OSes. Symbian is FOSS now. Nobody cares.
Taos was amazing and should have been the new Amiga.
QNX still is amazing, but Blackberry 10 flopped. Huawei's HarmonyOS is vapourware and marketing: it should just licence BB10. It's real and it works.
A lot of that was not technical, but I think I'd call roughly half of those random.
For example, I feel like what Linux did uniquely well was the development process. The Linux development process, circa 1995-2000, was unparalleled:
- The community was incredibly open and welcoming. Don't let brash and harsh language be confused with being unwelcoming. Anyone could join.
- There was thorough code review. Low-quality code was not let in. This was an amazing learning process.
- The design decisions were all brilliantly documented on LKML. The quality of discussion there was exceptional.
I attribute a lot of this to Linus' leadership, as well as that of lieutenants like Alan Cox. On top of that, you had really robust corporate support in Red Hat (and, for a while SuSE), and an amazing community, package management, etc. around Debian. There were friendly IRC channels for tech support.
The social components were super-important. And yes, licenses too: The GPL was the right license to prevent parasites.
It offloaded a huge cost of porting literally everything in the universe mostly for syntactic niceties like print and parenthesis. That was an astronomical unfunded downstream mandate. Now, it's paying the consequences. There are codebases which couldn't afford to do that, who are suffering. For example, there's a lot of academic code which was grant-funded, which is widely-used, but with no more funding. Python is used a lot for numerics. That code had better keep working, if you want scientific replicability and apples-to-apples comparison of results.
There are ones who _can't_ do that, for a variety of reasons (many regulatory). Like FORTRAN, Python 2 needs to be around for a long, long time. If I have a system with a multimillion dollar audit / certification process, I can't just up-and-port to Python 3 like a web app.
Bless Ubuntu and other organizations who simply didn't say F-U to all legacy codebases.
I might not even mind if Python 3 was designed to do anything useful: supporting compiled code better, multicore, GPGPU, JIT, or whatever. However, it was mostly minor syntactic sugar. The only breaking change in Python 3 which made sense was probably unicode, and that one didn't have to be so breaking.
It was just so random, stupid, and arbitrary.
Yes, the Python community should deal with "receive random bug reports, random feature requests, random crash reports" and other crap. This should be a very visible failure, or we'll see this again with other projects (not to mention Python 4).