[kamaal]

I imagine in case of a Kernel that should be like a rule beyond any scope of discussion. And may be for any piece of software that is at the bottom layers of any stack.

You just can't break backwards compatibility.

This was a big mistake Python 3 made.

[christophilus]

The problem with almost all non-lisps is that you have to add / modify syntax to get a lot of new features (think async/await), whereas in a lisp, most such changes are done via macros and require no breaking changes to the language itself.

[vorg]

> You just can't break backwards compatibility. This was a big mistake Python 3 made.

At least Python's backers had the guts to release a breaking version 3, unlike some other dynamic languages, specifically Ruby and Apache Groovy. Ruby's backers have tentatively slated 2020 for a (MRI) Ruby 3 release after two more Ruby 2.x versions. As for Groovy, 2 months ago its backers canceled its version 2.6 release which was to backport the planned features of Groovy 3 into a JDK 7 compatible release. Looks like we won't be seeing version 3 of either language anytime soon.

[dragonwriter]

> At least Python's backers had the guts to release a breaking version 3, unlike some other dynamic languages, specifically Ruby and Apache Groovy.

The backward-compatibility breaking Ruby release often held up as a parallel to Python 3—Ruby 1.9—was released December 25, 2007, about a year before Python 3.

If Python had as much guts as Ruby in this area we'd be seeing a backward-compatibility breaking Python 4 around 2021.

[kamaal]

>>At least Python's backers had the guts to release a breaking version 3,

If you have to absolutely must break backwards compatibility. Break it once, and never again, in a way you fix your language permanently. Like Perl 6.

Python broke backwards compatibility for a print statement.

[pjmlp]

Gradle is the only thing holding Groovy alive, beyond maintenance projects.

I remember attending JSF Days in 2009 and how we would be all writing Groovy JEE Beans in the near future, show in different ways across a few sessions, and here we are.

[vorg]

... and Android Studio could be the only thing keeping Gradle alive, beyond 20-liner build scripts. When Google finally releases Fuchsia, it will virtually replace Android's market share within 2-3 yrs. Because Flutter uses Dart for building apps, it'll probably go with Dart for specifying build scripts as well. Although one of the Apache Groovy PMC is probably busy politicking inside Google to get them to use Gradle for Flutter, hopefully that team won't repeat Android's mistake. Otherwise the 3rd-party app market for Fuchsia will be stillborn.

[pjmlp]

Yep, if it wasn't for Google having the silly idea of adopting Gradle, I would never bothered to learn it.

I don't suffer from XML allergy and am pretty fine with Maven in what concerns Maven projects.

Actually I was pretty happy with Eclipse + Ant + ndk-build as well.

To this day the NDK already went through a couple of Gradle build variations and it still doesn't support NDK libraries across NDK projects, as AAR only support binary libs for Java apps.

I follow Fuchsia with attention, but it might end up just as Brillo.

[natbobc]

there was freedom I feel Ruby had that Python didn't which allowed Ruby to successfully go from 1.8 to 1.9;

* Ruby had better dependency management. Python's only recently introduced Pipenv which provides a standard approach for managing development and prod dependencies. VirtualEnv's been around for a while but didn't handle scope.

* Ruby wasn't used as extensively as Python in stuff like system libs and start-up scripts so system packages weren't constrained in the same way.

* Ruby's ecosystem was more focused around web development. Rails in particular was fairly early in adopting new versions of Ruby within about 1 year of the release of Ruby 1.9. Django on the other hand was about 5 years before it had it's first Python 3 compatible release.

* "microbenchmarks" generally got better with Ruby releases whereas Python 3 seemed to have gotten worse from what I remember. I don't think microbenchmarks are terribly useful out of the context of an actual application but many people use them as indicators.

* subjectively I think the Ruby community was more committed to unit testing which made "fearless upgrades" a little more palatable.

[iwintermute]

That's why it's called Python 3 and not Python 2.x

Think about it like totally new language

[decebalus1]

If it's like totally a new language, it should be called TotallyNewython 1.0

[iwintermute]

AlmostTotallyNewthon would be better, correct. Would shift discussion from 'upgrade' to 'migration' without all the usual cries. Other than that it's understandable that some architecture decisions from 1991 haven't aged well.

[Roboprog]

I think “Perl 6” was already taken, so they couldn’t use that :-)