| HN Mirror

A language that is constantly changing in fundamental ways is not usable for serious software. Languages are largely constrained by the choices that were made when they were released. See Python 3 for an example of what happens when a language tries to make breaking changes — and Python's were relatively minor! No language could thrive while going through that sort of strife constantly.

danudey 4392 days ago

Objective-C has added several features in recent years; things like bindings, dot notation, etc. are relatively new, but pretty substantial changes in terms of code clarity.

As for Python 3, the Python community never intended for people to move over to Python 3 immediately, nor was it intended for people to move their Python projects from Python 2 to Python 3 unless they had some reason to (e.g. Django, public libraries, etc).

Objective-C has added things, but it has largely not changed existing things. As I said in another comment, languages generally do not change — they just accrete.

If you feel like a language is missing something, that might be fixable — but if you feel like a language either added the wrong thing or did something the wrong way (which is the OP's concern), that is a much more difficult problem, because after release you can't take away what's already there without making people angry.

wvenable 4392 days ago

Languages generally don't change -- they get added-to and very rarely removed-from but they don't often change. For example, it's usually very hard to add new keywords. So C++ for example, uses every possible keyword and symbol overloaded to provide new features to the point of ridiculousness. Java is on the same road. Python changed a few things and the split between 2.x and 3.x is still ongoing.

This was about the pace of language development once upon a time. C++ changed glacially for its first 20 years, for example. Perl, Ruby, and Python also took pretty conservative approaches to language evolution (but kudos to Python for finding a way to encode versioning information into the code itself). I think Objective-C didn't really change all that much until the mid-2000s, probably in prep for iOS?

So the expectation seems like it's probably based on real history, but things seem different now, though. Languages are evolving faster than before, even some of the ones that previously moved very slowly.

By and large, languages don't really change — they just accrete. It is very rare for something that used to work one way to later work another way. For example, languages generally do not go from being statement-based to being expression-based, or go from something being mutable to immutable, or eliminate operators. (MzScheme did the second one — it went from mutable to immutable defaults — and it was considered so significant that they stopped calling their language Scheme and renamed it Racket to avoid confusion!)

I agree this has historically been true, but again, I think this is changing. Ruby 1.9 and Python 3 both did more than accrete, they actively broke existing code in quite significant ways.

C++ has so far avoided completely breaking changes, but with all the accretion it's doing now it's probably only a matter of time before some significant breaking changes happen lest it become even more ridiculously complex than it is now.

Go has had breaking changes as well, I believe, but they have a smart upgrade tool to help with it. This is probably something that will catch on for other rapidly evolving languages.

I think we'll see a lot more of this kind of thing in the future.

Ruby and Python each did one release where they were willing to do significant changes. That's it — they're not doing it again for a good long while now. It's an isolated incident, not a trend in those languages' development practices that we can project into the future.

Go did breaking changes pre-1.0, but they are now committed to providing a stable platform that only accretes features (http://golang.org/doc/go1compat).

I don't expect either of them to do it for a while either. The trends I'm talking about are in PL development in general. More breaking evolution is taking place post-initial-development than ever has before in languages both new and old.

Note that the Go compat wiki you link to acknowledges a future Go 2 that may break compatibility. That's actually a pretty strongly pro-evolution statement compared to past languages.

The statement you were questioning is that Swift will be "fixed for a decade at least" after release. If you agree that it will be at least four years before Python makes any more changes like Python 3, then you are agreeing with the OP.

danudey 4392 days ago

Python 3 is also not meant to run Python 2.x code. It's pretty trivial to port most code from Python 2.x to Python 3, but there's no real reason to do so unless you're writing a library or other project for other people to take advantage of.

It was intentionally done as a 'clean break' release; 2.x keeps doing the same stuff it used to do, and Python 3 changes a bunch of stuff which, in hindsight, makes sense (such as a distinction between 'stream of bytes' and 'string of text', vs. 'stream of bytes which may or may not be ascii text' and 'string of unicode text').

That said, there's no real benefit to moving an existing project/codebase from 2.x to 3.x, and it was never intended that there would be one. Python 3 is for new projects; Python 2.x is for existing projects, or new projects which need deployment in older environments.

Which is one of the reasons I think it's a sign towards acceleration. People are developing strategies to deal with language evolution. This, "import from __future__", and gofmt are all tooling that helps you deal with a language that's still willing to evolve.

That said, I don't think this part of the python3 effort entirely succeeded. Unfortunately they share a package source, so a lot of projects do need to support both. But whatever failings there have been in the python jump to 3.0, they're nothing compared to the disaster that was ruby 1.9, even though I think 1.8.7 is truly relegated to legacy now.

As long as they learn from those issues, though, I think the future is bright for non-stagnating languages.

samth 4387 days ago

This isn't right about the renaming. The switch from mutable pairs to immutable pairs happened in release 4.0 of PLT Scheme, in June 2008. The name change was in May 2010.