That's nonsensical since it solves a completely unrelated problem.
2to3 was conceived of as a one-shot migration tool for Python 2 codebases. Not as a way to build cross-version Python, and not as something to use repeatedly (which it why it is relatively simplistic and fallible), just as a way to take an existing codebase and handle the first 90% of migrating to Python 3, leaving you with the second 90%.
2to3 was _conceived_ of as migration tool, as if evolutionary lineages just all stopped, changed their genome and restarted. Except it only refactored some easy stuff and let the rest fall on the floor.
Six has had way more impact than 2to3. It was only because core Python said running a single 2/3 codebase was UnPythonic that people didn't embrace it sooner. Running a single codebases across both 2 and 3 is now the defacto technique.
Writing in 3 and "compiling down" to 2 gets people writing 3 which is the point, even if they can't run CPython3.
There's nothing to be charitable about, GP's comment is rather clear and stands for itself.
> 2to3 was _conceived_ of as migration tool
And it still is one, so far as I know, which is actually part of the issues with it.
> Except it only refactored some easy stuff and let the rest fall on the floor.
I'm rather well aware of that, which you'd have noted if you'd actually read my comment.
> Six has had way more impact than 2to3. […] Running a single codebases across both 2 and 3 is now the defacto technique.
Sure, but again that has nothing whatsoever to do with the inanity of ranking 2to3 and backwards.
> Writing in 3 and "compiling down" to 2 gets people writing 3 which is the point, even if they can't run CPython3.
You do realise that's complementary to 2to3 right? A big issue of 2to3 is that once you've converted your P2 codebase to pure-3 (not a common subset of 2 and 3) you either have to maintain two different branches making packaging and backporting difficult or you just leave your P2 users out in the cold.
Once again, the purpose of 2to3 is not to regularly run 2to3 to generate a Python3 release from a Python2 source.
I don't think it is nonsensical. There was a tool 3to2 that took Python 3 code and produced Python 2 code. You had to write in a subset of Python 3 but the idea was sound, I think. Python 3 is more strict in ways than 2 and so downgrading the code was easier (e.g. no implicity str/unicode coercions). It looks like 3to2 is not maintained but I liked the idea.
If I were writing a library that needed both 2 and 3 compatibility, I would prefer to write Python 3 code and then run 3to2 on it, rather than going the other way. My own solution is to fork the library, provide only bug fixes for the 2.x version and do all new development in Python 3 only. Porting has taken some effort but not as much as I feared. In retrospect, I'm happy to have invested the time.
It is completely nonsensical in the context it is made in, the purpose of TFA has nothing to do with the purpose of 2to3, it makes no sense to "rank" them.
> There was a tool 3to2 that took Python 3 code and produced Python 2 code. You had to write in a subset of Python 3 but the idea was sound, I think. Python 3 is more strict in ways than 2 and so downgrading the code was easier (e.g. no implicity str/unicode coercions). It looks like 3to2 is not maintained but I liked the idea.
That still has nothing to do with 2to3, whose intended purpose once again is as a helper for a one-shot conversion of an existing Python 2 project to Python 3.
> If I were writing a library that needed both 2 and 3 compatibility, I would prefer to write Python 3 code and then run 3to2 on it, rather than going the other way.
And guess what? If you're starting from a Python 2 project you can use 2to3 to convert it to Python3 then use some other converter to provide P2 releases.
Forward adoption is one thing the JS ecosystem does well (probably due to the sheer necessity of it given the many varying platforms outside control of the developer). Transpiling supersets of the language to the lowest supportable denominator is such an elegant approach. Thanks to projects like babel, typescript, webpack, etc, you can write your code in the most modern language spec, but target platforms not yet supporting it.
I suspect this may be easier to achieve with JS than other languages because its prototypical nature makes it easy to compose dependencies. However, there is certainly a lot that other communities can learn from the JS community's solutions to this problem.
In theory more than practice. For example if you use array destructuring (`[x, y] = something`) and Babel cannot verify that `something` is always an array it inserts a 664 character function that handles generators and other things. That adds performance and size overhead (grep for _slicedToArray in your compiled js files), when all you really want is `x = something[0]...`
It's a solution to a problem that exists because browsers incrementally support different features of the language, instead of upgrading all at once. That and the older browsers in use when you want to maintain backwards compatibility.
The only reason the Python 2/3 issue exists is because of just enough breaking backwards compatibility in a way that a lot of people didn't agree with in the community. This isn't a thing in Ruby.
So the goal should be to have codebases in Python 3.
> python3 should be the input and not the output
How would this accomplish that goal? All the codebases would be in Python 2 then. I bet that removing barriers to run Python 2 code would cause it to stay around longer.
All the code bases are already in 2. This would allow someone to write new internal libraries in 3.6, or migrate existing code one at a time, but still have the project as a whole run on a python2.7 until it is ready.
Ever since mypy started gaining attention, I’ve been looking for a way to write using Python 3 annotations and strip the annotations for release so that the code can run on Python 2.
I never was able to find anyone talking about that goal with mypy, but it looks like this tool may be a solution.
A writeup on how you handled mapping bytes, bytearrays, unicode, etc, to 2.7 constructs would be interesting.
In code that supports both 2 and 3, I end up with ugly stuff testing sys.hexversion to deal with 3rd party libraries, like pyserial, that expect str in v2, bytearrays in v3.
I've scanned the source code and it doesn't look like it handles the bytes/unicode thing. Which, while the project is cool, will mean it won't work for a ton of Python 3 code.
The implicit coercion between bytes and unicode was one of the biggest hurtles I ran into when porting a relatively large (30k+ lines) code base from Python 2 to 3. Also challenging was the change in comparisons (None no longer smaller than all types, TypeError when trying to order disjoint types). While this looks like an interesting hack, I can't see anyone seriously using it to backport code to Python 2.7.
As I mentioned elsewhere, I created "ppython" to handle these two porting problems. The source code is on github: https://github.com/nascheme/ppython . If someone needs a Windows binary, I could build it if you ask politely. ;-) For my project, it has been a useful tool to speed up porting.
Porting code is not trivial and there is huge amount of Python 2 out there. I wish more effort had been spent building tools to help porting of code. Even simple things like disallowing the 'u' prefix on strings in Python 3 was a big mistake. Mostly those things have been corrected but it is still going to take a very long time to move the majority of the community to Python 3. There will be Python 2 running for the next 50 years easily, I'm sure, maybe 100 years. At least Python is open source and you will not be stranded like VB developers where when MS drastically changed Visual Basic.
Maybe not, but the incremental work was less than making a single codebase work for both 2 and 3. It's code that runs on openwrt, so forcing my ideas on the users is painful for them...they would have to build a whole new image.
"Why should I switch to 3.x?" You haven't provided a reason to switch.
I don't know if it was your intent, but your question assumes that Python 3.x is the "default", and that there is some sort of obligation to use it instead of 2.x. That's a premise that a lot of people do not share.
As an example, they may not view 3.x as an "upgrade", but as a different language. So you might as well ask anyone doing work in one language "What reason is there for you to still use language X? Why not use language Y?"
As long as some team is willing to support Python 2.7 (and possibly backport non-breaking features), 2.7 will live on, and there is no reason it should go away. The only strong reason for many to switch to 3.x is "I have a library I need to use that is supported only in 3.x". Or "I need to hire developers, and I can't find people who know 2.x, but I can find those who know 3.x"
Languages are tools. As long as the tool is more than adequate for the task, the burden is on others to justify a change in tool.
The Python project itself is dropping support for Python 2 by 2020. Lots of major libraries have pledged to stop supporting Python 2 by 2020 or even before.
Writing new code in Python 2 is completely nuts and just makes everyone's life harder (including your own).
>Writing new code in Python 2 is completely nuts and just makes everyone's life harder (including your own).
And when it makes someone's life harder, they will change. If it's too late for them and affects them, they will have themselves to blame.
This, really, is the only reason for people to change to Python3. Yet people keep evangelizing all its great features, etc as if the existence of a much better language suddenly obligates everyone to switch to it.
It's fine to point out potential repercussions of not switching now. But moralizing it is counterproductive.
Thank you for signing up to help in the porting effort for FreeCAD and Pyspread to Python 3. Free software projects depend on volunteer contributors like you.
So I've been following FreeCAD since it started. Even tried to get my head around the code base at one point, but the hacking guide was mostly nonexistent in those days, so I moved on to projects that did not skip that very important step.
Just to put a stake in the ground, when did FreeCAD start? When did 3.0 come out, along with the announced end-of-life date for 2.7?
You can absolutely do string handling in a common subset of Python 2 and Python 3. It won't work for legacy string handling code if you relied on implicit coercion, but then the same apps are unlikely to support unicode either.
> It won't work for legacy string handling code if you relied on implicit coercion
That's far from the only pitfall. For instance some of the P2 stdlib relies on implicit coercion to bytestrings whereas the Python 3 version has been properly converted to unicode (I discovered that when I tried to disable implicit coercion in one of the codebases I work with for cleanup/migration purposes).
So cross-platform string handling is not just a matter of properly splitting bytestrings and textstrings, but also finding out where "native strings" need to remain.
Distributing a tool to OS distributions that do not have Python3.x (e.g. RedHat/CentOS 7) where the requirements include 'Not installing any additional packages'
I hate it because Python2 is a flaming garbage heap, and 2.7 is what happens when you piss on a flaming garbage heap to put it out, whereas Python 3 is actually a nice language to work in, but until OS distributions get their act together on the Python front, some stuff is going to continue sucking.
2 and 3 have like a 90% overlap. How can you call one a flaming garbage heap and the other a nice language?
I worked in Python 2 daily for about a year and a half (doing side stuff in 3), and switched over to Python 3 a few months ago. It's just not that big of a deal. Maybe it's because I'm pretty shielded from the madness of strings vs. bytes (or just import from future).
>How can you call one a flaming garbage heap and the other a nice language?
----------
try:
import queue as queue
except ImportError:
import Queue as queue
----------
They couldn't even consistently name portions of the standard library. Python 2 has a few random things where the first letter is capitalized while 99% of everything else isn't.
There's all sorts of things like this, where poor design decisions cause you to sit there with question marks appearing over your head. That 10% takes Python from being a joy to work with to being incredibly frustrating. I would be happy if I could just work exclusively in 3, but making sure things are backwards compatible to 2.7 is a frequent source of driving me up the wall.
>Maybe it's because I'm pretty shielded from the madness of strings vs. bytes (or just import from future).
If you lose this shielding it will help you understand my vitriolic response as well.
> How can you call one a flaming garbage heap and the other a nice language?
The lack of UnicodeDecodeErrors.
Just because there is a 90% overlap doesn't mean one can't be a pile of trash (i think that's a bit harsh though). The improvements under the hood are huge (unicode, no 'new-style' classes, speed improvements etc) as well as syntactic sugar means that as time goes on Python 2 looks worse and worse.
Serious Answer: Because some of us have some rather large perfectly working systems written in 2.7 and using lots of different libraries that moving to 3 is a major project requiring a good amount of time and effort.
Not all of us have hobby-sized projects on the go.
OK, but this has been a very long transition. There has been lots of time to evolve to 3.x. Has the primary barrier been the slow transition of library dependancies?
I'd really like to understand this from a technology transition management standpoint.
If you'll allow a strained analogy, the 2.7 to 3.x transition appears in retrospect like an exercise in herding cats, because the transition was done using cat-herding style incentives. What should have been done differently to make the transition into greyhounds chasing a rabbit? What should the rabbit have been?
Any technology that lives long enough eventually has to transition its customer base. I'm trying to learn to identify rabbits.
It's been a long transition, yes. So long, that 4-5 years ago, developing a new application in 3.X was a non starter. Hence, 2.7.
Now, you have 2.7 apps that are very mature, stable and reliant on lots of little things. Okay, big job.
Now let's throw another wrinkle into this. You can pay good money to develop your next feature, or to port your system to 3.x. One adds real value to your customers and improves your bottom line. One lets you say you are on 3.x and gets nods of approval from random programmers.
What do you think a business is going to do? Exactly.
When will we eventually port? Sometime shortly after 3.x becomes the overall standard. THIS is actually what we are seeing right now, which is great, but that hasn't been the case up until the last year or so.
>Any technology that lives long enough eventually has to transition its customer base.
While true, this costs money, time, and effort. So you better be damn well sure there is a good ROI that is something more than "This code that nobody but programmers see is more elegant and properly structured".
There's a reason banks are still using COBOL after all...
Asking this question without disqualifying the answer "No breaking changes whatsoever" usually just generates a bunch of different ways of phrasing that.
No, I don't buy that. IBM successfully transitioned a huge customer base from 709/7090 36-bit words to System 360 32 bit words. The transition from 32 to 64 bit addresses was rocky for a lot of architectures. DEC Alpha, Intel Itanium. But are you still running a 32 bit OS?
K&R C no longer compiles. Not that the C to C++14 transition is what I would call an example of greatness, but it is somewhere along the success spectrum.
Seriously, you will someday need to transition your customer base. How do you plan to do it?
foo@virt-ubuntu:~/c$ gcc --version
gcc (Ubuntu 6.3.0-12ubuntu2) 6.3.0 20170406
Copyright (C) 2016 Free Software Foundation, Inc.
This is free software; see the source for copying conditions. There is NO warranty; not even for MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.
foo@virt-ubuntu:~/c$ cat 'k&r.c'
#include<stdio.h>
int main(argc, argv)
int argc;
char *argv[];
{
puts("hello, world!");
return 0;
}
foo@virt-ubuntu:~/c$ gcc 'k&r.c'
foo@virt-ubuntu:~/c$ ./a.out
hello, world!
Yes, I agree with you. That's why I said you should disqualify that answer. It's by far the most common answer, but it's also obviously not the right answer.
Mac still doesn't come with python 3 out of the box, so if i want to distribute a script to people without requiring them to install anything else, python 2 is my only option.
I don't get why Apple doesn't include python 3.x since they can easily slot it as they do for 2.6 and 2.7, and there was no licence change that I know of (in contrast with e.g bash 3.x vs 4.x) to justify that.
Slightly tangential but system ruby is stuck at 2.0.0p648 (which is well into unsupported) and perl v5.18.2 (which I don't know the status support of but is old enough to wonder).
They might as well rip them out and make them an optional package like CLI dev tools for possible backwards compatibility needs (easily installable either with something like xcode-select --install or the java/javac stubs).
Total guess, not a Mac guy, only know Apple by reputation: maybe they aren't interested in spending any resources for anyone to have modern tools that aren't Apple tools. "Y'all do what you want. If it doesn't come from us, you're on your own."
As others pointed out, the cost of moving from 2.7 to 3 may simply be to high. Remember that while much can be done automatically, you still need to test that all your code still behaves as expected afterwards. I was one a project that did migrate a rather large web application and it was pretty painless, but I can easily imagine more complex code bases having issues.
I don't really see the benefit of a Python 3.6 to 2.7 transpiler/compiler though. More realistically I would require that all new code be both 2 and 3 compatible, for when the eventual migration will happen. This isn't necessarily and easy task either, but then you're only depended on the Python project, not some random transpiler that might not see further development.
if you have a large py2 codebase in a shipping product, can you make an ROI argument that porting to py3 makes sense given the staff months required to do so?
I'm working on something right now in 2.7, a personal project. One of the 3rd party tools has a note about 3 with just enough of a "mostly" that I just don't want to encounter the speed bump or road block or whatever it's going to be. And also, the code isn't really the major part of the project.
> I like lambda a, b: a + b syntax better than lambda a_b: a_b[0] + a_b[1]
It already works this way.
Python 3.6.1 (v3.6.1:69c0db5, Mar 21 2017, 17:54:52) [MSC v.1900 32 bit (Intel)] on win32
Type "help", "copyright", "credits" or "license" for more information.
>>> add = lambda a, b: a + b
>>> add(1, 2)
3
> I prefer map/reduce/filter to return lists rather than iterable
You can produce a list from any iterable by passing it to `list()`. You cannot take a materialized list and make it lazy though.
> I prefer dict.keys/values/items to return sets/lists rather than iterables, unless I call dict.iter[keys/values/items]
Why? Sets are mutable. What happens when you mutate dict.values? It doesn't make sense.
lambda a, b: a + b kind of works, but you can't call it with tuple argument (which you could in Python 2.7). This hurts when e.g. you have a list of pairs and try to map when via lambda, e.g.: map(lambda a, b: a+b, [(1, 2), (3, 4)]). Even worse, this won't fail right away in Python 3 (thanks, lazy evaluation in map), but will raise when you try to use the result of map
Your first point is incorrect, lambda works with the first syntax for both python 2 and 3. What /has/ changed, however, is the implicit destructuring:
l = lambda (a, b): a + b
l((1, 2))
I suspect there are very good reasons not to allow something like this.
With regard to the second point, I also would have liked a more "gradual" step there, I find myself (especially in REPL environments) often doing `list(map(sth, sth))`. A `mapi`, `filteri` or something like this would probably not be zenny enough.
Both don't pose very strong points for python 2 > 3.
These seem like OK reasons, but I find myself continuously needing the destructuring idiom in lambdas, while these introspection and documentation concerns are just not there for me.
However tuple-parameters being an exception to rules (such as args and kwargs not being usable with them) is more compelling. And allowing destructuring only in lambdas would be weird, because they wouldn't be normal function objects any more.
Unless the destructuring was only syntactic sugar for the translation in the PEP? But that's again inconsistent.
Yes, lack of implicit tuple unpacking in lambda is what forces me to write `lambda a_b`.
Python 2.7 has imap and ifilter in itertools. Python 3 could have imported them into global namespace to simplify usage without breaking map/filter functions.
In my case: MacOS has python 2.7 pre-installed, and python3's lack of backward compatibility is mildly annoying while giving very little actual improvements in return.
> Python 2 is like FORTRAN. It might not be sexy anymore but it's not going anywhere.
Except Fortran is, and is going to be, further developed. (So, if a new useful programming concept appears, or a major design mistake is discovered, it always can be patched.)
Python 2 is going to be abandoned in 2020. Python 3 will supersede it.
Abandoned by whom? ;-) I don't plan to abandon it.
Elsewhere in the thread xrange has mentioned Tauthon. And there are more Python 2 interpreters than just the main C Python.
Pypy may stop supporting 2 syntax officially, but I doubt it. Since it never changes, there's next to no overhead to keeping it available.
And further, if your language isn't changing then maintenance and debugging become much easier. Conceivably the codebase(s) can asymptotically approach 100% correctness.
Well now you're getting onto my personal reason why I don't adopt Python 3. Python 2 will remain just as useful as it's been. Python 3 will not significantly extend that usefulness.
So I don't think there's much to be gained.
The 3 syntax and new features are not massively better than the sweet spot hit by 2, they're not even incrementally better. They're just massively more complicated.
So the cost/benefit ratio of Python 2 to 3 just isn't there. People are changing up because it's "the done thing" not because they really gain anything.
I was surprised as hell when I realized I wouldn't use Python 3, but it is a rational and considered opinion.
I think it means that you tell the compiler what version of Python the original source code is in, so it knows how to compile it. So, if your codebase has f''-strings then the compiler needs to know.
We've already had tools like 2to3, but they weren't all that popular, and this won't be either. Most things are ported to Python 3 now and most new projects just use Python 3. Never fear.
I never understood the insistence in this. A transpiler is "a compiler that targets a high-level language." It's a term that adds information above and beyond compiler. All transpilers are compilers, not all compilers are transpilers.
You say it doesn't make sense, and yet you were able to figure out how to apply it correctly to the example you gave, so you have to have some understanding of the idea.
Like, in most American homes, there's a bathtub with a showerhead in it. When we're using the showerhead, we'll say, "I was in the shower..." When we're using the faucet, we'll say "I was in the bathtub." This sort of thing happens all the time. It makes plenty of sense that something is a transpiler when it's transpiling and not when it isn't.
But isn't that what this is? I was under the impression that a compiler turns source code into machine code whereas a transpiler turns source code into differnet source code?
All compilers turn one form of code into another. Whether that's C to assembly (to machine code), or JavaScript to bytecode to machine code, or Python to Python. Transpiler is just a term referring to a compiler taking one high-level language to another. It's unnecessary because a Coffeescript to JavaScript compiler is a transpiler already; saying "transpiler" there tells you nothing you don't already know.
Interesting but I wonder when you would need a tool like that. I have something related, a fork of Python 3.6 that adds a bunch of backwards compatible behaviour so 2.7 code runs more easily, https://github.com/nascheme/ppython .
