First, there's no easy way for library developers to maintain a codebase that supports both 2/3 development in parallel. Yes, it's easy to use a 2to3/3to2 tool but unlike transpilers in Javascript the output code may still require fine-tuning after the tranform.
This could be solved by adding preprocessor support to python (ie for conditional loading of code) but GvR is adamantly against this approach use to the way many devs write unmaintainable code in c via the preprocessor.
I actually published an OSS project called pypreprocessor that was supposed to solve this issue. It works to add preprocessor conditionals to code but won't work with both 2/3 code in the same file because there's no hook to inline code before lexer runs and throws syntax errors.
Second, package management in python still sucks. Do you use distutils, distutils2, setup.py, etc? There's no single standardized approach that everybody agrees on and each tool follows a unique approach or needs to be supported because of legacy code.
Unlike the Ruby, Javascript ecosystem where package management and publishing is relatively straightforward/easy, it's an absolute PITA to publish packages to PYPI.
Back in 2011, the answer was. Just wait 5 years for everybody to adapt their libraries to support Python 3. 5 years later, nothing has changed. The existing libraries don't want to piss off users by dropping Python2 support and maintaining two separate branches of the code in parallel sucks. So, since most everybody that uses python in production still relies on Python2, Python2 is still the default.
The library ecosystem makes the platform. The tooling sucks so library devs don't want to make the switch.
Until there is a major cultural shift python devs have 2 options. Use a better version of the language without the support of a mature library ecosystem, or use an outdated version of the language with a rich ecosystem.
The ecosystem is rotting for 2 reasons.
First, there's no easy way for library developers to maintain a codebase that supports both 2/3 development in parallel. Yes, it's easy to use a 2to3/3to2 tool but unlike transpilers in Javascript the output code may still require fine-tuning after the tranform.
This could be solved by adding preprocessor support to python (ie for conditional loading of code) but GvR is adamantly against this approach use to the way many devs write unmaintainable code in c via the preprocessor.
I actually published an OSS project called pypreprocessor that was supposed to solve this issue. It works to add preprocessor conditionals to code but won't work with both 2/3 code in the same file because there's no hook to inline code before lexer runs and throws syntax errors.
Second, package management in python still sucks. Do you use distutils, distutils2, setup.py, etc? There's no single standardized approach that everybody agrees on and each tool follows a unique approach or needs to be supported because of legacy code.
Unlike the Ruby, Javascript ecosystem where package management and publishing is relatively straightforward/easy, it's an absolute PITA to publish packages to PYPI.
Back in 2011, the answer was. Just wait 5 years for everybody to adapt their libraries to support Python 3. 5 years later, nothing has changed. The existing libraries don't want to piss off users by dropping Python2 support and maintaining two separate branches of the code in parallel sucks. So, since most everybody that uses python in production still relies on Python2, Python2 is still the default.
The library ecosystem makes the platform. The tooling sucks so library devs don't want to make the switch.
Until there is a major cultural shift python devs have 2 options. Use a better version of the language without the support of a mature library ecosystem, or use an outdated version of the language with a rich ecosystem.