Why Do Projects Continue To Support Old Python Releases?

On Planet Python, Gregory Szorc asks why many projects continue to support Python releases prior to 2.7, when they are all EOLed (2.4 in 2008, 2.5 in 2011, and 2.6 last October), and Python 2.7 provides a clear upgrade path to Python 3. Quoting: "I think maintainers of Python projects should seriously consider dropping support for Python 2.6 and below. Are there really that many people on systems that don't have Python 2.7 easily available? Why are we Python developers inflicting so much pain on ourselves to support antiquated Python releases? As a data point, I successfully transitioned Firefox's build system from requiring Python 2.5+ to 2.7.3+ and it was relatively pain free." Shortly after posting, other developers responded with their reasons for using older Python releases. First, Rob Galanakis of CCP (EVE Online) explains the difficulties involved in upgrading a mature commercial project embedding Python. Nathan Froyd adds "I think this list of reasons to upgrade misses the larger point in providing software for other people: You do not get to tell your users what to do. ... Maybe those users don’t have sufficient control over their working environments to install a new version of Python. ... Maybe those users rely on certain APIs only available in older versions of Python and don’t wish to take an indeterminate amount of time to rewrite (retest, recertify, etc. etc.) their software. ... Maybe those users are just rationally lazy and don’t want to deal with downloading, configuring, and installing a new version of Python, plus dealing with inevitable fallout, when the old version has worked Just Fine for everything else."

Comments

When upgrades break code

[fyngyrz]

The problem with upgrades is the developers don't take nearly enough care to ensure compatibility with existing software. For instance, I moved a Python script from an older machine (redhat 9) to an Ubuntu 12 system with a later 2.x series Python, and the script immediately bombed out. Turned out that I'd used "as" as a variable name, and it had become a reserved word in the interim. Elsewhere, I was adding ints to floats or vice-versa; now that produces an error. Somewhere, the behavior of "global" changed. I had a procedure written with global VariableName up front so the procedure could see a global switch. I handed the script to someone else, and they had to *remove* that to make it work.

I moved a large system depending on perl scripts (yes, I know, not Python, but the issues are identical) across the same machine pair and it was so broken under the new perl that it took me two months (and cost my client a great deal of money) to adequately debug and update it. From external module changes (use DBI) to how references to hashes were handled, it was one thing after another. What made me fume the entire time is that this stuff worked *perfectly* under redhat 9 with the earlier perl -- it really wasn't broken. Bloody perl was broken! Worse, I'd abandoned Perl for Python years ago, and my Perl skills were very rusty... that's tough when you're trying to work through tens of thousands of lines of sophisticated code. I truly regret every line of Perl I ever wrote, and cringe at the idea of having to revisit it for any reason at all, because Perl, whatever its true merits, is a language that I was destined to hate to my very $core;

This kind of thing makes me extremely unwilling to upgrade; for that matter, it can send me hunting for the old version to install in new machines, although that can present its own set of problems and requires at least some of a different skill set than programming that really leans more into system administration, which I am not in the least enthused about. For some strange reason, enjoying research on auroras does not automatically carry with it a predilection for meddling with OS environments.

This is all part of a larger gripe set I have where makers of languages and APIs break existing code in favor of new ideas. Here's my position: If you ship the language with X features, you have NO BUSINESS breaking ANY of those features. You have a new idea? Fine. Implement it as something completely new within the language or the API. If it's got new keywords, provide the programmer a switch to enable them -- words that were not reserved before may have crept into the user's namespace(s.) Never take a call out of an API. Never make an API call work differently. If you think you need to do that, then *I* think you need an entirely new API. If you even use the word "deprecated", I will burn you in effigy and spit on the ashes. Furthermore, if you do it right, you will NEVER need to say "end of life" because the most current version will still run the first code ever written for API or language 1.0; upgrades will actually be upgrades rather than incompatible replacements.

Windows got this very nearly right for a long time, then with XP, they began to break everything from window metrics to system calls. By the time Windows 7 rolled around, apps that worked perfectly looked like trash and worked even worse, and I tossed my Windows machine in the can both for development and as a user. Apple's guilty of this too, even going so far as to obsolete their own applications and feature-sets. One really irritating example of this is having bought Aperture 1, 2, and 3 for OSX 10.6, I bought a new camera, a Canon EOS 6D. Works *fabulous*, pretty much the first DSLR I've ever been actually satisfied with. Both Lightroom and Aperture required an update to read the RAW images. Lightroom upgrade went smoothly. The Aperture upgrade? Told me that in order to install the updated RAW support, I'd have to upgrade the OS. But 10.7 and later are not highly compatible upgrades, I have to support OSX