Teaching seems to be one way to avoid this. You have to keep defending your decisions. Switching specializations (or creating new ones) is another. The only other way I know is to seek outside help, and that level of humility is rare in experts.
We're currently re-building all our systems to target RHEL6 because that's what the execs decided, if we're lucky the project will be done before it's end of lifed. They also decided that we'd be using modern c++, more modern than the GCC that ships with RHEL6. Thanks to these largely incompatible requirements the team responsible has practically built their own distro on top of RHEL6, with a newer gcc and a rebuild of all the libraries we use complete with it's own package management system. I think it's customized enough that a highly doubt redhat could off us any support anyway.
We just got off of RHEL5, so I can understand why there are people still running 6.
I recently interviewed at a startup (though a reasonably far-along/mature, if not spectacularly successful by VC standards) that was both running RHEL6 and depending a remarkable amount on "enterprise" gear, software, and support, which seemed incongruous for any small company, let alone a startup.
They did see it as worthwhile to "migrate" to CentOS6, but I'm pretty sure the interviewer who was the incumbent rejected me quickly and strenuously after I admitted a preference for Ubuntu (due to its package ecosystem better catering to a wider array of developers, making my DevOps culture evangelism and implementation that much easier) and a focus on frugality first when it comes to hardware (that being a huge benefit of it being a commodity).
> They have more important things to do.
Poe's Law aside, I do actually find this to be a valid argument, if it's presented eyes-open, rather than just through inertia.
That is, there's a difference between having considered the migration and deciding it's not worth it (or that the "more important things" are more worth it and take priority) and not having considered the migration at all because those "more important things" completely consumed everyone's attention. I don't mean to present these as a false dichotomy, just two extremes that illustrate my point.
I've barely been in the industry for 90 days and I've already had to deal with the lack of the above in RHEL6.
(Indirectly, mind you; but I believe that the transition between RHEL6 and CentOS6 is negligible for the current topic.)
That said, I also hate working on RHEL 6.
... unless the software's needs evolve over time such as to require them.
> you know if your gray box ships with a 10gbe card or not
Yes; and lucky we do know. Else, our upcoming model of gray boxes that - unlike their predecessors - aren't approaching EOL but are only available with 10gbe cards ... would already be shipping with broken networking.
> all Python packages
That's a weakness in the language. You trade such a simple syntax for dependency hell. Oh, you wanted XML-derp-3.5.2.9.6? Well that requires XML-slurp-2.6.5.9.77 and you only have '75, sorry bud, can't possibly work. Oh but if you upgrade XML-slurp that breaks pycairo-2.7.55-1 and all your GUI code will stop working. Fun stuff.
(An IT illustration: there are still places running RHEL6 in production and not seeing an effort to migrate from it worth the while. They have more important things to do.)