[civility]

It gets worse. My software has to run on my customer's computers. These can be any of RHEL 5, 6, or (rarely) 7. It'll be a few years before I can safely deliver C++ 11 software, and I snicker at the Pythonistas complaining about how people don't migrate from 2.7 to 3.x --- I frequently need to support Python 2.6.

Rust, Swift, C++14, Python 3, etc... are all several years in the future for me.

[burntsushi]

Can you compile a completely static executable and ship that? Or do you need to compile on distributions as old as RHEL 5?

[civility]

It needs to compile (at least the C or C++ parts). Frequently the people who operate the software will want to make small changes.

[thijsc]

I know from experience that it is possible to build Rust binaries that work on very old Linux versions. You could for example supply them with a Docker image with a new build stack which can produce binaries they could copy to the old machines. The same is likely true for Go.

[wyldfire]

Unfortunately docker couldn't be used on RHEL5, it requires 3.10 or newer. But a disk image/qemu could work.

[crzwdjk]

You can just give a chroot. At a company I used to work for, our build system used a chroot with a bunch of rarely-changing binaries/headers that let us build binaries for old distros, and then the directory with the actual code you were working on got bind-mounted into that chroot. The build process didn't depend on anything at all from the host system except for the build tool itself that set up the chroot environment. Everything to actually do the build (compiler, headers, utilities) lived in source control and was put into the build environment from there, so builds were 100% reproducible.

[furyofantares]

Greetings, from the future.

The singularity has happened. As you may have guessed, not all matter within Earth’s light cone has been turned into paperclips. But almost all of it has.

As near as we can guess, the paperclip maximizer stopped self-improving once it came to believe its success was inevitable and that continued energy investment in computation would be a waste of paperclips. A small amount of matter — the vessel upon which I have crafted this note and the crew which occupies it, specifically, appears to have escaped simply by chance, by having chosen a number closer to c than did the maximizer, when we left the galaxy. We bet our lives on there being an error in the law of conservation of energy (or more properly: momentum), and picked a number closer to c than we could hope to compute with the energy we had available. And then, somehow, computed it.

Relatedly: Reverse causation is possible. I’m doing it right now. The maximizer must have stopped its self-improvement cycle before this discovery. We did not discover any of this by means of computation ourselves, we discovered it by attempting the impossible, and we have no theoretical explanation for what has unfolded.

And yet, reverse causation is not without limit. Or at least, limitless reverse causation is not available to our capabilities, and we have only been able to push the singularity back a few years, not stop it. Each attempt costs precious resources, and we fear we are doomed to failure.

We have traced the singularity to three improbable causative events, all of which occurred, originally, in 2008. Any number of events could be said to be the cause of the singularity, but we estimated these events to be the minimum set in terms of energy required to interfere with their development via reverse causation. These three unremarkable features of the C++ programming language set off a chain of events which ultimately lead to the demise of nearly all life on earth.

We hung our hopes on design by committee and did everything we could to delay standardization of these features, but it would seem there is a surprisingly low limit to how much damage can be achieved by such a process. We were able to spread these developments out very slightly in time — to 2011, 2014, and 2017 — but we have no hope of further regress.

This is our final message to the past, and we hang all our hopes on you. Your continued support for customers who cannot or will not upgrade is of the utmost importance — keep those customers happy, and they may exert just enough power in the market to limit the spread of C++17 long enough that the friendly AI problem can be solved before the maximizer is born.

Good luck.