Re^5: The future of Perl?

[mr_mischief]

I think QNX would be a poor target for removal up front. It's a mostly POSIX-compatible system that runs on modern embedded hardware very well. ( http://www.qnx.com/products/neutrino-rtos/neutrino-rtos.html#POSIX ) If your desire is to make Perl more relevant, abandoning a system with decent although small market share on ARM Core and Intel Atom systems is probably not a good way to do that. Should the GNU-userland Linux systems, the BSD systems, OSX, Windows, and maybe Android be higher priority? Sure. But to get to all of those portably gets you a pretty long way toward QNX. If support for it fell off it may not be a catastrophe. Writing it off before any attempt to gauge the work necessary to keep it seems premature.

VMS may not be cutting edge, but it's still used in some important places.

I don't imagine it would be hard with a Windows version and a GNU userland installed on top of their OS to keep the eComStation die-hards interested in maintaining the OS/2 and eComStation portability. It's not a very big community, but it's still fairly active.

I think if you wanted a better list of systems to abandon for being irrelevant you could start with MS-DOS/PC-DOS, AmigaOS, Haiku, Mac OS Classic, and BS2000. I almost included RISC OS. Their community seems pretty headstrong and there's still work going into the OS.

[BrowserUk]

See points 1 through 3.

[mr_mischief]

++ and I'd already read that. My point is less that I'd have picked differently and more that how much effort a minor platform takes is perhaps as important as how many users it has.

Starting with a cleaner, simpler, more easily managed core could make managing ports around the edges of the code simpler. Starting with an example POSIX machine (and sticking to mostly POSIX), be that GNU/Linux, FreeBSD, Darwin, or whatever and then adding thin, isolated compatibility layers around the file handling and signals on not-so-POSIX systems goes a long way toward portability. One or two platforms is plenty for a first model, but choosing not to support low-hanging fruit beyond that just because the fruit isn't very popular seems silly. Now, if there's a platform that absolutely nobody will step up to support, that's an issue for that platform.

[BrowserUk]

One or two platforms is plenty for a first model,

That is exactly the point. Reduce, reduce, reduce. Refactor & test; refactor & test. (Minimal platforms for tests means least hysteresis for testing.)

Then, if the project achieves anything, and if there are sufficient people willing to contribute to a given platform, then add it back.

Remember, nothing changes in the existing P5 line -- unless p5p decide to change it.

Less platforms at the start means less people required; that means it is easier to achieve agreement at each step; less chance of getting held up by one platform.

One of my steps would also be 64-bit only.