Comment by bigfatkitten
11 hours ago
They exist but they’re in the minority. Compare this with the Linux world, where userspace compatibility between one major distro release and the next, 12 months apart is very much a roll of the dice.
11 hours ago
They exist but they’re in the minority. Compare this with the Linux world, where userspace compatibility between one major distro release and the next, 12 months apart is very much a roll of the dice.
This is true.
To be fair, if you use only the Linux syscall interface, then a program that you compiled on x86 in the 90s will probably still run anywhere today. Linus is adamant about this.
But if you want to use... anything else, then it's unlikely to work at all unless you are very specific about your target. There isn't one company deciding that glibc or mesa or whatever is binary backward compatible on every kernel for every platform forever. Microsoft is, somewhat, one such company. That's why System32\*.dll have such stable interfaces -- it's their job to translate whatever late 90s system/graphics facilities some boomer dreamed up into whatever the current Windows hodgepodge of system services support. It's no wonder Microsoft is trying to drop support for hardware like crazy.
This implicit compatibility isn't true for all Windows programs, though. Consider Visual Studio. Couldn't compile my console program on my computer and run it on my dad's computer. He had to first install the "Redistributable," which for him and most people might as well be a rootkit and a super scary virus program bad guy.
> To be fair, if you use only the Linux syscall interface, then a program that you compiled on x86 in the 90s will probably still run anywhere today. Linus is adamant about this.
This is frankly the opposite of how it should've been. Who cares that the kernel ABI doesn't change from release to release? As an end-user, I couldn't care less. Even as a developer I care very little indeed, because I'm not writing drivers. I'm almost never programming directly against the kernel's interface but rather using my language's standard library, which is already an abstraction over the kernel interface.
Platform vendors (this includes distro maintainers) should recompile and re-package libraries for each new ABI in each OS release, done. End-users won't ever notice, they can run their applications portably because the user-mode library ABIs haven't changed... Which is what Win32 has done, and what glibc has consistently failed to do.
The funny thing about the glibc situation is that glibc really doesn’t provide improvements to any of the kernel interfaces or really any comfort stuff for developers. If anything the fact that it is bound by the C standards makes it worse than using the kernel API directly besides that for whatever reason the kernel headers seem to lack a fair number of definitions. That could just be a me problem though.
I think Windows also discourages people to use their syscall directly. It could change from version to version so they should use the API.
The biggest problem with user space compatibility on Linux are binaries compiled and run on older releases, not future ones.