Earlier quoted context omitted.
Drivers for virtualized hardware is one big one. Parallels tools have to be recompiled for pretty much every new kernel version because Linux refuses to stabilize the kernel API. Often I have to edit the drivers by hand (example: replacing "sem" to "lock" for kernel 5.8 because apparently driver compatibility isn't as important as catering to kernel devs who don't know what a semaphore is?). In case you don't think v…
> Parallels tools have to be recompiled for pretty much every new kernel version because Linux refuses to stabilize the kernel API. I don't think that's a fair comparison. Parallels comes with dkms module and while the process of installing involves compiling under the covers, it's just part of installation. It's not the late 90s "compile that driver yourself completely manually" situation. If Parallels doesn't work…
Both approaches have their upsides, but I've very rarely had drivers stop working between various windows release. Of course, there are exceptions to the rule here - Yes, I also have had some drivers stop working since Windows XP. But those are rare, and getting rarer as time goes on and MS locks the kernel up, preventing drivers from doing stupid things.
On Linux, the problem only exists for out-of-tree drivers, but those will break fairly frequently for all but the simplest drivers.
As I said in my sibling comment, I don't know how I feel about a stable ABI - I do fear it would make the situation worse with more out-of-tree/proprietary drivers. But it is worth keeping in mind that it does have its upsides.
Of course, MacOS being what it is, it chose the "worst of all worlds" scenario: Proprietary drivers with no ABI stability guarantee. It's not really worth discussing as a point of comparison, MacOS has the worst driver story from the lot, only saved by its intended minimal target hw support.