2 ms·

And without backwards compatibility you would have never heard of Linux.

geofft 9y ago
That's a very strong claim. Some evidence in favor is that, at least in the early days of Linux, Microsoft took the same strategy of prioritizing backwards compatibility over security - and reaped the rewards by becoming extremely popular and extremely full of security holes. So clearly the strategy worked for MS. On the other hand, MS did respond and prioritize security, and was able to pull it off without compromising backwards compatibility too much. (For instance, last week's stack-clash vulnerability straight up doesn't exist on Windows because MSVC and the NT kernel have been doing the right thing with stack probes for years.) But some real evidence against is that this whole backwards-compatibility thing is a kernel policy, not a userspace policy; no distro cares nearly as much. With the a.out to ELF transition and libc5 to libc6 transition back in the day, and to this day with OpenSSL versions, the GCC 5 libstdc++ ABI change, etc., there's not a ton of backwards compatibility in what binaries you can actually run on a real-world Linux system. It seems hard to believe that the kernel-to-userspace compatibility story is what made Linux popular, given the vast amount of userspace-to-other-userspace incompatibility.
- hacknat 9y ago
  I don't think it's "very" strong. It might just be "strong". Rehashing MS's 90s-00s history of prioritizing security creates an unfair assumed comparison. Linux doesn't get to control userland the way MS does. I don't want to belittle MS's efforts, but the attack vector is a lot smaller in NT. Linux has way more features and use cases than the NT kernel ever has (maybe by an order of magnitude). We also don't have the complete picture on NT because of the source being closed. > With the a.out to ELF transition and libc5 to libc6 transition back in the day, and to this day with OpenSSL versions, the GCC 5 libstdc++ ABI change, etc You need to remember that Linux's use cases are way bigger than being able to build C binaries and stay forward with SSL. It's easy to forget, but Linux is hardly just servers, they are probably the biggest embedded foot print outside the no-OS or RTOS space, tons non-PC peripheral and consumer electronic applications. You're calling out one set of features that a huge swath of Linux consumers probably never touched for a decade (remember its only been recently that embedded applications have communicated over a network, or had to do so securely).

That's a very strong claim. Some evidence in favor is that, at least in the early days of Linux, Microsoft took the same strategy of prioritizing backwards compatibility over security - and reaped the rewards by becoming extremely popular and extremely full of security holes. So clearly the strategy worked for MS. On the other hand, MS did respond and prioritize security, and was able to pull it off without compromising backwards compatibility too much. (For instance, last week's stack-clash vulnerability straight up doesn't exist on Windows because MSVC and the NT kernel have been doing the right thing with stack probes for years.) But some real evidence against is that this whole backwards-compatibility thing is a kernel policy, not a userspace policy; no distro cares nearly as much. With the a.out to ELF transition and libc5 to libc6 transition back in the day, and to this day with OpenSSL versions, the GCC 5 libstdc++ ABI change, etc., there's not a ton of backwards compatibility in what binaries you can actually run on a real-world Linux system. It seems hard to believe that the kernel-to-userspace compatibility story is what made Linux popular, given the vast amount of userspace-to-other-userspace incompatibility.

hacknat 9y ago
I don't think it's "very" strong. It might just be "strong". Rehashing MS's 90s-00s history of prioritizing security creates an unfair assumed comparison. Linux doesn't get to control userland the way MS does. I don't want to belittle MS's efforts, but the attack vector is a lot smaller in NT. Linux has way more features and use cases than the NT kernel ever has (maybe by an order of magnitude). We also don't have the complete picture on NT because of the source being closed. > With the a.out to ELF transition and libc5 to libc6 transition back in the day, and to this day with OpenSSL versions, the GCC 5 libstdc++ ABI change, etc You need to remember that Linux's use cases are way bigger than being able to build C binaries and stay forward with SSL. It's easy to forget, but Linux is hardly just servers, they are probably the biggest embedded foot print outside the no-OS or RTOS space, tons non-PC peripheral and consumer electronic applications. You're calling out one set of features that a huge swath of Linux consumers probably never touched for a decade (remember its only been recently that embedded applications have communicated over a network, or had to do so securely).