3 ms·

There are plenty of optimizations in Linux, they just pick ones that don't break userspace. Similarly compiler writers could do optimizations without breaking

by unnah 5y ago

There are plenty of optimizations in Linux, they just pick ones that don't break userspace. Similarly compiler writers could do optimizations without breaking user programs. Although I guess testing compiler backward compatibility would probably be more difficult than kernel backward compatibility, since compilers present a larger interface (multiple entire programming languages).

UncleMeat 5y ago
> Similarly compiler writers could do optimizations without breaking user programs. This isn't possible, since "breaking user programs" in this context means "violating the intuition of the programmer", who is an unknown person who has never written their intuition down to be observed by the compiler authors. If you are actually talking about breaking user programs, both C and C++ already have ABI stability as a design goal, which hugely limits changes to the languages but enables people to keep linking against binaries that were compiled ages ago.
- unnah 5y ago
  GCC and clang developers already do regularly compile and test a limited number of open source packages, and investigate any newly occurring failures. For now they ignore failures if the code does not conform to standards (except for notifying upstream developers). In principle, in such cases they could expend effort to find similar (possibly simpler) optimizations that do not change the existing behavior of the code. Alternatively, they could disable the relevant optimizations at -O2 (and maybe -O3) level. I believe MSVC developers do put more effort than GCC/Clang developers to maintain source-level backwards compatibility. Alternative development strategies are certainly possible, but I understand that the GCC/Clang developers do face pressure from a significant subset of users to maximize performance at any cost.
  - josephcsible 5y ago
    > In principle, in such cases they could expend effort to find similar (possibly simpler) optimizations that do not change the existing behavior of the code. This wouldn't just be expending effort. It would be making all code GCC/Clang ever compiled be slower. > Alternatively, they could disable the relevant optimizations at -O2 (and maybe -O3) level. Ditto.

by unnah 5y ago

UncleMeat 5y ago

> Similarly compiler writers could do optimizations without breaking user programs. This isn't possible, since "breaking user programs" in this context means "violating the intuition of the programmer", who is an unknown person who has never written their intuition down to be observed by the compiler authors. If you are actually talking about breaking user programs, both C and C++ already have ABI stability as a design goal, which hugely limits changes to the languages but enables people to keep linking against binaries that were compiled ages ago.

unnah 5y ago
GCC and clang developers already do regularly compile and test a limited number of open source packages, and investigate any newly occurring failures. For now they ignore failures if the code does not conform to standards (except for notifying upstream developers). In principle, in such cases they could expend effort to find similar (possibly simpler) optimizations that do not change the existing behavior of the code. Alternatively, they could disable the relevant optimizations at -O2 (and maybe -O3) level. I believe MSVC developers do put more effort than GCC/Clang developers to maintain source-level backwards compatibility. Alternative development strategies are certainly possible, but I understand that the GCC/Clang developers do face pressure from a significant subset of users to maximize performance at any cost.
- josephcsible 5y ago
  > In principle, in such cases they could expend effort to find similar (possibly simpler) optimizations that do not change the existing behavior of the code. This wouldn't just be expending effort. It would be making all code GCC/Clang ever compiled be slower. > Alternatively, they could disable the relevant optimizations at -O2 (and maybe -O3) level. Ditto.