How does this relate to the address sanitizer? I believe ASan can catch all of the issues caught by _FORTIFY_SOURCE. Does the latter have fewer checks and better runtime performance? I'm guessing it is considered by most to be fine using _FORTIFY_SOURCE in production binaries but too much overhead for ASan?
_FORTIFY_SOURCE is a rather weak but fast mitigation for a narrow class of security issues. AddressSanitizer is a bug finding tool (2x CPU overhead!), for a different and much wider class of issues.
Given both are open source, neither of them is going to be able to leapfrog the other for long, at least in terms of core C/C++ language features, without the other copying and implementing something similar. Both have sufficient developers to keep up with the pace of development. What's more interesting, at least to me, is what impact this competition is having on the wider ecosystem. Is it leading to more features and better compilation than would be the case if there was just one compiler toolchain?
> Is it leading to more features and better compilation than would be the case if there was just one compiler toolchain?
I would argue that the pace of C++ compiler development has significantly increased since clang was introduced.
I would surmise that elimination of GCC or Clang -- in the way that Chrome has resulted in the elimination of more than one competing browser -- would be detrimental to C++ as a whole.
And therefore, I would say: yes, it's leading to more features and better compilation that if it was just one compiler toolchain.
However, because Clang, we have finally IDEs caughting up with Energize C++ and VA for C++ v4 ideas, a couple of decades after they shown what would be possible with C++ when one isn't constrained to UNIX mentality of development.
Neither of those are significant -- both can be forked if it becomes necessary. GCC has a long history of forks, and half the reason clang is becoming popular is because the downstream vendors didn't like GPL and wanted to make proprietary forks.
Not literal copying, no, since the two codebases differ substantially, but they can look at the code for what the other has done to see how they did it and they can re-implement it without fear of patent infringement claims.
It's bad for GCC, but good for C++ developers. GCC and MSVC will die and there will be only one compiler on all platforms (which is also the best one).
Just have a look at Chrome and Firefox. They already use Clang on all platforms. On Windows they are working on replacing link with lld. Android has deprecated GCC completely. In the future it's gonna be clang + lld + libc++ all the way. And yeah, GDB and WinDbg will also be replaced by LLDB.
This is the longest shot of what you mentioned. I can see a future where the llvm toolchain dominates but people still use GDB because of muscle memory. (Unless LLDB changes its commands to match GDB.)
Also GCC is guaranteed to live for at least 50 yrs because of freetards. GNU has die hard users and developers who would rather die than leave. LLVM has money, big money. It remains to be seen which matters more.
> GCC and MSVC will die and there will be only one compiler on all platforms
That sounds like bad for C++ developers: competition is what made clang as good as it is now, and GCC improved a lot since Clang's threat happened. Also, Clangs bugs would become the de facto standard (this is happening right now on the web with Chrome).
Competition is good, sure. But focusing the limited developing resources on one compiler, too. Without parallel universes it's hard to know what actually leads to a better compiler.
Remember that these are both Linux benchmarks; GCC on windows is a bit sketchy with MinGW. Things can usually be made to work, but Clang is the only real competitor to MSVC (unless you're counting Intel's crazy-expensive offering, which is not intended for general use).
I doubt that you can fault clang with this behavior, because clang has a proper constexpr detection in C also, gcc only on C++.
So clang detects much more optimization opportunities with constant arguments, making it about 20x faster on the typical memcpy sizeof,... It's even faster than the glibc HW specific vectorized assembler optimizations, because the compiler knows much more than glibc, e.g. alignment, size. asm is an optimization barrier.
This is probably the reason clang has a different bos check than gcc.
Clang has mostly caught up to GCC and suppressed it in many others ways. It's probably somewhat bad in the long run since LLVM/Clang seems to have more devs contributing to it now.