Gregory's Server
 42 comments
@isagalaev hm… maybe ‘unreachable’ is clang hardcoded function name that will be called in case of infinity loop is happened? 🤨  @dev The situation is so bad that AFAIK even though LLVM has improved in this area, if you want to compile a language (unlike C++ or C prior to C11) that has well-defined infinite loops on top of LLVM, you pretty much still have to use a dummy asm block to prevent this kind of insanity.  @pervognsen @dev Yes, Rust had a problem with this: https://blog.rust-lang.org/inside-rust/2020/03/19/terminating-rust.html @dev oh I see you have a type error that for some reason the compiler is ignoring. Try -Werror=return-type ? @dev Oh - I guess that’s not firing because the *compiler* looks at the control flow and says “you don’t actually reach the closing brace” and then something is happening lower down in LLVM to delete the loop? @elfprince13 working theory is the since the ub loop is optimized out, somehow it keeps going beyond main  @dev @elfprince13 Yeah, you can take a look in Compiler Explorer and see that that is indeed the case. main is just empty, so it falls through to unreachable. @dev if you don’t write your infinite loops as the Collatz Conjecture that’s on you ;) but it doesn’t seem like the compiler should be allowed to treat the loop as infinite for the purposes of control flow analysis to suppress a warning about missing “return 0”, but then decide the behavior is undefined and can be finite. @elfprince13 @dev As of C99 and C++ it is optional to have a return statement in main(). If control reaches the end of main() with no explicit return, it returns 0. @elfprince13 @dev ah, but is the collatz conjecture good enough when c compilers are smart enough to disprove fermat's last theorem? @elfprince13 @dev oh i see you've seen this already nevertheless, a shining testament to the field of computer science. those mathematicians can't even disprove fermat's last theorem, smh  @dev If main never calls unreachable, how does that code get executed? Is it something weird like falling through from main without a return, and then running the following code? Hmm I saw some of the chatter on Twitter from JF but don't know enough about C++ semantics  @jawnsy @dev basically in c++ infinite loops that do nothing are undefined behavior and LLVM assumes undefined behavior cannot occur, and just stops emitting code for main after the prolog.
This includes the function return so it literally just falls through the next code, and that happens to be unreachable() @dev Endless loops with no side effects is undefined behavior, Clang decided to be funny about it. @zombierustpunk oh it is just fish shell and I've replaced cat with bat, which does the code highlighting @dev I want to know what variant of  @dev @david @ionchy oh it might only be C++ too |
@dev but...why??