llvm dev - Feb 2015 - [LLVMdev] Semantics of __builtin_setjmp/__builtin_longjmp

Hi all,
currently we lower the __builtin_setjmp/__builtin_longjmp intrinsincs to
SjLj exception style intrinsics. For ARM on non-Darwin, they end up as
nops, breaking Ruby. This leads to the question of what the defined
semantics of them is.

(a) Are they supposed to be compatible with libc? PPC disagrees in
comments.

(b) Are they supposed to be compatible with GCC? If not with libc, why
with GCC?

Joerg

> (b) Are they supposed to be compatible with GCC?
I think they're probably not supposed to be compatible with libc, but
should be with GCC.

Libc gets to choose its implementation in setjmp.h. It might implement
them completely separately (which would still work however we
implement the __builtins), or it might make use of the __builtins (in
which case we should certainly make sure they're compatible).
> If not with libc, why with GCC?
Because the GCC builtins are an ABI issue controlled by the compiler:
we need to be compatible to mix object files on the platform.

Cheers.

Tim.

Correct on all counts.

-Jim
> On Feb 10, 2015, at 1:16 PM, Tim Northover <t.p.northover at
gmail.com> wrote:
> 
>> (b) Are they supposed to be compatible with GCC?
> 
> I think they're probably not supposed to be compatible with libc, but
> should be with GCC.
> 
> Libc gets to choose its implementation in setjmp.h. It might implement
> them completely separately (which would still work however we
> implement the __builtins), or it might make use of the __builtins (in
> which case we should certainly make sure they're compatible).
> 
>> If not with libc, why with GCC?
> 
> Because the GCC builtins are an ABI issue controlled by the compiler:
> we need to be compatible to mix object files on the platform.
> 
> Cheers.
> 
> Tim.
> _______________________________________________
> LLVM Developers mailing list
> LLVMdev at cs.uiuc.edu         http://llvm.cs.uiuc.edu
> http://lists.cs.uiuc.edu/mailman/listinfo/llvmdev