[llvm-dev] [RFC] Requiring explicit address space arguments for PointerType

[Nicolai Hähnle via llvm-dev]

Hi Alex,

I'd be very much in favor of this, thanks for pushing ;)

On Wed, Jul 22, 2020 at 1:55 PM Alexander Richardson via llvm-dev
<llvm-dev at lists.llvm.org> wrote:
> (5?) Depending on how we feel about out-of-tree consumers, remove the
> LLVM_DEFAULT_AS_PARAM(AddressSpace) macros and require the argument to
> always be present. Adding the argument is compatible with older LLVM
> versions so it should be possible to do so without version #ifdefs. I
> think we should probably reduce the amount of changes required by
> out-of-tree consumers and keep the optional default, but it would
> obviously be nice to eventually remove the #ifdefs from inside LLVM.

As you say, since consumers can be changed in a way where the new
consumer version still builds against older LLVM versions, I think
this is reasonable. It would be great to find a way where consumers
get deprecation warnings for some period of time, but I can't think of
an easy way to do that.

Cheers,
Nicolai


>
> Does this sound like a sensible approach? Should I also attempt to
> implement steps 4 and 5 or is 1-3 sufficient?
>
> I would be great if we could land this upstream as it will
> significantly reduce the maintenance burden for our CHERI LLVM and
> should also prevent AVR issues such as e.g.
> https://reviews.llvm.org/rG215dc2e203341f7d1edc4c4a191b048af4ace43d
>
> Thanks,
> Alex
>
> [1] CHERI is currently available for MIPS and RISC-V and will soon be
> available on the upcoming Arm Morello board
> (https://developer.arm.com/architectures/cpu-architecture/a-profile/morello).
> _______________________________________________
> LLVM Developers mailing list
> llvm-dev at lists.llvm.org
> https://lists.llvm.org/cgi-bin/mailman/listinfo/llvm-dev



-- 
Lerne, wie die Welt wirklich ist,
aber vergiss niemals, wie sie sein sollte.