benedekaibas wrote:
> I think we might want to take a step back and discuss at a higher level what
> we want to do here. It is possible that I am misunderstanding what this PR is
> trying to achieve.
>
> Consider the following example:
>
> ```
> struct S {
> int *owned;
> int *view;
> S();
> S(const S&);
> ~S();
> };
>
> S getS(S input [[clang::lifetimebound]]);
> ```
>
> Here, imagine `S` has both owned data, and data that refers to some other
> unowned storage. The `lifetimebound` here expressed that the output `S`
> depends on whatever the input `S` depended on. If we recursively do the
> binding for all fields, we might end up introducing a false dependency on
> `owned`. So I think the right thing to do in this case is to consult what the
> input was depending on and propagate the same dependence.
I see your point and I agree that this can introduce many FPs. The prupose of
this PR was just to introduce aggregate value tracking and then incrementally
improve it in further PRs.
https://github.com/llvm/llvm-project/pull/214823
_______________________________________________
cfe-commits mailing list
[email protected]
https://lists.llvm.org/cgi-bin/mailman/listinfo/cfe-commits
