On Wed, Aug 14, 2024 at 05:03:21PM +0200, Miguel Ojeda wrote: > On Wed, Aug 14, 2024 at 4:00 PM Danilo Krummrich <dakr@xxxxxxxxxx> wrote: > > > > If we keep them, we'd consequently also need to add them for vrealloc() and > > kvrealloc(). But again, they don't do anything for us, and hence are more > > misleading than helpful IMO. > > In general, they could do something (e.g. `noreturn`), perhaps in the future. Indeed, and I think once they're honored we should add them again. It's just that I think as long as compiler attributes aren't honored, we should not have them in the first place to avoid confusion about whether they do or do not have any effect. > > Apart from being potentially misleading, do we gain something by > removing them? It's not so much that I want to remove them for krealloc(), it's that I don't want to intentionally add them for the vrealloc() and kvrealloc() helpers, knowing that they don't do anything (yet). And I think it's even more confusing if the krealloc() helper has those compiler attributes, but the vrealloc() and kvrealloc() helpers do not. > I guess simplicity in the file, but it is also simpler > to keep them aligned to the C side (which I guess is Alice's point), > and avoids having to keep track of what could have a present or future > impact in `bindgen`. > > Cheers, > Miguel >
