Antoine Lavandier (Oct 03 2025 at 08:34): Antoine Lavandier (Oct 03 2025 at 08:34):I have been wondering for a while why the bound on Store<T> is T: 'static. instead of having a Store<T, 'a> with the bound T: 'a instead. I don't understand why T has to outlive the Store that uses it. Is it because the store isn't capable to drop T when it is itself dropped ? Since T is taken by value at the store construction there isn't a way for that value to be used again once the store is dropped AFAIU...
Is the current because it was/is too complicated to propagate the lifetime constraints all over the machinery or is my understanding wrong ?
Alex Crichton (Oct 03 2025 at 15:28):Propagating a lifetime everywhere would be pretty complicated yeah, but it also has to do with the component-model-async implementation where it, to us at least, unnecesarily requires that T: 'static when otherwise we shouldn't need it
Pat Hickey (Oct 03 2025 at 15:28):Design rationale is here https://github.com/bytecodealliance/wasmtime/pull/10760
Pat Hickey (Oct 03 2025 at 15:28):ah, you beat me to it
Alex Crichton (Oct 03 2025 at 15:29):oh no that PR is better heh
Alex Crichton (Oct 03 2025 at 15:29):I keep forgetting about that...
Antoine Lavandier (Oct 03 2025 at 15:52):I still don't understand why it's needed in the first place. From that PR the only thing I got is that at some point (or maybe from the start of wasmtime) the compiler started asking for T: 'static bound and you tried to resist it by adding some transmuteto alleviate this but it ended up overwhelming so you decided instead to put T: 'static bounds everywhere (which is completely fair) but I don't understand the underlying reason that caused any T: 'static bound to show up in the first place
Lann Martin (Oct 03 2025 at 15:59):We added this constraint to Spin's embedding of Wasmtime long ago. One of the biggest problems for async is that futures need to be 'static in a lot of situations like for tokio::task::spawn.
Alex Crichton (Oct 03 2025 at 16:10):The main cause of the compiler wanting this bound is that we store, for example, fn(T) in data structures where T is a generic parameter. That data structure isn't 'static unless T: 'static
Alex Crichton (Oct 03 2025 at 16:11):(according to rustc, which we feel is a bug but can't do much about it)
Chris Fallin (Oct 03 2025 at 16:19):Yeah, the rust-lang folks point to the well-formedness check on the type (any contained type must outlive the type) which makes sense when the value of that type is actually stored inside but a function arg is not that. I suppose the most one could push things if the restriction didn't exist is that a fn(T) of lifetime 'static could live beyond when any valid T exists that would fit, which is odd (uninhabitable callsite param) but not unsound. shrug
Chris Fallin (Oct 03 2025 at 16:21):(of course I could be missing something subtle here but my point is that at least it doesn't seem like an obvious "type system insists, sorry end of story")
Last updated: Dec 06 2025 at 06:05 UTC