add C API for epoch_deadline_callback and wasmtime_error_t creation

[theothergraham]

issue #6277

This adds C API interfaces to:

1. Set the epoch deadline callback to a C function
2. Create a wasmtime_error_t from C

[github-actions]

Subscribe to Label Action

cc @peterhuene

Details

This issue or pull request has been labeled: "wasmtime:c-api"

Thus the following users have been cc'd because of the following labels:

• peterhuene: wasmtime:c-api

To subscribe or unsubscribe from this label, edit the .github/subscribe-to-label.json configuration file.

Learn more.

[alexcrichton]

Looks reasonable to me, thanks!

[alexcrichton]

I think instead of :? this should use .to_str() on the pointer received to avoid Rust-level string escaping which shows up in the debug representation.

[theothergraham]

.to_str() leaves me with the potential to get a Utf8Error instead of a &str. I was not sure what the preferred way to deal with such bad input would be, but based on what I see in crates/c-api/src/trap.rs, I think I will go with String::from_utf8_lossy(). Sound good?

[alexcrichton]

Yeah I think the _lossy variant is fine. That's a reality of working with cross-language FFI here but in the context of "just an error message" the lossy variant should be fine.

[theothergraham]

Thanks, revised.

[kpreisser]

Hi! From my testing, it seems the new wasmtime_error_new function in the C API doesn't copy the string passed by the const char* argument when being called; instead the supplied char pointer is dereferenced later after wasmtime_error_new has already returned, e.g. when returning the wasmtime_error_t* in the epoch deadline callback. Is this intended?

I would have expected a similar behavior as wasmtime_trap_new, which copies the string when being called, so that the memory holding the string can be freed by the caller after the function returns. Otherwise, it's not quite clear to me when the memory can be freed.

Thanks!

[alexcrichton]

Ah thanks for testing! That's a mistake in the API and a subtle detail with the implementation. It's definitely not intended that the pointer should remain live, the contents should be copied into the error.

The line that needs to change is this one, namely the anyhow!(msg_string) should become anyhow!("{msg_string}"). @theothergraham or @kpreisser would one of y'all be up for a PR along those lines?

[theothergraham]

🤦 Sorry, that's my Rust newbieness showing. I missed that String:: from_utf8_lossy() only makes a copy if it finds invalid uft-8. I'll make a PR. Thank you for catching that, @kpreisser!