feat: upgrade to uniffi25

[dignifiedquire]

Upgrades to NordSecurtiy/[email protected]
Based on #13.

TODOs

• Async FFI Refactor: Updated Future FFI mozilla/uniffi-rs#1767
• Reuse upstreamed code for binding generation and config merging
• fix & update tests

[dignifiedquire]

this is ready for review, only the latest commit is new

[arg0d]

Looks good, I left a small nits, but one thing that stood out was the object.pointer == nil check. Could you provide more context why it was added?

[arg0d]

Do you have time the make the remaining changes? It looks like the only major issue is error handling in async code. We have a requirement to have 0.25 ready in one weeks time, so I'm also invested to have this finished, and could offer some help.

[dignifiedquire]

I don't see any changes.

should be removed now

[dignifiedquire]

Will you change this to reuse the sequence template, or do you still see possible optimizations here?

I tried to do it, but ran into a bunch of import issues, where templates were either missing or included twice, even though I used the include_once_check. So I would leave it for now.

[arg0d]

I tried to do it, but ran into a bunch of import issues, where templates where either missing or included twice, even though I used the include_once_check. So I would leave it for now.

You are right, looks like my "trick" is actually buggy. It doesn't work when UInt8 isn't used in definitions somewhere, and also clashes with Vec<u8> type. I will need to do a fix for C# aswell 😅

[kegsay]

This almost compiles for me, just missing #21 (comment) - having a struct X with a field called Error, where X meets the error interface, causes names to collide as the error interface means there is a function called Error() and a field called Error which isn't valid.

I can also confirm this makes async functions work correctly (previously on the 0.24 PR they would block indefinitely).

[kegsay]

However, using bleeding edge rust code which uses a newer commit of uniffi-rs seems to fail to link:

ld: Undefined symbols:
  _ffi_matrix_sdk_ffi_rust_future_continuation_callback_set, referenced from:
      __cgo_8643f67aa596_Cfunc_ffi_matrix_sdk_ffi_rust_future_continuation_callback_set in 000001.o

I think it's because of mozilla/uniffi-rs@25ba41d (after much git bisecting..). I think this is because that commit removes what was a global callback to be per-crate, and that is not being defined.

[arg0d]

About the Error variant issue, it sounds like a separate issue unrelated to 0.25 update.

[dignifiedquire]

However, using bleeding edge rust code which uses a newer commit of uniffi-rs seems to fail to link:

yeah bleeding edge won't work until we upgrade to the next released version, due to breaking changes under the hood

[arg0d]

Hmm, this is not nice. mozilla/uniffi-rs@25ba41d is tagged with 0.25.1, so it should be compatible with 0.25.

[arg0d]

False alarm. The commit you specified is from 0.25.1 branch, and 0.25.1 lib interops fine with 0.25.0 Go bindings. I think you meant mozilla/uniffi-rs@eb97592 from main, and I can confirm that indeed 0.25.0 Go bindings don't interop with that commit from main branch. But this is expected, Go bindings aren't supposed to interop correctly with an arbitrary commit in main branch. Users should only use release tags, and match compatible lib/binding uniffi versions.

[arg0d]

I'm proceeding to merge this into main. This has been open for quite a while, and I think its time to merge this. I'm also under pressure to get this ready for our internal use 😅

Huge thanks @dignifiedquire for your efforts on this!!