chore(ffi): use ruma's built-in uniffi support to eliminate some event type enums by Johennes · Pull Request #6161 · matrix-org/matrix-rust-sdk

Johennes · 2026-02-11T13:10:00Z

This is the smallest possible way that I could find to start using ruma/ruma#2338 in the FFI bindings. It eliminates the types StateEventType, MessageLikeEventType and RoomAccountDataEventType from the FFI crate and instead uses the Ruma types directly .

The only difference of the two types in the foreign language should be the use of Custom vs. Other.

Public API changes documented in changelogs (optional)

codecov · 2026-02-11T13:28:41Z

Codecov Report

✅ All modified and coverable lines are covered by tests.
✅ Project coverage is 89.89%. Comparing base (7b2b439) to head (bdb87ce).
⚠️ Report is 5 commits behind head on main.
✅ All tests successful. No failed tests found.

Additional details and impacted files

@@            Coverage Diff             @@
##             main    #6161      +/-   ##
==========================================
+ Coverage   89.86%   89.89%   +0.02%     
==========================================
  Files         396      396              
  Lines      110260   110260              
  Branches   110260   110260              
==========================================
+ Hits        99088    99113      +25     
+ Misses       7395     7378      -17     
+ Partials     3777     3769       -8

☔ View full report in Codecov by Harness.
📢 Have feedback on the report? Share it here.

codspeed-hq · 2026-02-11T13:32:37Z

Merging this PR will not alter performance

✅ 50 untouched benchmarks

_{Comparing Johennes:johannes/uniffi (bdb87ce) with main (e8673e4)}

Johennes · 2026-02-12T14:31:47Z

This is failing the complement-crypto tests now. These downgrade uniffi to 0.28.3. This no longer works because ruma's new uniffi support depends on changes from uniffi 0.31.0.

In particular:

doc(hidden) doesn't work in 0.28.3
uniffi::export only works on functions and impl blocks in 0.28.3 (ruma uses uniffi::export(Display, Eq, Hash) on an enum)
In 0.28.3 custom_type! only takes two type parameters and then you need to implement UniffiCustomTypeConverter. In 0.31.0 the macro takes the converters directly in a third argument

I can think of the following ways to resolve this:

Split the uniffi implementation in ruma into two features, one that works with 0.28.3 and one that uses 0.31.0
Change the rebuild_rust_sdk.sh script from complement to hack the changes into a local ruma checkout

I'm not sure which of these is worse but maybe I've missed a better option?

jplatte · 2026-02-12T19:38:08Z

The better option is to help move NordSecurity/uniffi-bindgen-go#86 forward, IMHO.

Johennes · 2026-02-13T09:37:17Z

The better option is to help move NordSecurity/uniffi-bindgen-go#86 forward, IMHO.

That's still two minor versions behind what ruma needs but, yes, ideally the complement crypto tests wouldn't have to downgrade uniffi in the first place.

jplatte · 2026-02-13T11:39:25Z

Yeah, but it would unblock additional upgrades (which may be much easier than that one, if you're lucky ^^).

poljar · 2026-04-02T13:17:03Z

This should be unblocked as matrix-org/complement-crypto#233 is now merged.

Johennes · 2026-05-27T05:58:27Z

Looks like this still fails the complement-crypto tests. Will have to investigate what's wrong there.

 Error: Failed to extract data from archive member `ruma_events-2c1d40fe02f50a4b.ruma_events.21d0e9ca7cef90a2-cgu.09.rcgu.o`

Caused by:
    0: extracting metadata for 'UNIFFI_META_RUMA_EVENTS_CONSTRUCTOR_EPHEMERALROOMEVENTTYPE_FROM_STRING'
    1: Constructor return type must be Self or Arc<Self>

Hywan

Thank you for your contributions!

Hywan · 2026-05-28T07:34:04Z

  complement-crypto:
    name: "Run Complement Crypto tests"
-    uses: matrix-org/complement-crypto/.github/workflows/single_sdk_tests.yml@399a1deeab0d7e4fa9604cbe83b1df6058c40193 # main
+    uses: matrix-org/complement-crypto/.github/workflows/single_sdk_tests.yml@1b1f86348079608fbc5276fe77c54dd66638a622 # main


Why this change?

This is the commit that updated complement-crypto to uniffi 0.31.0 (which is the version the uniffi support in Ruma requires) and it is slightly ahead of the currently used SHA: https://github.com/matrix-org/complement-crypto/commits/main/

Johennes · 2026-05-28T18:10:03Z

Looks like this still fails the complement-crypto tests. Will have to investigate what's wrong there.

 Error: Failed to extract data from archive member ``ruma_events-2c1d40fe02f50a4b.ruma_events.21d0e9ca7cef90a2-cgu.09.rcgu.o``

Caused by:
    0: extracting metadata for 'UNIFFI_META_RUMA_EVENTS_CONSTRUCTOR_EPHEMERALROOMEVENTTYPE_FROM_STRING'
    1: Constructor return type must be Self or Arc<Self>

The error appears to refer to this method. But confusingly that does already return Self. 🤷‍♂️

Johennes · 2026-05-28T18:17:50Z

Sigh. I think #[uniffi::constructor] may only work on objects but not on enumerations. So we'll probably need to fix this in Ruma.

Johennes · 2026-06-09T06:57:01Z

The cargo-deny error seems to also exist on main. Apart from that, I believe this should finally be ready for review now.

andybalaam

This looks reasonable to me, without much context on the complement-crypto changes etc. Waiting for Ivan to comment, and asked one question about what I think is a merge mistake.

andybalaam

Good with me but I think there are unresolved conversations with other reviewers.

Johennes · 2026-06-10T10:58:59Z

Sorry, I'm now realizing the bindings tests have uncovered a few more name clashes such as:

/Users/runner/work/matrix-rust-sdk/matrix-rust-sdk/bindings/apple/generated/swift/matrix_sdk_ffi.swift:40996:30: error: 'TimelineEventType' is ambiguous for type lookup in this context

I'm not sure why these didn't show up before. I think this will require renaming the exported FFI type names or trying to eliminate the FFI types in favour of the Ruma types.

Johennes · 2026-06-11T06:26:30Z

Moving back to draft as this seems to require still further work. 😕

Johennes · 2026-06-20T12:58:53Z

 #[derive(Clone, uniffi::Enum, PartialEq, Eq, Hash)]
 #[uniffi::export(Eq, Hash)]
-pub enum TimelineEventType {
+pub enum FfiTimelineEventType {


This required renaming because it caused a name-clash with the identically named Ruma type which is now also exported. Ruma's type is flat. So this wouldn't have been a drop-in replacement.

I tried renaming just the exported type name but that broke auto-generating the Eq and Hash implementations.

I tried renaming just the exported type name but that broke auto-generating the Eq and Hash implementations.

Not sure what this means exactly, but sounds like an upstream bug? Did you file an issue already?

I first tried adding #[uniffi::export(name = "FfiTimelineEventType")] but that then failed when generating bindings claiming that it cannot locate TimelineEventType. I have a hunch this might be because of the exported Eq and Hash implementations but haven't verified that. Reporting upstream sounds sensible assuming we can reproduce it in a minimal example.

Tried to reproduce it in a minimal example but couldn't. So I may have hit something else or maybe it's already been fixed.

Johennes · 2026-06-20T19:33:31Z

It looks like this is finally ready. The remaining lint error also exists on main.

poljar

Looks good to me. We should just mention the rename of the exported type in the changelog as well.

poljar · 2026-06-24T11:57:09Z

 #[derive(Clone, uniffi::Enum, PartialEq, Eq, Hash)]
 #[uniffi::export(Eq, Hash)]
-pub enum TimelineEventType {
+pub enum FfiTimelineEventType {


That's a bit of an ugly name, hopefully the users of the bindings won't mind too much.

The more important bit here is that this is a breaking change though. We should mention this in the changelog.

Yeah, it's not pretty. Eventually this should just be replaced with the ruma type. That one has a different structure though. So doing that would have been a bigger breaking change than the renaming.

Maybe CC @stefanceriu for visibility on this.

And @jmartinesp as well.

Beware there's an annoying rename in this PR.

Argh, sorry for forgetting the Android side. I'm lacking coffee. 🙈 ☕

…t type enums Signed-off-by: Johannes Marbach <n0-0ne+github@mailbox.org>

jmartinesp · 2026-06-24T14:26:20Z

I feel like I've said this quite a few times, but next time we do something breaking in the FFI side, could we have some time to test it before merging something 😅 ?

This broke a few unit tests we have on Android, since now StateEventType and MessageLikeEvenType have one case which is an Object (Custom) and that means it'll automatically try to load the native library (which isn't even built for the host's architecture) and the test won't even start. It's not like they're important tests so we can just remove them, but it would have been better to know about this in advance.

poljar · 2026-06-24T14:34:06Z

This broke a few unit tests we have on Android, since now StateEventType and MessageLikeEvenType have one case which is an Object (Custom) and that means it'll automatically try to load the native library (which isn't even built for the host's architecture) and the test won't even start. It's not like they're important tests so we can just remove them, but it would have been better to know about this in advance.

Oh, I thought about requesting a real life test for it but underestimated how much this changed things.

Sorry.

If it's problematic we can revert this and reaply later, as it's a single commit.

jmartinesp · 2026-06-24T14:35:54Z

No, I think it's fine, it's just a few unit tests we'll disable.

Johennes · 2026-06-24T15:54:57Z

Apologies. I probably also should have pinged you about this much earlier than today. Sorry for the extra hassle.

Johennes force-pushed the johannes/uniffi branch 4 times, most recently from e0e9256 to 5acec1d Compare February 11, 2026 13:16

Johennes mentioned this pull request Feb 12, 2026

common: Add hidden macro to generate PrivOwnedStr structs ruma/ruma#2361

Merged

Johennes mentioned this pull request Feb 12, 2026

uniffi attributes ruma/ruma#2070

Open

Johennes force-pushed the johannes/uniffi branch 6 times, most recently from 3ae18f2 to 41aaf22 Compare May 26, 2026 18:20

Hywan reviewed May 28, 2026

View reviewed changes

Johennes mentioned this pull request May 28, 2026

events: remove #[uniffi::constructor] as it doesn't appear to work on enums ruma/ruma#2501

Merged

Johennes force-pushed the johannes/uniffi branch from 41aaf22 to 3cf16bb Compare June 1, 2026 18:38

Johennes commented Jun 1, 2026

View reviewed changes

Comment thread Cargo.toml Outdated

Johennes force-pushed the johannes/uniffi branch 4 times, most recently from f5efb05 to c8e7f7a Compare June 1, 2026 18:52

Johennes requested a review from andybalaam June 9, 2026 06:56

Hywan self-requested a review June 9, 2026 08:45

andybalaam reviewed Jun 9, 2026

View reviewed changes

Comment thread .deny.toml

Johennes force-pushed the johannes/uniffi branch from a5c9a26 to 9353167 Compare June 9, 2026 17:22

andybalaam approved these changes Jun 10, 2026

View reviewed changes

Johennes force-pushed the johannes/uniffi branch from 9353167 to b95e1d4 Compare June 10, 2026 19:09

Johennes marked this pull request as draft June 11, 2026 06:26

Johennes force-pushed the johannes/uniffi branch from b95e1d4 to 6dd33bd Compare June 20, 2026 12:35

Johennes marked this pull request as ready for review June 20, 2026 12:41

Johennes force-pushed the johannes/uniffi branch 2 times, most recently from 23d068d to a496a7a Compare June 20, 2026 12:57

Johennes commented Jun 20, 2026

View reviewed changes

Johennes force-pushed the johannes/uniffi branch 2 times, most recently from 89765ef to 74c3f60 Compare June 20, 2026 13:12

Johennes changed the title ~~chore(ffi): use ruma's built-in uniffi support to eliminate StateEventType and MessageLikeEventType~~ chore(ffi): use ruma's built-in uniffi support to eliminate some event type enums Jun 20, 2026

Johennes force-pushed the johannes/uniffi branch from 74c3f60 to 521bfb0 Compare June 20, 2026 19:01

poljar requested changes Jun 24, 2026

View reviewed changes

chore(ffi): use ruma's built-in uniffi support to eliminate some even…

bdb87ce

…t type enums Signed-off-by: Johannes Marbach <n0-0ne+github@mailbox.org>

Johennes force-pushed the johannes/uniffi branch from 521bfb0 to bdb87ce Compare June 24, 2026 12:06

Johennes requested a review from poljar June 24, 2026 12:08

poljar approved these changes Jun 24, 2026

View reviewed changes

poljar merged commit d895344 into matrix-org:main Jun 24, 2026
54 of 55 checks passed

Uh oh!

Conversation

Johennes commented Feb 11, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

codecov Bot commented Feb 11, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Codecov Report

Uh oh!

codspeed-hq Bot commented Feb 11, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Merging this PR will not alter performance

Uh oh!

Johennes commented Feb 12, 2026

Uh oh!

jplatte commented Feb 12, 2026

Uh oh!

Johennes commented Feb 13, 2026

Uh oh!

jplatte commented Feb 13, 2026

Uh oh!

poljar commented Apr 2, 2026

Uh oh!

Johennes commented May 27, 2026

Uh oh!

Hywan left a comment

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Johennes commented May 28, 2026

Uh oh!

Johennes commented May 28, 2026

Uh oh!

Uh oh!

Johennes commented Jun 9, 2026

Uh oh!

andybalaam left a comment

Choose a reason for hiding this comment

Uh oh!

Uh oh!

andybalaam left a comment

Choose a reason for hiding this comment

Uh oh!

Johennes commented Jun 10, 2026

Uh oh!

Johennes commented Jun 11, 2026

Uh oh!

Johennes Jun 20, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Johennes commented Jun 20, 2026

Uh oh!

poljar left a comment

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Uh oh!

jmartinesp commented Jun 24, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

poljar commented Jun 24, 2026

Uh oh!

jmartinesp commented Jun 24, 2026

Johennes commented Feb 11, 2026 •

edited

Loading

codecov Bot commented Feb 11, 2026 •

edited

Loading

codspeed-hq Bot commented Feb 11, 2026 •

edited

Loading

Johennes Jun 20, 2026 •

edited

Loading

jmartinesp commented Jun 24, 2026 •

edited

Loading