Port to matrix-rust-sdk-crypto

[TobiasFella]

No description provided.

[redstrate]

Initial review drop, will do more when I'm feeling better

[redstrate]

nit: Do these functions (and ditto) need to use clone? Clippy would tell you I guess.

[redstrate]

Do we want #863 in before this or can they be merged unrelated to each other?

[TobiasFella]

Do we want #863 in before this or can they be merged unrelated to each other?

I guess that doesn't matter too much, whichever goes in later must be adapted

[nvrWhere]

So currently when I try to verify my session with NeoChat it all starts fine but no emojis appear once both sides have started

[nvrWhere]

I've been using this for months and haven't noticed which in this case is great. It's worlds better than where we are now. I think we ship this.

[KitsuneRal]

Functionally, this looks very reasonable overall by now. Thanks for the huge work done! The only thing I discovered is actually in the build configuration. There's something off with the path to lib.rs.h because my initial attempt to build the PR failed because the compiler couldn't find the header (despite it having been generated obviously). Also, the fact that it relies on a very new Corrosion version that has to be manually installed is a bit unfortunate; but I really don't know if we should wait for it to land in distro repos or just commit the generated files instead... which might solve the lib.rs.h path problem along the way, too?

[TobiasFella]

Ugh, thought the header generation thing was fixed. Weird that it works fine on the CI, then... I'll have a look tomorrow.

WRT corrosion, i think we should just ship it like it is now. The best way to make a distro update a dependency is to just enforce it.

[TobiasFella]

I really don't know what's up with lib.rs.h... It works completely fine here. Is there anything you can say about when it is (and isn't) failing for you?

[KitsuneRal]

Right; I really should've given more than just a "doesn't build here" note. Buut, now that I tried to build it from clean slate, one more thing surfaced:

   Compiling matrix-sdk-sqlite v0.16.0
error[E0308]: mismatched types
    --> src/cryptomachine.rs:1887:37
     |
1887 | ...                   pickle: session.pickle(),
     |                               ^^^^^^^^^^^^^^^^ expected `SessionPickle`, found `vodozemac::olm::SessionPickle`
     |
note: there are multiple different versions of crate `vodozemac` in the dependency graph
    --> /var/home/axy/dev/.cargo/registry/src/index.crates.io-1949cf8c6b5b557f/vodozemac-0.9.0/src/olm/session/mod.rs:481:1
     |
 481 | pub struct SessionPickle {
     | ^^^^^^^^^^^^^^^^^^^^^^^^ this is the expected type
     |
    ::: /var/home/axy/dev/.cargo/registry/src/index.crates.io-1949cf8c6b5b557f/vodozemac-0.10.0/src/olm/session/mod.rs:510:1
     |
 510 | pub struct SessionPickle {
     | ------------------------ this is the found type
     = help: you can use `cargo tree` to explore your dependency tree

cargo tree shows that indeed, we get the older vodozemac with matrix-sdk crates but then also get vodozemac-0.10 via our own Cargo.toml. That was what I could see from absent ~/.cargo, no Cargo.lock, and with a freshly done CMake configuration. If I manually remove the new vodozemac from Cargo.lock then the build goes through (even without that former error about lib.rs.h I've got, which is also peculiar). In any case, that definitely needs fixing: we should only pull the version of vodozemac that other crates are using, not just anything with 0 major version. I'm not sure if pinning down vodozemac 0.9 in Cargo.toml is the best way but that's as good as I can come up with for now.

[redstrate]

Without testing, IIRC you can specify a "*" and Cargo should be smart enough to pick the version used by the other crates.

[TobiasFella]

Fixed by not depending on vodozemac directly.

[KitsuneRal]

Hm, that seemed to work although when I tried the same locally before it was complaining about not finding the crate. Whatever works.
I'll give it a couple more runs but (aside from panics in case of some errors) this does seem good to go.

[KitsuneRal]

I don't want to hold it off for another month while I'm away; so I guess that's it.