Always serialize and deserialize ids the same way

[nical]

Description

RawId's serialization/deserialization currently depends on what cargo features are enabled. If trace is enabled it is serialized via the SerialId type and if replay is enabled it is deserialized from SerialId.

This means that if an application serializes and deserializes Ids (like gecko does for some of the remoting), then having only one of trace and replay features enabled is broken. Firefox only uses the trace feature.

For example:

let id = wgc::id::AdapterId::zip(0, 1, wgt::Backend::Vulkan);
let mut tmp = Vec::new();
bincode::serialize_into(&mut tmp, &id).unwrap();
let de = bincode::deserialize::<wgc::id::AdapterId>(tmp.as_slice()).unwrap(); // Boom.

It's unclear why it used to work and only started breaking at the latest update.

With this patch PR, ids are always serialized via SerialId. It's probably not what we want long term, but we need the fix rather urgently so it will have to do for now.

Checklist

• Run cargo fmt.
• Run cargo clippy.
• Run cargo xtask test to run tests.

[teoxoy]

It's unclear why it used to work and only started breaking at the latest update.

#5149 changed some of the features but I don't see anything off that could have caused this.

It's worth noting that the serial-pass feature was replaced by serde but if either replay or trace are enabled it should also get enabled.

Firefox only uses the trace feature.

I see we also use the replay feature.

This change looks fine to me either way.

[nical]

I see we also use the replay feature.

This change looks fine to me either way.

Ah, that explains it then. I'm continuing Jim's work-in-progress wgpu update, I'm on top of https://phabricator.services.mozilla.com/D200266 where trace is enabled but not replay.

[nical]

Let's get rid of the footgun anyway.