[1.30 beta] Unresolved import errors in a few crates

[pietroalbini]

Some crates are failing in 1.30 beta with errors like this one:

 error[E0432]: unresolved import `geometry::Isometry`
   --> /cargo-home/registry/src/gh.neting.cc-1ecc6299db9ec823/nalgebra-0.15.3/src/base/cg.rs:15:16
    |
 15 | use geometry::{Isometry, IsometryMatrix3, Orthographic3, Perspective3, Point, Point3, Rotation2,
    |                ^^^^^^^^ no `Isometry` in `geometry`. Did you mean to use `isometry`?

• casimir-fdfd 0.2.0 regressed from stable to beta (build log) cc @ThomasdenH
• ipfs-api 0.5.0-alpha2 regressed from stable to beta (build log) cc @ferristseng
  • world_id 0.1.0 (depends on ipfs-api) regressed from stable to beta (build log) cc @icefoxen
• riscan-pro 0.2.1 regressed from stable to beta (build log) cc @gadomski
• Enet4/propan regressed from stable to beta (build log) cc @Enet4

cc @petrochenkov

[petrochenkov]

The root issue is in the nalgebra crate and it's something that was fixed between nalgebra 0.16.0 and nalgebra 0.16.1.

EDIT: nalgebra 0.16.1 removed use serde; imports, almost certainly as a response to this regression.

[icefoxen]

The ipfs-api crate does not use nlagebra at all. All of these build logs seem to have something wrong with serde involved; for example, ipfs-api gets a lot of errors associated with its use of #[serde(rename_all = "PascalCase")], along the lines of:

Sep 21 17:09:54.166 INFO kablam! error: cannot determine resolution for the attribute macro `serde`
Sep 21 17:09:54.166 INFO kablam!   --> /cargo-home/registry/src/gh.neting.cc-1ecc6299db9ec823/nalgebra-0.15.3/src/linalg/symmetric_tridiagonal.rs:16:5
Sep 21 17:09:54.166 INFO kablam!    |
Sep 21 17:09:54.166 INFO kablam! 16 |     serde(
Sep 21 17:09:54.166 INFO kablam!    |     ^^^^^
Sep 21 17:09:54.166 INFO kablam!    |
Sep 21 17:09:54.166 INFO kablam!    = note: import resolution is stuck, try simplifying macro imports

All the other build logs appear to have "undefined type or module" along with the error import resolution is stuck, try simplifying macro imports. Removing the serde annotations from ipfs-api lets it build successfully, though the protocol it speaks is then incorrect.

[petrochenkov]

Yes, ipfs-api is a separate case (probably with the same underlying issue in rustc) and it's already tracked separately in #54386, the other regressed crates use nalgebra.

[icefoxen]

Aha, thanks!

[petrochenkov]

Minimized:

#![allow(unused)]

#[macro_use]
extern crate serde_derive;

use self::one::*;
use self::two::*;

mod serde {}

mod one {
    use serde;

    #[derive(Serialize)]
    #[serde]
    struct One;
}

mod two {
    use serde;

    #[derive(Serialize)]
    #[serde]
    struct Two;
}

fn main() {}

[petrochenkov]

Fixed in #54518
All the crates listed in #54471 (comment) build successfully now.