Manifests in rustc-src distributed package are not valid any more

[RalfJung]

The Miri build against latest rustc master is failing with the following error:

error: failed to parse manifest at `/home/runner/.rustup/toolchains/miri/lib/rustlib/rustc-src/rust/compiler/rustc/Cargo.toml`

Caused by:
  error inheriting `lints` from workspace root manifest's `workspace.lints`

The problem is that that Cargo.toml file appears inside a workspace in this repo, but not in rustc-src. So this breaks every useage of cargo on the distributed rustc-src artifacts.

This is caused by #138084. Cc @nnethercote @jieyouxu. I think that PR should be reverted to minimize downstream breakage while we consider our options.

[jieyouxu]

... Yeah. I'll post a revert to buy time to reconsider this.

[jieyouxu]

Revert: #138306.

Looks like we may have to go the magical-env-var-hack or sth (cf. #138106) afterall...

[bjorn3]

Why is the root Cargo.toml not included in rustc-src? Is it because rustc-src doesn't contain other projects that are part of the root workspace too?

[jieyouxu]

Why is the root Cargo.toml not included in rustc-src? Is it because rustc-src doesn't contain other projects that are part of the root workspace too?

I suspect this is the case if rustc-src wants to focus on being the compiler sources, because the root manifest also has stuff for tools, bootstrap bits and library bits.

[RalfJung]

yeah the root Cargo.toml contains a lot of members that are not part of rustc-src. If we wanted to include it, we'd have to filter the members list, as cargo AFAIK will error out when a member does not exist.

[tgross35]

In addition to lints, workspace dependencies may be nice in the future. If fixing up the workspace Cargo.toml before release is an option, that seems like a nice way forward so we keep its advantages.

It may also be possible to split the workspaces based on what gets distributed, like Bjorn did for library/ somewhat recently.

[jieyouxu]

The revert landed so closing this specific instance as fixed. However, this is one of those general #136822 problems.

And yeah, having multiple dist components that serve different consumers but that we don't really test is really not great.