`warning: doc tests are not supported for crate type(s) cdylib` even with explicit `doctest = false`

[sourcefrog]

With this Cargo.toml

[package]
name = "rustdoc-warning-repro"
version = "0.1.0"
edition = "2024"

[lib]
crate-type = [ "cdylib" ]
doctest = false

Run

; cargo t --doc --workspace
warning: doc tests are not supported for crate type(s) `cdylib` in package `rustdoc-warning-repro`

In the case of an isolated crate I suppose it makes sense to tell you there's nothing to run, but if this is inside a workspace and you're running cargo t --workspace --doc the error message is noise since you've asked for them not to be run.

I realize this is very minor, feel free to close if it seems too trivial. But maybe it's easy to suppress the message if doctest = false?

Meta

rustc --version --verbose:

; rustc --version --verbose
rustc 1.90.0-beta.7 (fb918cec0 2025-08-29)
binary: rustc
commit-hash: fb918cec013920472f6975b2009b0339d9e8cc9c
commit-date: 2025-08-29
host: aarch64-apple-darwin
release: 1.90.0-beta.7
LLVM version: 20.1.8

[weihanglo]

I am really sorry that I didn't look closer to the behavior change and the original issue. The proposed change seems wrong here?

cargo test --doc means you intentionally want to doctest the lib, regardless doctest = false or not. The doctest field means by default (without any target selection flags) cargo test will run doctest for it or not.
https://doc.rust-lang.org/nightly/cargo/reference/cargo-targets.html#the-doctest-field

So this is an intentional behavior because the user intent is doctest (--doc is present). One thing we could do in the future is leverage the cargo linting system (-Zcargo-lints) or some similar warning controls.

[weihanglo]

User may set doctest = false for different reasons, like doctest is slow and they want to run it in a separate job. If the warning were suppressed it may be a bit hard to figure out doctest were not run.

[sourcefrog]

Yes, I see what you mean. For one single package when you say doctest = false in the config and --doc on the command line it seems reasonable to warn.

However when you have a workspace with many packages and some can never support doctests, it's nice not to warn about them every time?

The reason to run cargo test --workspace --doc is that nextest doesn't support doctests, so being able to run only the doctests for the whole workspace is useful.

[weihanglo]

However when you have a workspace with many packages and some can never support doctests, it's nice not to warn about them every time?

To me it is still useful. One may accidentally change the crate-type field and didn't know it implicitly makes the lib doctestable. Was the original problem a verbosity issue like having the same warning for different packages too many times? If that we might want to consolidate to one. If not, we could work towards a new lint which can also be tracked in #12235?

[weihanglo]

The reason to run cargo test --workspace --doc is that nextest doesn't support doctests, so being able to run only the doctests for the whole workspace is useful.

BTW, this is tracked in rust-lang/testing-devex-team#5, and rust-lang/rust#134529 already paved a possible path towards it. I meant we eventually want to make doctest less special, and just works as other tests.