Regression in 1.26: cannot find function `assert_test_result` in module `self::test`

[emilyalbini]

Multiple crates are failing with the error cannot find function 'assert_test_result' in module 'self::test' .

• html5ever regressed from stable to beta (build log) cc @SimonSapin
• idna regressed from stable to beta (build log) cc @SimonSapin
• rustc-test regressed from stable to beta (build log) cc @SimonSapin
• url regressed from stable to beta (build log) cc @SimonSapin
• xml5ever regressed from stable to beta (build log) cc @SimonSapin
• xmlparser regressed from stable to beta (build log) cc @RazrFalcon

[Mark-Simulacrum]

This feels like a problem on our part, maybe related to some libsyntax or resolution related changes, cc @petrochenkov @alexcrichton...

[SimonSapin]

This is probably due to using https://crates.io/crates/rustc-test/0.2.0 which has [lib] name = "test" and so ends being used by code generated rustc --test instead of of the standard library’s test crate. In retrospect this "clever hack" was obviously a mistake since it ends up relying on unstable implementation details of that crate.

https://crates.io/crates/rustc-test/0.3.0 fixes this (by not overloading the test crate name, it is imported as extern crate rustc_test; instead). Each crate listed above has already updated to 0.3 at least in their git repository (only some of them in the latest crates.io version). Some of the build logs use old crate versions.

I don’t know if there’s anything for the Rust project to do here other than maybe trying to patch that "hole" in the stability system. (Maybe by somehow having crates from the sysroot take priority over Cargo dependencies? Or reject conflicts? Make Cargo pass --extern to rustc for sysroot crates?)

[alexcrichton]

Ah this is unfortunate, @SimonSapin's diagnosis sounds spot on. Ideally we never would have allowed this in the first place but I think at this point it's basically "this is technically out of our stability guarantees so not much we can do"

[SimonSapin]

I think that just because I managed (more or less accidentally) to use on the Stable channel some features that were really unstable doesn’t mean that we should consider them de-facto stable.

[alexcrichton]

This was discussed during libs triage and we've decided to close this as "wontfix"