don't use force-unstable-if-unmarked with x test in standard library doctests

[onur-ozkan]

Using force-unstable-if-unmarked on library/ makes bootstrap ignoring doctests errors, potentially leading to the merging of PRs that do not compile successfully.

For testing this change, apply the following patch(or remove one of the error_in_core in library/core/src/error.rs):

diff --git a/library/core/src/error.rs b/library/core/src/error.rs
index 1170221c10c..b0ab2b04032 100644
--- a/library/core/src/error.rs
+++ b/library/core/src/error.rs
@@ -130,7 +130,6 @@ fn cause(&self) -> Option<&dyn Error> {
     ///
     /// ```rust
     /// #![feature(error_generic_member_access)]
-    /// #![feature(error_in_core)]
     /// use core::fmt;
     /// use core::error::{request_ref, Request};
     ///
@@ -361,7 +360,6 @@ pub fn sources(&self) -> Source<'_> {
 ///
 /// ```rust
 /// # #![feature(error_generic_member_access)]
-/// # #![feature(error_in_core)]
 /// use std::error::Error;
 /// use core::error::request_value;
 ///
@@ -385,7 +383,6 @@ pub fn request_value<'a, T>(err: &'a (impl Error + ?Sized)) -> Option<T>
 ///
 /// ```rust
 /// # #![feature(error_generic_member_access)]
-/// # #![feature(error_in_core)]
 /// use core::error::Error;
 /// use core::error::request_ref;
 ///
@@ -457,7 +454,6 @@ fn request_by_type_tag<'a, I>(err: &'a (impl Error + ?Sized)) -> Option<I::Reifi
 ///
 /// ```
 /// #![feature(error_generic_member_access)]
-/// #![feature(error_in_core)]
 /// use core::fmt;
 /// use core::error::Request;
 /// use core::error::request_ref;
@@ -528,7 +524,6 @@ fn new<'b>(erased: &'b mut (dyn Erased<'a> + 'a)) -> &'b mut Request<'a> {
     ///
     /// ```rust
     /// #![feature(error_generic_member_access)]
-    /// #![feature(error_in_core)]
     ///
     /// use core::error::Request;
     ///
@@ -563,7 +558,6 @@ pub fn provide_value<T>(&mut self, value: T) -> &mut Self
     ///
     /// ```rust
     /// #![feature(error_generic_member_access)]
-    /// #![feature(error_in_core)]
     ///
     /// use core::error::Request;
     ///
@@ -599,7 +593,6 @@ pub fn provide_value_with<T>(&mut self, fulfil: impl FnOnce() -> T) -> &mut Self
     ///
     /// ```rust
     /// #![feature(error_generic_member_access)]
-    /// #![feature(error_in_core)]
     ///
     /// use core::error::Request;
     ///
@@ -632,7 +625,6 @@ pub fn provide_ref<T: ?Sized + 'static>(&mut self, value: &'a T) -> &mut Self {
     ///
     /// ```rust
     /// #![feature(error_generic_member_access)]
-    /// #![feature(error_in_core)]
     ///
     /// use core::error::Request;
     ///
@@ -698,7 +690,6 @@ fn provide_with<I>(&mut self, fulfil: impl FnOnce() -> I::Reified) -> &mut Self
     /// it.
     ///
     /// ```rust
-    /// #![feature(error_generic_member_access)]
     /// #![feature(error_in_core)]
     ///
     /// use core::error::Request;
@@ -787,7 +778,6 @@ pub fn would_be_satisfied_by_value_of<T>(&self) -> bool
     ///
     /// ```rust
     /// #![feature(error_generic_member_access)]
-    /// #![feature(error_in_core)]
     ///
     /// use core::error::Request;
     /// use core::error::request_ref;

then run x test library/core --doc

Related #114838

[rustbot]

r? @Mark-Simulacrum

(rustbot has picked a reviewer for you, use r? to override)

[onur-ozkan]

I don't think the block I removed makes any sense.. But still, I am not certain about regressions with this change.

@bors rollup=never

[onur-ozkan]

Okay I see how it makes sense...

[RalfJung]

It certainly makes no sense to remove this without also removing the code that receives this env var

rust/src/bootstrap/bin/rustc.rs

Lines 158 to 160 in 795ade0

	if env::var_os("RUSTC_FORCE_UNSTABLE").is_some() {
	cmd.arg("-Z").arg("force-unstable-if-unmarked");
	}

rust/src/bootstrap/bin/rustdoc.rs

Lines 55 to 57 in 795ade0

	if env::var_os("RUSTC_FORCE_UNSTABLE").is_some() {
	cmd.arg("-Z").arg("force-unstable-if-unmarked");
	}

But this is old code, and I think it is very important. For instance it ensures that if people load the libc crate from the sysroot, they get stability errors, even though libc does not contain any unstable markers.

It's generally a bad idea to remove code if you cannot explain why it was there. :)

[RalfJung]

However, if removing this env var helps, that should hopefully get us one step closer to solving the puzzle. Just, the env var has to stay.

[onur-ozkan]

Yeah, removing it indeed was a bad idea. Will fix the problem by updating the condition instead of removing it.

It's generally a bad idea to remove code if you cannot explain why it was there. :)

Very true, at the first look it didn't make any sense. But there is the CI report so.. 😄

[onur-ozkan]

Just put the PR mode in draft. Tomorrow I will consider the best way to handle this.

[onur-ozkan]

@rustbot ready

[RalfJung]

I'm not sure this is the right appraoch. Shouldn't rustc be fixed to still complain about the missing feature gate despite -Zforce-unstable-if-unmarked?

[onur-ozkan]

I'm not sure this is the right appraoch. Shouldn't rustc be fixed to still complain about the missing feature gate despite -Zforce-unstable-if-unmarked?

Regarding to your comment here #114838 (comment), no it should not.

If you meaning fixing this from the compiler side, I can add a FIXME note to the change I did. This PR at least ensures not merging the broken changes until the rustc fix

[onur-ozkan]

If you meaning fixing this from the compiler side, I can add a FIXME note to the change I did. This PR at least ensures not merging the broken changes until the rustc fix

I guess this will make things more complicated. I will keep this PR draft and let the compiler team fix this problem with the right approach