rewrite no-builtins-lto to rmake

src/tools/tidy/src/allowed_run_make_makefiles.txt

@@ -146,7 +146,6 @@ run-make/native-link-modifier-verbatim-rustc/Makefile
 run-make/native-link-modifier-whole-archive/Makefile
 run-make/no-alloc-shim/Makefile
 run-make/no-builtins-attribute/Makefile
-run-make/no-builtins-lto/Makefile
 run-make/no-duplicate-libs/Makefile
 run-make/obey-crate-type-flag/Makefile
 run-make/optimization-remarks-dir-pgo/Makefile

tests/run-make/lto-avoid-object-duplication/rmake.rs

@@ -1,3 +1,4 @@
+// ignore-tidy-tab
 // Staticlibs don't include Rust object files from upstream crates if the same
 // code was already pulled into the lib via LTO. However, the bug described in
 // https://github.com/rust-lang/rust/issues/64153 lead to this exclusion not

tests/run-make/no-builtins-lto/rmake.rs

@@ -0,0 +1,20 @@
+// The rlib produced by a no_builtins crate should be explicitely linked
+// during compilation, and as a result be present in the linker arguments.
+// See the comments inside the test for more details.
+// See https://github.com/rust-lang/rust/pull/35637
+
+use run_make_support::{rust_lib_name, rustc};
+
+fn main() {
+    // Compile a `#![no_builtins]` rlib crate
+    rustc().input("no_builtins.rs").run();
+    // Build an executable that depends on that crate using LTO. The no_builtins crate doesn't
+    // participate in LTO, so its rlib must be explicitly
+    // linked into the final binary. Verify this by grepping the linker arguments.
+    rustc()
+        .input("main.rs")
+        .arg("-Clto")
+        .print("link-args")
+        .run()
+        .assert_stdout_contains(rust_lib_name("no_builtins"));
+}