Auto merge of rust-lang#127060 - Oneirical:testificate, r=jieyouxu

Migrate `symbol-visibility` `run-make` test to rmake

Part of rust-lang#121876 and the associated [Google Summer of Code project](https://blog.rust-lang.org/2024/05/01/gsoc-2024-selected-projects.html).

Pretty scary!

- The expected number of symbols on each check has been changed slightly to reflect the differences between `llvm_readobj` and `nm`, as I think the former will print hidden symbols once and visible symbols twice, while the latter will only print visible symbols.
- The original test ran the same exact checks on `cdylib` twice, for seemingly no reason. I have removed it.
- This may be possible to optimize some more? `llvm_readobj` could get called only once for each library type, and the regex could avoid being created repeatedly. I am not sure if these kinds of considerations are important for a `run-make` test.

Demands a Windows try-job.

try-job: x86_64-mingw

src/tools/compiletest/src/command-list.rs

@@ -117,6 +117,7 @@ const KNOWN_DIRECTIVE_NAMES: &[&str] = &[
     "ignore-watchos",
     "ignore-windows",
     "ignore-windows-gnu",
+    "ignore-windows-msvc",
     "ignore-x32",
     "ignore-x86",
     "ignore-x86_64",

src/tools/run-make-support/src/command.rs

@@ -166,11 +166,19 @@ pub struct CompletedProcess {
 
 impl CompletedProcess {
     #[must_use]
+    #[track_caller]
     pub fn stdout_utf8(&self) -> String {
         String::from_utf8(self.output.stdout.clone()).expect("stdout is not valid UTF-8")
     }
 
     #[must_use]
+    #[track_caller]
+    pub fn invalid_stdout_utf8(&self) -> String {
+        String::from_utf8_lossy(&self.output.stdout.clone()).to_string()
+    }
+
+    #[must_use]
+    #[track_caller]
     pub fn stderr_utf8(&self) -> String {
         String::from_utf8(self.output.stderr.clone()).expect("stderr is not valid UTF-8")
     }

src/tools/tidy/src/allowed_run_make_makefiles.txt

@@ -49,7 +49,6 @@ run-make/split-debuginfo/Makefile
 run-make/stable-symbol-names/Makefile
 run-make/staticlib-dylib-linkage/Makefile
 run-make/symbol-mangling-hashed/Makefile
-run-make/symbol-visibility/Makefile
 run-make/sysroot-crates-are-unstable/Makefile
 run-make/thumb-none-cortex-m/Makefile
 run-make/thumb-none-qemu/Makefile

tests/run-make/symbol-visibility/rmake.rs

@@ -0,0 +1,179 @@
+// Dynamic libraries on Rust used to export a very high amount of symbols,
+// going as far as filling the output with mangled names and generic function
+// names. After the rework of #38117, this test checks that no mangled Rust symbols
+// are exported, and that generics are only shown if explicitely requested.
+// See https://github.com/rust-lang/rust/issues/37530
+
+//@ ignore-windows-msvc
+
+//FIXME(Oneirical): This currently uses llvm-nm for symbol detection. However,
+// the custom Rust-based solution of #128314 may prove to be an interesting alternative.
+
+use run_make_support::{bin_name, dynamic_lib_name, is_darwin, is_windows, llvm_nm, regex, rustc};
+
+fn main() {
+    let cdylib_name = dynamic_lib_name("a_cdylib");
+    let rdylib_name = dynamic_lib_name("a_rust_dylib");
+    let exe_name = bin_name("an_executable");
+    let combined_cdylib_name = dynamic_lib_name("combined_rlib_dylib");
+    rustc().arg("-Zshare-generics=no").input("an_rlib.rs").run();
+    rustc().arg("-Zshare-generics=no").input("a_cdylib.rs").run();
+    rustc().arg("-Zshare-generics=no").input("a_rust_dylib.rs").run();
+    rustc().arg("-Zshare-generics=no").input("a_proc_macro.rs").run();
+    rustc().arg("-Zshare-generics=no").input("an_executable.rs").run();
+    rustc()
+        .arg("-Zshare-generics=no")
+        .input("a_cdylib.rs")
+        .crate_name("combined_rlib_dylib")
+        .crate_type("rlib,cdylib")
+        .run();
+
+    // Check that a cdylib exports its public #[no_mangle] functions
+    symbols_check(&cdylib_name, SymbolCheckType::StrSymbol("public_c_function_from_cdylib"), true);
+    // Check that a cdylib exports the public #[no_mangle] functions of dependencies
+    symbols_check(&cdylib_name, SymbolCheckType::StrSymbol("public_c_function_from_rlib"), true);
+    // Check that a cdylib DOES NOT export any public Rust functions
+    symbols_check(&cdylib_name, SymbolCheckType::AnyRustSymbol, false);
+
+    // Check that a Rust dylib exports its monomorphic functions
+    symbols_check(
+        &rdylib_name,
+        SymbolCheckType::StrSymbol("public_c_function_from_rust_dylib"),
+        true,
+    );
+    symbols_check(
+        &rdylib_name,
+        SymbolCheckType::StrSymbol("public_rust_function_from_rust_dylib"),
+        true,
+    );
+    // Check that a Rust dylib does not export generics if -Zshare-generics=no
+    symbols_check(
+        &rdylib_name,
+        SymbolCheckType::StrSymbol("public_generic_function_from_rust_dylib"),
+        false,
+    );
+
+    // Check that a Rust dylib exports the monomorphic functions from its dependencies
+    symbols_check(&rdylib_name, SymbolCheckType::StrSymbol("public_c_function_from_rlib"), true);
+    symbols_check(&rdylib_name, SymbolCheckType::StrSymbol("public_rust_function_from_rlib"), true);
+    // Check that a Rust dylib does not export generics if -Zshare-generics=no
+    symbols_check(
+        &rdylib_name,
+        SymbolCheckType::StrSymbol("public_generic_function_from_rlib"),
+        false,
+    );
+
+    // FIXME(nbdd0121): This is broken in MinGW, see https://github.com/rust-lang/rust/pull/95604#issuecomment-1101564032
+    // if is_windows() {
+    //     // Check that an executable does not export any dynamic symbols
+    //     symbols_check(&exe_name, SymbolCheckType::StrSymbol("public_c_function_from_rlib")
+    //, false);
+    //     symbols_check(
+    //         &exe_name,
+    //         SymbolCheckType::StrSymbol("public_rust_function_from_exe"),
+    //         false,
+    //     );
+    // }
+
+    // Check the combined case, where we generate a cdylib and an rlib in the same
+    // compilation session:
+    // Check that a cdylib exports its public #[no_mangle] functions
+    symbols_check(
+        &combined_cdylib_name,
+        SymbolCheckType::StrSymbol("public_c_function_from_cdylib"),
+        true,
+    );
+    // Check that a cdylib exports the public #[no_mangle] functions of dependencies
+    symbols_check(
+        &combined_cdylib_name,
+        SymbolCheckType::StrSymbol("public_c_function_from_rlib"),
+        true,
+    );
+    // Check that a cdylib DOES NOT export any public Rust functions
+    symbols_check(&combined_cdylib_name, SymbolCheckType::AnyRustSymbol, false);
+
+    rustc().arg("-Zshare-generics=yes").input("an_rlib.rs").run();
+    rustc().arg("-Zshare-generics=yes").input("a_cdylib.rs").run();
+    rustc().arg("-Zshare-generics=yes").input("a_rust_dylib.rs").run();
+    rustc().arg("-Zshare-generics=yes").input("an_executable.rs").run();
+
+    // Check that a cdylib exports its public #[no_mangle] functions
+    symbols_check(&cdylib_name, SymbolCheckType::StrSymbol("public_c_function_from_cdylib"), true);
+    // Check that a cdylib exports the public #[no_mangle] functions of dependencies
+    symbols_check(&cdylib_name, SymbolCheckType::StrSymbol("public_c_function_from_rlib"), true);
+    // Check that a cdylib DOES NOT export any public Rust functions
+    symbols_check(&cdylib_name, SymbolCheckType::AnyRustSymbol, false);
+
+    // Check that a Rust dylib exports its monomorphic functions, including generics this time
+    symbols_check(
+        &rdylib_name,
+        SymbolCheckType::StrSymbol("public_c_function_from_rust_dylib"),
+        true,
+    );
+    symbols_check(
+        &rdylib_name,
+        SymbolCheckType::StrSymbol("public_rust_function_from_rust_dylib"),
+        true,
+    );
+    symbols_check(
+        &rdylib_name,
+        SymbolCheckType::StrSymbol("public_generic_function_from_rust_dylib"),
+        true,
+    );
+
+    // Check that a Rust dylib exports the monomorphic functions from its dependencies
+    symbols_check(&rdylib_name, SymbolCheckType::StrSymbol("public_c_function_from_rlib"), true);
+    symbols_check(&rdylib_name, SymbolCheckType::StrSymbol("public_rust_function_from_rlib"), true);
+    symbols_check(
+        &rdylib_name,
+        SymbolCheckType::StrSymbol("public_generic_function_from_rlib"),
+        true,
+    );
+
+    // FIXME(nbdd0121): This is broken in MinGW, see https://github.com/rust-lang/rust/pull/95604#issuecomment-1101564032
+    // if is_windows() {
+    //     // Check that an executable does not export any dynamic symbols
+    //     symbols_check(&exe_name, SymbolCheckType::StrSymbol("public_c_function_from_rlib")
+    //, false);
+    //     symbols_check(
+    //         &exe_name,
+    //         SymbolCheckType::StrSymbol("public_rust_function_from_exe"),
+    //         false,
+    //     );
+    // }
+}
+
+#[track_caller]
+fn symbols_check(path: &str, symbol_check_type: SymbolCheckType, exists_once: bool) {
+    let mut nm = llvm_nm();
+    if is_windows() {
+        nm.arg("--extern-only");
+    } else if is_darwin() {
+        nm.arg("--extern-only").arg("--defined-only");
+    } else {
+        nm.arg("--dynamic");
+    }
+    let out = nm.input(path).run().stdout_utf8();
+    assert_eq!(
+        out.lines()
+            .filter(|&line| !line.contains("__imp_") && has_symbol(line, symbol_check_type))
+            .count()
+            == 1,
+        exists_once
+    );
+}
+
+fn has_symbol(line: &str, symbol_check_type: SymbolCheckType) -> bool {
+    if let SymbolCheckType::StrSymbol(expected) = symbol_check_type {
+        line.contains(expected)
+    } else {
+        let regex = regex::Regex::new(r#"_ZN.*h.*E\|_R[a-zA-Z0-9_]+"#).unwrap();
+        regex.is_match(line)
+    }
+}
+
+#[derive(Clone, Copy)]
+enum SymbolCheckType {
+    StrSymbol(&'static str),
+    AnyRustSymbol,
+}