rewrite macos-deployment-target to rmake

Author: Oneirical

src/tools/tidy/src/allowed_run_make_makefiles.txt

@@ -16,7 +16,6 @@ run-make/libtest-junit/Makefile
 run-make/libtest-thread-limit/Makefile
 run-make/long-linker-command-lines-cmd-exe/Makefile
 run-make/long-linker-command-lines/Makefile
-run-make/macos-deployment-target/Makefile
 run-make/min-global-align/Makefile
 run-make/native-link-modifier-bundle/Makefile
 run-make/no-alloc-shim/Makefile

tests/run-make/macos-deployment-target/Makefile

@@ -0,0 +1,29 @@
+// Check that a set deployment target actually makes it to the linker.
+// This is important since its a compatibility hazard. The linker will
+// generate load commands differently based on what minimum OS it can assume.
+// See https://github.com/rust-lang/rust/pull/105123
+
+//@ only-macos
+
+use run_make_support::{cmd, rustc};
+
+fn main() {
+    rustc()
+        .env("MACOSX_DEPLOYMENT_TARGET", "10.13")
+        .input("with_deployment_target.rs")
+        .output("with_deployment_target.dylib")
+        .run();
+    let pattern = if cmd("uname").arg("-m").run().stdout_utf8().contains("arm64") {
+        "minos 11.0"
+    } else {
+        "version 10.13"
+    };
+    // XXX: The check is for either the x86_64 minimum OR the aarch64 minimum
+    // (M1 starts at macOS 11). They also use different load commands, so we let that change with
+    // each too. The aarch64 check isn't as robust as the x86 one, but testing both seems unneeded.
+    cmd("vtool")
+        .arg("-show-build")
+        .arg("with_deployment_target.dylib")
+        .run()
+        .assert_stdout_contains("pattern");
+}