Tests for field is never read diagnostic

Author: sunjay

src/test/ui/lint/dead-code/drop-only-field-issue-81658.rs

@@ -0,0 +1,42 @@
+//! The field `guard` is never used directly, but it is still useful for its side effect when
+//! dropped. Since rustc doesn't consider a `Drop` impl as a use, we want to make sure we at least
+//! produce a helpful diagnostic that points the user to what they can do if they indeed intended to
+//! have a field that is only used for its `Drop` side effect.
+//!
+//! Issue: https://github.com/rust-lang/rust/issues/81658
+
+#![deny(dead_code)]
+
+use std::sync::{Mutex, MutexGuard};
+
+/// Holds a locked value until it is dropped
+pub struct Locked<'a, T> {
+    // Field is kept for its affect when dropped, but otherwise unused
+    guard: MutexGuard<'a, T>, //~ ERROR field is never read
+}
+
+impl<'a, T> Locked<'a, T> {
+    pub fn new(value: &'a Mutex<T>) -> Self {
+        Self {
+            guard: value.lock().unwrap(),
+        }
+    }
+}
+
+fn main() {
+    let items = Mutex::new(vec![1, 2, 3]);
+
+    // Hold a lock on items while doing something else
+    let result = {
+        // The lock will be released at the end of this scope
+        let _lock = Locked::new(&items);
+
+        do_something_else()
+    };
+
+    println!("{}", result);
+}
+
+fn do_something_else() -> i32 {
+    1 + 1
+}

src/test/ui/lint/dead-code/drop-only-field-issue-81658.stderr

@@ -0,0 +1,14 @@
+error: field is never read: `guard`
+  --> $DIR/drop-only-field-issue-81658.rs:15:5
+   |
+LL |     guard: MutexGuard<'a, T>,
+   |     ^^^^^^^^^^^^^^^^^^^^^^^^
+   |
+note: the lint level is defined here
+  --> $DIR/drop-only-field-issue-81658.rs:8:9
+   |
+LL | #![deny(dead_code)]
+   |         ^^^^^^^^^
+
+error: aborting due to previous error
+

src/test/ui/lint/dead-code/field-used-in-ffi-issue-81658.rs

@@ -0,0 +1,50 @@
+//! The field `items` is being "used" by FFI (implicitly through pointers). However, since rustc
+//! doesn't know how to detect that, we produce a message that says the field is unused. This can
+//! cause some confusion and we want to make sure our diagnostics help as much as they can.
+//!
+//! Issue: https://github.com/rust-lang/rust/issues/81658
+
+#![deny(dead_code)]
+
+/// A struct for holding on to data while it is being used in our FFI code
+pub struct FFIData<T> {
+    /// These values cannot be dropped while the pointers to each item
+    /// are still in use
+    items: Option<Vec<T>>, //~ ERROR field is never read
+}
+
+impl<T> FFIData<T> {
+    pub fn new() -> Self {
+        Self {items: None}
+    }
+
+    /// Load items into this type and return pointers to each item that can
+    /// be passed to FFI
+    pub fn load(&mut self, items: Vec<T>) -> Vec<*const T> {
+        let ptrs = items.iter().map(|item| item as *const _).collect();
+
+        self.items = Some(items);
+
+        ptrs
+    }
+}
+
+extern {
+    /// The FFI code that uses items
+    fn process_item(item: *const i32);
+}
+
+fn main() {
+    // Data cannot be dropped until the end of this scope or else the items
+    // will be dropped before they are processed
+    let mut data = FFIData::new();
+
+    let ptrs = data.load(vec![1, 2, 3, 4, 5]);
+
+    for ptr in ptrs {
+        // Safety: This pointer is valid as long as the arena is in scope
+        unsafe { process_item(ptr); }
+    }
+
+    // Items will be safely freed at the end of this scope
+}

src/test/ui/lint/dead-code/field-used-in-ffi-issue-81658.stderr

@@ -0,0 +1,14 @@
+error: field is never read: `items`
+  --> $DIR/field-used-in-ffi-issue-81658.rs:13:5
+   |
+LL |     items: Option<Vec<T>>,
+   |     ^^^^^^^^^^^^^^^^^^^^^
+   |
+note: the lint level is defined here
+  --> $DIR/field-used-in-ffi-issue-81658.rs:7:9
+   |
+LL | #![deny(dead_code)]
+   |         ^^^^^^^^^
+
+error: aborting due to previous error
+