[crashtracker] Retrieve panic message when crashing

bin_tests/Cargo.toml

@@ -23,6 +23,7 @@ libc = "0.2"
 nix = { version = "0.29", features = ["signal", "socket"] }
 hex = "0.4"
 os_info = "3.7.0"
+regex = "1.0"
 
 [dev-dependencies]
 serial_test = "3.2"

bin_tests/src/bin/crashing_test_app.rs

@@ -14,6 +14,8 @@ mod unix {
     use anyhow::ensure;
     use anyhow::Context;
     use std::env;
+    use std::sync::atomic::{AtomicBool, Ordering};
+    use std::sync::Arc;
     use std::time::Duration;
 
     use libdd_common::{tag, Endpoint};
@@ -23,8 +25,25 @@ mod unix {
 
     const TEST_COLLECTOR_TIMEOUT: Duration = Duration::from_secs(10);
 
-    #[inline(never)]
-    unsafe fn fn3() {
+    #[derive(Debug, Clone, Copy, PartialEq, Eq)]
+    enum CrashType {
+        Segfault,
+        Panic,
+    }
+
+    impl std::str::FromStr for CrashType {
+        type Err = anyhow::Error;
+        fn from_str(s: &str) -> Result<Self, Self::Err> {
+            match s {
+                "segfault" => Ok(CrashType::Segfault),
+                "panic" => Ok(CrashType::Panic),
+                _ => anyhow::bail!("Invalid crash type: {s}"),
+            }
+        }
+    }
+
+    #[inline(always)]
+    unsafe fn cause_segfault() {
         #[cfg(any(target_arch = "x86", target_arch = "x86_64"))]
         {
             std::arch::asm!("mov eax, [0]", options(nostack));
@@ -37,13 +56,25 @@ mod unix {
     }
 
     #[inline(never)]
-    fn fn2() {
-        unsafe { fn3() }
+    fn fn3(crash_type: CrashType) {
+        match crash_type {
+            CrashType::Segfault => {
+                unsafe { cause_segfault() };
+            }
+            CrashType::Panic => {
+                panic!("program panicked");
+            }
+        }
     }
 
     #[inline(never)]
-    fn fn1() {
-        fn2()
+    fn fn2(crash_type: CrashType) {
+        fn3(crash_type);
+    }
+
+    #[inline(never)]
+    fn fn1(crash_type: CrashType) {
+        fn2(crash_type);
     }
 
     #[inline(never)]
@@ -53,6 +84,7 @@ mod unix {
         let output_url = args.next().context("Unexpected number of arguments 1")?;
         let receiver_binary = args.next().context("Unexpected number of arguments 2")?;
         let output_dir = args.next().context("Unexpected number of arguments 3")?;
+        let crash_type = args.next().context("Unexpected number of arguments 4")?;
         anyhow::ensure!(args.next().is_none(), "unexpected extra arguments");
 
         let stderr_filename = format!("{output_dir}/out.stderr");
@@ -88,6 +120,19 @@ mod unix {
             .collect(),
         };
 
+        let crash_type = crash_type.parse().context("Invalid crash type")?;
+        let is_panic_mode = matches!(crash_type, CrashType::Panic);
+
+        let called_panic_hook = Arc::new(AtomicBool::new(false));
+        let old_hook = std::panic::take_hook();
+        if is_panic_mode {
+            let called_panic_hook_clone = Arc::clone(&called_panic_hook);
+            std::panic::set_hook(Box::new(move |panic_info| {
+                called_panic_hook_clone.store(true, Ordering::SeqCst);
+                old_hook(panic_info);
+            }));
+        }
+
         crashtracker::init(
             config,
             CrashtrackerReceiverConfig::new(
@@ -100,7 +145,13 @@ mod unix {
             metadata,
         )?;
 
-        fn1();
+        fn1(crash_type);
+
+        // If the panic hook was chained, it should have been called.
+        anyhow::ensure!(
+            !is_panic_mode || called_panic_hook.load(Ordering::SeqCst),
+            "panic hook was not called"
+        );
         Ok(())
     }
 }

bin_tests/src/lib.rs

@@ -46,6 +46,7 @@ pub struct ArtifactsBuild {
     pub artifact_type: ArtifactType,
     pub build_profile: BuildProfile,
     pub triple_target: Option<String>,
+    pub panic_abort: Option<bool>,
 }
 
 fn inner_build_artifact(c: &ArtifactsBuild) -> anyhow::Result<PathBuf> {
@@ -58,6 +59,19 @@ fn inner_build_artifact(c: &ArtifactsBuild) -> anyhow::Result<PathBuf> {
         ArtifactType::ExecutablePackage | ArtifactType::CDylib => build_cmd.arg("-p"),
         ArtifactType::Bin => build_cmd.arg("--bin"),
     };
+
+    if let Some(panic_abort) = c.panic_abort {
+        if panic_abort {
+            let existing_rustflags = std::env::var("RUSTFLAGS").unwrap_or_default();
+            let new_rustflags = if existing_rustflags.is_empty() {
+                "-C panic=abort".to_string()
+            } else {
+                format!("{} -C panic=abort", existing_rustflags)
+            };
+            build_cmd.env("RUSTFLAGS", new_rustflags);
+        }
+    }
+
     build_cmd.arg(&c.name);
 
     let output = build_cmd.output().unwrap();

bin_tests/src/modes/behavior.rs

@@ -134,6 +134,9 @@ pub fn get_behavior(mode_str: &str) -> Box<dyn Behavior> {
         "runtime_callback_frame_invalid_utf8" => {
             Box::new(test_012_runtime_callback_frame_invalid_utf8::Test)
         }
+        "panic_hook_after_fork" => Box::new(test_013_panic_hook_after_fork::Test),
+        "panic_hook_string" => Box::new(test_014_panic_hook_string::Test),
+        "panic_hook_unknown_type" => Box::new(test_015_panic_hook_unknown_type::Test),
         _ => panic!("Unknown mode: {mode_str}"),
     }
 }

bin_tests/src/modes/unix/mod.rs

@@ -13,3 +13,6 @@ pub mod test_009_prechain_with_abort;
 pub mod test_010_runtime_callback_frame;
 pub mod test_011_runtime_callback_string;
 pub mod test_012_runtime_callback_frame_invalid_utf8;
+pub mod test_013_panic_hook_after_fork;
+pub mod test_014_panic_hook_string;
+pub mod test_015_panic_hook_unknown_type;

bin_tests/src/modes/unix/test_013_panic_hook_after_fork.rs

@@ -0,0 +1,120 @@
+// Copyright 2025-Present Datadog, Inc. https://www.datadoghq.com/
+// SPDX-License-Identifier: Apache-2.0
+//
+// Test that panic hooks registered before fork() continue to work in child processes.
+// This validates that:
+// 1. The panic hook survives fork()
+// 2. The panic message is captured in the child process
+// 3. The crash report is correctly generated
+use crate::modes::behavior::Behavior;
+use libdd_crashtracker::{self as crashtracker, CrashtrackerConfiguration};
+use nix::sys::wait::{waitpid, WaitStatus};
+use nix::unistd::Pid;
+use std::fs;
+use std::path::Path;
+use std::time::{Duration, Instant};
+
+pub struct Test;
+
+impl Behavior for Test {
+    fn setup(
+        &self,
+        _output_dir: &Path,
+        _config: &mut CrashtrackerConfiguration,
+    ) -> anyhow::Result<()> {
+        Ok(())
+    }
+
+    fn pre(&self, output_dir: &Path) -> anyhow::Result<()> {
+        pre(output_dir)
+    }
+
+    fn post(&self, output_dir: &Path) -> anyhow::Result<()> {
+        post(output_dir)
+    }
+}
+
+fn pre(output_dir: &Path) -> anyhow::Result<()> {
+    let old_hook = std::panic::take_hook();
+    let output_dir = output_dir.to_path_buf();
+
+    // Set up a panic hook BEFORE crashtracker::init to verify the hook chain works
+    std::panic::set_hook(Box::new(move |panic_info| {
+        // Mark that our custom hook was called by writing a marker file
+        // This works across fork() because it's persistent storage
+        let marker_path = output_dir.join("panic_hook_called.marker");
+        let _ = fs::write(marker_path, "hook was called");
+
+        // Call the previous hook (usually the default panic hook)
+        old_hook(panic_info);
+    }));
+
+    Ok(())
+}
+
+fn post(output_dir: &Path) -> anyhow::Result<()> {
+    match unsafe { libc::fork() } {
+        -1 => {
+            anyhow::bail!("Failed to fork");
+        }
+        0 => {
+            // Child - panic with a specific message
+            // The crashtracker should capture both the panic hook execution
+            // and the panic message
+            crashtracker::begin_op(crashtracker::OpTypes::ProfilerCollectingSample)?;
+
+            // Give parent time to set up wait
+            std::thread::sleep(Duration::from_millis(10));
+
+            panic!("child panicked after fork - hook should fire");
+        }
+        pid => {
+            // Parent - wait for child to panic and crash
+            let start_time = Instant::now();
+            let max_wait = Duration::from_secs(5);
+
+            loop {
+                match waitpid(Pid::from_raw(pid), None)? {
+                    WaitStatus::StillAlive => {
+                        if start_time.elapsed() > max_wait {
+                            anyhow::bail!("Child process did not exit within 5 seconds");
+                        }
+                        std::thread::sleep(Duration::from_millis(10));
+                    }
+                    WaitStatus::Exited(_pid, exit_code) => {
+                        // Child exited - this is what we expect after panic
+                        eprintln!("Child exited with code: {}", exit_code);
+                        break;
+                    }
+                    WaitStatus::Signaled(_pid, signal, _) => {
+                        // Child was killed by signal (also acceptable for panic)
+                        eprintln!("Child killed by signal: {:?}", signal);
+                        break;
+                    }
+                    _ => {
+                        // Other status - continue waiting
+                    }
+                }
+            }
+
+            // Verify that our custom panic hook was called by checking for the marker file
+            // This proves that the hook chain works correctly:
+            // crashtracker's hook -> our custom hook -> default hook
+            let marker_path = output_dir.join("panic_hook_called.marker");
+
+            if !marker_path.exists() {
+                anyhow::bail!(
+                    "Custom panic hook was not called - hook chaining failed! \
+                    Expected marker file at: {}",
+                    marker_path.display()
+                );
+            }
+
+            // Parent exits with error code to indicate test completion
+            // The test harness will verify the crash report contains the panic message
+            unsafe {
+                libc::_exit(1);
+            }
+        }
+    }
+}

bin_tests/src/modes/unix/test_014_panic_hook_string.rs

@@ -0,0 +1,31 @@
+// Copyright 2025-Present Datadog, Inc. https://www.datadoghq.com/
+// SPDX-License-Identifier: Apache-2.0
+//
+// Test that panic hooks work correctly with String payloads (not just &str).
+// This validates that:
+// 1. String panic payloads are correctly captured
+// 2. The panic message format is "Process panicked with message: <msg>"
+use crate::modes::behavior::Behavior;
+use libdd_crashtracker::CrashtrackerConfiguration;
+use std::path::Path;
+
+pub struct Test;
+
+impl Behavior for Test {
+    fn setup(
+        &self,
+        _output_dir: &Path,
+        _config: &mut CrashtrackerConfiguration,
+    ) -> anyhow::Result<()> {
+        Ok(())
+    }
+
+    fn pre(&self, _output_dir: &Path) -> anyhow::Result<()> {
+        Ok(())
+    }
+
+    fn post(&self, _output_dir: &Path) -> anyhow::Result<()> {
+        let dynamic_value = 42;
+        panic!("Panic with value: {}", dynamic_value);
+    }
+}

bin_tests/src/modes/unix/test_015_panic_hook_unknown_type.rs

@@ -0,0 +1,30 @@
+// Copyright 2025-Present Datadog, Inc. https://www.datadoghq.com/
+// SPDX-License-Identifier: Apache-2.0
+//
+// Test that panic hooks work correctly with unknown types via panic_any.
+// This validates that:
+// 1. panic_any() with non-string types is handled gracefully
+// 2. The message format is: "Process panicked with unknown type (<location>)"
+use crate::modes::behavior::Behavior;
+use libdd_crashtracker::CrashtrackerConfiguration;
+use std::path::Path;
+
+pub struct Test;
+
+impl Behavior for Test {
+    fn setup(
+        &self,
+        _output_dir: &Path,
+        _config: &mut CrashtrackerConfiguration,
+    ) -> anyhow::Result<()> {
+        Ok(())
+    }
+
+    fn pre(&self, _output_dir: &Path) -> anyhow::Result<()> {
+        Ok(())
+    }
+
+    fn post(&self, _output_dir: &Path) -> anyhow::Result<()> {
+        std::panic::panic_any(42i32);
+    }
+}

bin_tests/src/test_runner.rs

@@ -190,7 +190,9 @@ where
 /// This is more flexible than `run_crash_test_with_artifacts` and allows for:
 /// - Custom binary selection (e.g., crashing_test_app instead of crashtracker_bin_test)
 /// - Custom command arguments
-/// - Custom exit status expectations
+///
+/// Note: This function always expects the test to crash (exit with non-success status).
+/// All current uses of this function test crash scenarios, not successful exits.
 ///
 /// # Example
 /// ```no_run
@@ -223,7 +225,6 @@ where
 ///             .arg(&artifacts_map[&receiver])
 ///             .arg(&fixtures.output_dir);
 ///     },
-///     false, // expect crash (not success)
 ///     |payload, _fixtures| {
 ///         // Custom validation
 ///         Ok(())
@@ -235,7 +236,6 @@ where
 pub fn run_custom_crash_test<CB, V>(
     binary_path: &std::path::Path,
     command_builder: CB,
-    expect_success: bool,
     validator: V,
 ) -> Result<()>
 where
@@ -251,13 +251,10 @@ where
 
     let exit_status = crate::timeit!("exit after signal", { p.wait()? });
 
-    // Validate exit status
-    let actual_success = exit_status.success();
+    // Validate exit status - custom crash tests always expect failure
     anyhow::ensure!(
-        expect_success == actual_success,
-        "Exit status mismatch: expected success={}, got success={} (exit code: {:?})",
-        expect_success,
-        actual_success,
+        !exit_status.success(),
+        "Expected test to crash (non-success exit), but it succeeded with code: {:?}",
         exit_status.code()
     );