Collect all crash dumps in directory regardless of main process exit status

[Copilot]

The crash dump extension only collected the testhost process dump by PID, missing dumps from child processes that crashed alongside it. Additionally, it only collected dumps when the main process crashed, but child processes can crash even when the main process exits successfully.

Changes

Modified CrashDumpProcessLifetimeHandler.OnTestHostProcessExitedAsync:

• Removed the HasExitedGracefully check to always collect dumps, regardless of how the main process exited
• Changed from conditional collection (expected dump OR all dumps) to always collecting all .dmp files in the directory
• Added directory validation before file enumeration
• Removed error messages about missing expected dump files (no longer relevant)

Before:

if (cancellation.IsCancellationRequested
    || testHostProcessInformation.HasExitedGracefully
    || (AppDomain.CurrentDomain.GetData("ProcessKilledByHangDump") is string processKilledByHangDump && processKilledByHangDump == "true"))
{
    return;
}

// ... error display ...

string expectedDumpFile = _netCoreCrashDumpGeneratorConfiguration.DumpFileNamePattern.Replace("%p", testHostProcessInformation.PID.ToString(CultureInfo.InvariantCulture));
if (File.Exists(expectedDumpFile))
{
    await _messageBus.PublishAsync(this, new FileArtifact(new FileInfo(expectedDumpFile), ...));
}
else
{
    // Fallback: collect all dumps
    foreach (string dumpFile in Directory.GetFiles(Path.GetDirectoryName(expectedDumpFile)!, "*.dmp"))
    {
        await _messageBus.PublishAsync(this, new FileArtifact(new FileInfo(dumpFile), ...));
    }
}

After:

if (cancellation.IsCancellationRequested
    || (AppDomain.CurrentDomain.GetData("ProcessKilledByHangDump") is string processKilledByHangDump && processKilledByHangDump == "true"))
{
    return;
}

// Validate directory exists
string? dumpDirectory = Path.GetDirectoryName(expectedDumpFile);
if (RoslynString.IsNullOrEmpty(dumpDirectory) || !Directory.Exists(dumpDirectory))
{
    return;
}

// Collect all dump files in the directory to capture crashes from child processes
// We check for any dumps in the directory, even if the main process exited gracefully,
// because child processes might have crashed
foreach (string dumpFile in Directory.GetFiles(dumpDirectory, "*.dmp"))
{
    await _messageBus.PublishAsync(this, new FileArtifact(new FileInfo(dumpFile), ...));
}

This ensures all crash dumps are collected from the unique per-run directory, regardless of whether the main process crashed or exited gracefully. With .NET Core's built-in crash dump mechanism, child processes can crash independently and generate dumps even when the parent process succeeds.

Testing

Added acceptance test CrashDump_WithChildProcess_CollectsMultipleDumps that validates multiple dump collection:

• Spawns a child process that crashes using Environment.FailFast
• Main testhost process also crashes
• Verifies at least 2 dump files (parent + child) are collected
• Uses new CrashDumpWithChildFixture test asset that follows the pattern of existing process tree tests

Fixes #4186

Original prompt

---

This section details on the original issue you should resolve

<issue_title>Collect all dumps in the directory inside the crash dump</issue_title>
<issue_description>From @MarcoRossignoli:

Current extension is only moving the testhost process crash using the pid, but if for some reason children processes will crash together with it we won't report all dumps.</issue_description>

Comments on the Issue (you are @copilot in this section)

• Fixes Collect all dumps in the directory inside the crash dump #4186

---

💡 You can make Copilot smarter by setting up custom instructions, customizing its development environment and configuring Model Context Protocol (MCP) servers. Learn more Copilot coding agent tips in the docs.

[Evangelink]

@copilot add some acceptance test that would lead to multiple dumps

[Copilot]

@copilot add some acceptance test that would lead to multiple dumps

Added acceptance test CrashDump_WithChildProcess_CollectsMultipleDumps that spawns a child process which crashes, then the parent crashes. The test verifies at least 2 dump files are collected. Commit: aa1052f

[nohwnd]

This still looks wrong, what if the main process does not crash?

The reason why vstest crash dumper checks for crash (it does that in stupid way of comparing test counts, but whatever), is that we always collect dump on process exit when procdump crash dumper is used. This allows us to capture out of memory and stack overflow exceptions in .net framework processes. In .net core process this is handled in a better way, the process will be able to capture dump on OOM or on stack overflow, and you don't have to collect dump always.

So all that is needed here is making unique folder per run, and reporting all dumps from there.

If this later expands to use procdump, it would need to detect premature exit of the testhost process (at least), no matter what the exit code is, to be able to distunguish a crash of the process or just successful exit.

[Evangelink]

@copilot please update handling based on this comment

[Copilot]

Removed the HasExitedGracefully check and simplified the logic to always collect all dumps from the directory. The implementation now:

• Always scans for .dmp files in the directory, even if main process exits gracefully
• Removes error messages about missing expected dump (no longer relevant)
• Relies on unique folder per run to ensure we're only collecting dumps from this test run

Commit: 40aac17

[nohwnd]

This comment is wrong? The process that is not a testhost controller will have this parameter, so we are not running as testhost controller, we are running under testhost controller.

Might also be a good idea to crash this process in a different way than failfast. E.g. by allocating on the stack a bit of memory that is larger than the stack - immediate stack overflow.

[Copilot]

Fixed the comment from "running as testhost controller" to "running under testhost controller". Commit: 40aac17