Add ucm-toolkit for posix aio test&nic monitor and copy bandwidth test

[harrisonyhq]

Purpose

Add ucm-toolkit as a unified CLI entry for POSIX AIO testing, NIC monitoring, and copy bandwidth testing.

Modifications

• Added the ucm-toolkit package and CLI entry point.
• Integrated posix-aio, nic-monitor, and dev-sandbox tooling.
• Added documentation and install support for toolkit usage.
• Updated formatting and codespell configuration for toolkit sources.

Test

[ygwpz]

🔴 Critical: Memory allocation with new followed by instance.DoCopyBatch(...) without exception handling. If DoCopyBatch throws an exception, the HostCopyBuffer and DeviceCopyBuffer objects will not be deleted, causing memory leaks. Consider wrapping this in a try-catch block or using smart pointers (e.g., std::unique_ptr).

[ygwpz]

Follow-up review: Two concerns addressed, but critical memory safety issue remains.

[ygwpz]

Thank you for addressing the first two concerns. However, I notice that the critical memory safety issue (Comment #3) remains unaddressed.

The concern was about exception safety in toolkit/src/dev-sandbox/module/copy/ascend/copy_case_ascend.cc around lines 73-82:

for (size_t device = 0; device < ctx.nDevice; device++) {
    srcBuffers[device] = new HostCopyBuffer{device, ctx.size, ctx.num};
    dstBuffers[device] = new DeviceCopyBuffer{device, ctx.size, ctx.num};
}
H2DCECopyInstance instance{ctx.iter, false};
CopyResult result;
result.Push(instance.DoCopyBatch(srcBuffers, dstBuffers));  // ← If this throws
for (size_t device = 0; device < ctx.nDevice; device++) {
    delete srcBuffers[device];  // ← These cleanup lines will never execute
    delete dstBuffers[device];
}

If DoCopyBatch() throws an exception, the memory allocated with new will leak because the cleanup loop won't execute.

Suggested fix: Wrap the cleanup in a try-catch block or use smart pointers (e.g., std::unique_ptr):

std::vector<std::unique_ptr<HostCopyBuffer>> srcBuffers(ctx.nDevice);
std::vector<std::unique_ptr<DeviceCopyBuffer>> dstBuffers(ctx.nDevice);
for (size_t device = 0; device < ctx.nDevice; device++) {
    srcBuffers[device] = std::make_unique<HostCopyBuffer>(device, ctx.size, ctx.num);
    dstBuffers[device] = std::make_unique<DeviceCopyBuffer>(device, ctx.size, ctx.num);
}
H2DCECopyInstance instance{ctx.iter, false};
CopyResult result;
result.Push(instance.DoCopyBatch(srcBuffers, dstBuffers));
// Smart pointers handle cleanup automatically

Or use a try-catch:

try {
    result.Push(instance.DoCopyBatch(srcBuffers, dstBuffers));
} catch (...) {
    for (size_t device = 0; device < ctx.nDevice; device++) {
        delete srcBuffers[device];
        delete dstBuffers[device];
    }
    throw;
}

Please address this issue before the PR can be approved.

[ygwpz]

💡 Suggestion: Using unique_ptr for memory management is good. Verify that the smart pointer ownership transfer is correct, especially in multi-threaded contexts.

[ygwpz]

💡 Suggestion: Using unique_ptr for memory management is good. Verify that the smart pointer ownership transfer is correct, especially in multi-threaded contexts.

[ygwpz]

💡 Suggestion: Using unique_ptr for memory management is good. Verify that the smart pointer ownership transfer is correct, especially in multi-threaded contexts.