How to solve C/C++ memory leaking (in Linux)?


My hobby project Med is written in C++. A lot of implementations need to use dynamic memory allocation and instantiation. Because complex data is impractical to be passed by value, like the case of JavaScript object and array. Since C++ doesn’t have garbage collection, it is possible that the developer doesn’t free/delete the dynamically created memory properly.

As in my project Med, the program will retrieve the memory from other process. That means, it needs to make a copy of the scanned memory. And this will involve creating dynamic memory (using new operator). When using the program to filter the memory to get the target result, it needs to get a new copy of memory with the updated values, then compare with the previous copy. The unmatched will need to be discarded (free/delete); the matched will need to replace the old one (also free/delete the old one, because the new one is dynamically allocated).

Though it can be easily expressed in verbal form, writing the algorithm is usually inducing some human mistakes, such as using wrong variable name. As a result, we may access the memory that is not dynamically allocated, or freeing the memory twice. These will cause segmentation fault, and C++ compiler will not tell you where the error comes from. Debugging is hellish in this case. As a result, I used several solutions to fix memory leaking.

Adopt TDD

I have covered this in my previous post. Smaller function is easier to be tested. Just make sure your functions or methods are testable.

Valgrind (Linux only)

Compile your application with debugging information (g++ with -g option). Then run your test suite with valgrind.

valgrind --leak-check=yes ./testSnapshot

It will tell you which function access the non-allocated memory, how many bytes are untracked, and so on.

Valgrind is super useful to check memory leaking.

Memory Manager

Because I wrote the code with some mistakes, so the memory is freed twice and causes the error. But I failed to find where is the cause (which finally I found it).

Therefore, I wrote a memory manager to make sure the memory will not be freed more than once. However, this is actually not a good solution to avoid memory leaking. It just prevents the program to free the memory twice.

ByteManager::ByteManager() {}

ByteManager::~ByteManager() {
  clear();
}

// Singleton
ByteManager& ByteManager::getInstance() {
  static ByteManager instance;
  return instance;
}

Byte* ByteManager::newByte(int size) {
  Byte* byte = new Byte[size];
  recordedBytes[byte] = byte;
  return byte;
}

void ByteManager::deleteByte(Byte* byte) {
  auto found = recordedBytes.find(byte);
  if (found != recordedBytes.end()) {
    delete[] recordedBytes[byte];
    recordedBytes.erase(found);
  }
}

void ByteManager::clear() {
  for (auto it = recordedBytes.begin(); it != recordedBytes.end(); ++it) {
    delete[] it->second;
  }
  recordedBytes.clear();
}

map ByteManager::getRecordedBytes() {
  return recordedBytes;
}

Then, I refactored both new operator and delete operator to be replaced by newByte() and deleteByte() methods. And I use a std::map to store all the created memory.

Smart pointer

C++11 introduced smart pointers shared_ptr and unique_ptr. unique_ptr can be replaced by auto_ptr before C++11. By using smart pointer, we can use new operator to instantiate an object, then we need not to delete it. Because it will be deleted automatically when it lost the reference. For example,

typedef std::shared_ptr<SnapshotScan> SnapshotScanPtr;

// some where else
SnapshotScanPtr matched = SnapshotScanPtr(new SnapshotScan(curr.getAddress() + i + currOffset, scanType));

So, program will instantiate a new SnapshotScan object as the shared_ptr. Then it will be stored in a std::vector. When it lost the reference, such as removed from the std::vector, it will be deleted automatically.

In my opinion, it is a better solution than the Memory Manager above. However, my existing project can be hardly refactored to use smart pointer.

Advertisements

C++ Unit Test and Dependency Injection


TDD (test driven development) is widely adopted in modern development such as web development. Because it allows the developers to test the solution robustly in order to produce a more stable product.

Higher level programming languages like JavaScript and Ruby allows the developers to easily mock the functions and data to test the target specification. However, programming language like C++ is not designed for TDD. It will be more complex if you want to mock functions.

In order to adopt TDD, we need to write the function as small as possible, so that it can be easily mocked. As a result, the design of the object class needs to be testable. The methods that we are going to test needs to be publicly accessible.

Unlike JavaScript, C++ is prototype-based programming language. JavaScript can easily be mocked by overriding methods. In order to mock the C++ class, I have to use inheritance and mock the public methods. Example from my project Med.

class SnapshotScanTester : public SnapshotScan {
public:
  SnapshotScanTester() : SnapshotScan() {
    scannedValue = NULL;
  }
  virtual Bytes* getValueAsNewBytes(long pid, ScanType scanType) {
    Byte* data = bm.newByte(8);
    memset(data, 0, 8);
    data[0] = 20;
    Bytes* bytes = new Bytes(data, 8);
    return bytes;
  }
};

In order to mock the class SnapshotScan, I need to make the getValueAsNewBytes method as virtual public. Then using CxxTest framework to test with SnapshotScanTester.

Dependency Injection

I learnt the term Dependency Injection when I was developing with AngularJS project. By using dependency injection solution, we can create the services and inject to the client (object).

Med project is complex. The main feature of Med is to scan the memory of other process. Mocking other process is impractical for the test. In order to solve this, I refactored the code that involves PID (process ID) as the parameter, and write them into a service. For example,

class SnapshotScanService {
public:
  SnapshotScanService();
  virtual ~SnapshotScanService();

  virtual bool compareScan(SnapshotScan* scan, long pid, const ScanParser::OpType& opType, const ScanType& scanType);

  virtual void updateScannedValue(SnapshotScan* scan, long pid, const ScanType& scanType);
};

Before the refactoring, my code is doing something like

snapshotScan.compareScan(pid, opType, scanType);

By refactoring them into a service, then I can mock the service to produce any value for the testing.

In order to inject the service, I wrote the constructor that can accept the service,

class Snapshot {
public:
  Snapshot(SnapshotScanService* service = NULL);
  // ...
};

If service is not provided to the constructor, then default service will be instantiated. Therefore, if I want to mock the service, I just create a new class derived from SnapshotScanService. For example,

class SnapshotScanServiceTester : public SnapshotScanService {
public:
  SnapshotScanServiceTester() {}

  virtual bool compareScan(SnapshotScan*, long, const ScanParser::OpType&, const ScanType&) {
    return true;
  }
  virtual void updateScannedValue(SnapshotScan* scan, long pid, const ScanType& scanType) {
    Bytes* currentBytes = Bytes::create(20);
    currentBytes->getData()[0] = 60;
    scan->freeScannedValue();
    scan->setScannedValue(currentBytes);
  }
};

Then in the test,

    SnapshotScanService* service = new SnapshotScanServiceTester();
    SnapshotTester* snapshot = new SnapshotTester(service);

So, by using the dependency injection, I can finally test my class properly to make sure its functionality reliable. Without unit test, meaning I need to test my code by running thousands of time to create different situations. I am not genius, I don’t think that my code without proper test can function properly.

Firefox Legacy version 56.0.2


The latest Firefox version 57 and above, a.k.a Firefox Quantum, it is fast, but… that is not what I need.

As a developer, I favoured Chromium more than Firefox. And I use Firefox mainly for downloading. The addon DownThemAll is the must. The greatest feature I love is the ability to highlight and download the selected hyperlinks as batch. And I can name the downloaded files by original filename or based on the text in HTML.

DownThemAll in Firefox 56
DownThemAll in Firefox 56

This is neither Google Chrome nor Firefox Quantum can do.

With Firefox Quantum, there is no more support for legacy addons. As a result, unless the addon developers update their project, nothing more we can do. We can either install other Firefox variants such as Firefox ESR, Waterfox, Palemoon, etc (as suggested by one of the DTA reviewers). In my opinion, with WebExtensions API, DTA developer may face some issues to develop for Quantum support.

I tried to install Firefox ESR (which is actually 52.5.0) in order to use DTA. However, it is extremely slower comparing to Firefox 56. When Firefox 54 was released, it was the best Firefox ever and it was really fast comparing to the older versions. Therefore, in order to use the latest Firefox that supports legacy addon, I choose to install Firefox 56 without removing the latest Firefox Quantum.

How to install older Firefox (for Linux only)?

  1. Use Google to search an older Firefox, you should get a URL like this: https://ftp.mozilla.org/pub/firefox/releases/56.0.2/
  2. Choose the link (directory) that matches your target machine, Windows 32-bit / 64-bit, Linux 32-bit / 64-bit, etc. Then choose a language that you prefer. For my case, I downloaded Firefox 56.0.2 for Linux 64-bit.
  3. Download and extract it to some location, such as /opt/firefox56.

Now, you should able to invoke Firefox 56 by (but don’t run it now)

/opt/firefox56/firefox

Run the above command will use the same profile as your default Firefox (if you installed Firefox Quantum).

Create another profile

Your Firefox profile should locate at ~/.mozilla/firefox/, named as xxxxx.profilename.

(NOTE: Suggest to make a backup of your default profile.)

Easiest way to create another profile for another version of Firefox,

  1. Create a new profile using command firefox -ProfileManager. Because this will update profile.ini as well.
  2. Exit Firefox Profile Manager.
  3. Locate to the directory of the profile, then delete all the files and directories within the profile, i.e, make it empty. (NOTE: Do not delete your default profile.)
  4. Copy all the files and directories of your default profile to the empty profile in step 3.
  5. Now you can run Firefox with the new profile, like /opt/firefox56/firefox -P profilename.

Create shortcut (application menu, aka desktop entries, for Linux only)

Because the Firefox we installed is just by extracting. There is no desktop entry file created. Therefore, we need to create our own desktop entry file. This can be done by creating a file as ~/.local/shared/applications/firefox-legacy.desktop.

The following is the content,

Name=Firefox Legacy
GenericName=Web Browser
Icon=firefox
Type=Application
Categories=Application;Network;
MimeType=text/html;text/xml;application/xhtml+xml;application/vnd.mozilla.xul+xml;text/mml;x-scheme-handler/http;x-scheme-handler/https;
Encoding=UTF-8
Exec=/opt/firefox56/firefox -P legacy %u
Terminal=false
MultipleArgs=false
StartupNotify=false
StartupWMClass=Firefox

where the /opt/firefox56/firefox -P legacy %u, with the legacy as the profile name. You can modify this line to your target profile name.

Then, restart your gnome-panel or whatever panel, it should generate this menu item named as Firefox Legacy. Now, you can run two different versions of Firefox simultaneously.

 

(Side note: WordPress with Markdown mode enabled is the best!)