Fix For Collection Was Modified And Enumeration Not Executed In .NET

Introduction

In .NET, this collection-modified exception can happen in dictionaries when there are two tasks. One iterating through the dictionary and another is adding or removing items from the dictionary. The tasks that are looping through the dictionary will throw this exception. The dictionary is not a thread-safe collection type but there are some ways around this issue by copying the parts of the dictionary to another collection type. Let's look at those in detail.

Motivation

How does this issue appear? The dictionary can be read by multiple threads but if there's a write operation by adding or removing items from the dictionary then it can not at the same time be in a loop. You could add locks so that only the thread is working with the dictionary at a time but for now we're going to assume that isn't the case.

Collection Was Modified And Enumeration Operation May Not Execute Exception Example Code

This code has two dictionaries, one is the primary dictionary and the secondary dictionary. There are two threads. One thread will be looping through the dictionary and printing values from the primary dictionary. The other thread will be looping through the secondary dictionary and add those items from the secondary to the primary dictionary. This is the write operation.

Each dictionary has a set of countries and populations as the key and values. Let's look at the code.

Dictionary<string, double> primaryDictionary = new Dictionary<string, double>()
{
    { "China", 1393.78 },
    { "India", 1369.87 },
    { "United States", 329.74 },
    { "Indonesia", 269.54 },
    { "Pakistan", 220.89 },
    { "Brazil", 213.99 },
};//Initialize dictionary

Dictionary<string, double> secondaryDictionary = new Dictionary<string, double>()
{
   { "Mexico", 126.58 },
    { "Japan", 125.96 },
    { "Philippines", 108.11 },
    { "Egypt", 101.33 },
    { "Ethiopia", 109.22 },
};//Initialize dictionary

PrintDictionaryAsync();//Start reader task
await AddToDictionaryAsync();//Start writer tasks


async Task PrintDictionaryAsync()
{
    try
    {
        foreach (string country in primaryDictionary.Keys)//Loop through the primary dictionary keys
        {
            await Task.Delay(400);//Add slight delay so that the writer will modify the dictionary while in the loop
            Console.WriteLine($"key:{country}");//Pring dictionary key
        }
    }
    catch (Exception ex)
    {
        Console.WriteLine(ex.Message);//Write exception to screen
    }
    return;
}

async Task AddToDictionaryAsync()
{
    foreach (KeyValuePair<string, double> countryKeyValue in secondaryDictionary)//Loop through the secondary dictionary key value pairs
    {
        await Task.Delay(500);//Add slight delay
        primaryDictionary.Add(countryKeyValue.Key, countryKeyValue.Value);//Add secondary key-value pair to the primary dictionary. This is a write operation
        Console.WriteLine("Add Country To List:" + countryKeyValue.Key);//Print a message to the screen saying add successfully
    }
    return;
}

Code Output

Reader Thread:key:China
Writer Thread:Add Country To List:Mexico
Reader Thread:key:India
Reader Thread:Collection was modified; enumeration operation may not execute.
Writer Thread:Add Country To List:Japan
Writer Thread:Add Country To List:Philippines
Writer Thread:Add Country To List:Egypt
Writer Thread:Add Country To List:Ethiopia

As this run, the reader thread that is looping through the primary dictionary will throw an exception after the writer thread has added an entry into the primary dictionary. After that, you can see that the reader thread stops. But the writer thread continues to add entries into the dictionary.

LINQ ToList Solution

Now that we have an understanding of the problem. The first solution is to copy the dictionary's keys to a list and then loop through the list. This works because even if the dictionary is modified we're not looping through the dictionary anymore but a copy of the keys.

LINQ ToList Example Code

The code change is happening with LINQ's ToList being applied to the keys and it returns a list of strings. This could be applied to values as well.

async Task PrintDictionaryAsync()
{
    try
    {
        List<string> list = primaryDictionary.Keys.ToList();//Copy dictionary keys to a list
        foreach (string country in list)//Loop through the list which is a copy of the dictionary keys
        {
            await Task.Delay(400);//Add slight delay so that the writer will modify the dictionary while in the loop
            Console.WriteLine($"Reader Thread:key:{country}");//Pring dictionary key
        }
    }
    catch (Exception ex)
    {
        Console.WriteLine("Reader Thread:" + ex.Message);//Write exception to screen
    }
    return;
}

Code Output

Reader Thread:key:China
Writer Thread:Add Country To List:Mexico
Reader Thread:key:India
Writer Thread:Add Country To List:Japan
Reader Thread:key:United States
Writer Thread:Add Country To List:Philippines
Reader Thread:key:Indonesia
Writer Thread:Add Country To List:Egypt
Reader Thread:key:Pakistan
Reader Thread:key:Brazil
Writer Thread:Add Country To List:Ethiopia

No exceptions are being thrown in this method now.

LINQ ToList Speed Test

What I like to do is test the performance of the method and how they compare with other methods. I will do an average of 10 tests. Each test will have a loop that calls the test method 100 times. Then each test method will have 1 million objects in the dictionary to work with.

LINQ ToList Speed Test Code

The test method takes the dictionary of 1 million objects and performs copies of the keys to the list.

void TestMethod(Dictionary<string, string> dictionary)
{
    List<string> list = dictionary.Keys.ToList();//Create a new List using LINQ ToList based on the Keys
}

Code Output

Test 1:Function Calls:100, In 0m 1s 41ms
Test 2:Function Calls:100, In 0m 0s 981ms
Test 3:Function Calls:100, In 0m 1s 44ms
Test 4:Function Calls:100, In 0m 0s 959ms
Test 5:Function Calls:100, In 0m 0s 976ms
Test 6:Function Calls:100, In 0m 0s 845ms
Test 7:Function Calls:100, In 0m 0s 958ms
Test 8:Function Calls:100, In 0m 0s 964ms
Test 9:Function Calls:100, In 0m 0s 849ms
Test 10:Function Calls:100, In 0m 0s 946ms
LINQ ToList Average Speed:957ms, In 10 Tests

This test was completed in just under a second which is pretty good. Let's see how it compares with the other methods.

using System.Diagnostics;

int numberOfTests = 10;//Number of tests 
int numberOfFunctionCalls = 100;//Number of function calls made per test
int numberOfObjectsToCreate = 1000000;//Number test objects
int lengthOfRandomString = 50;
string testName = "LINQ ToList";//Test name to print to average
List<string> keysList = new List<string>();
void TestMethod(Dictionary<string, string> dictionary)
{
    List<string> list = dictionary.Keys.ToList();//Create a new List using LINQ ToList based on the Keys
}

List<double> testSpeedList = new List<double>();
for (int testIndex = 0; testIndex < numberOfTests; testIndex++)
{
    testSpeedList.Add(StartTest(testIndex));
}
Console.WriteLine($"{testName} Average Speed:{Math.Round(testSpeedList.Average())}ms, In {numberOfTests} Tests");

double StartTest(int testIndex)
{
    Stopwatch stopwatch = new Stopwatch();
    Dictionary<string, string> testData = GetDictionaryData();//Get intial random generated data
    for (int i = 0; i < numberOfFunctionCalls; i++)
    {
        stopwatch.Start();//Start the Stopwatch timer
        TestMethod(testData);//
        stopwatch.Stop();//Stop the Stopwatch timer
    }
    stopwatch.Stop();//Stop the Stopwatch timer
    Console.WriteLine($"Test {testIndex + 1}:Function Calls:{numberOfFunctionCalls}, In {stopwatch.Elapsed.Minutes}m {stopwatch.Elapsed.Seconds}s {stopwatch.Elapsed.Milliseconds}ms");
    return stopwatch.Elapsed.TotalMilliseconds;
}

Dictionary<string, string> GetDictionaryData()
{
    Dictionary<string, string> testData = new Dictionary<string, string>(StringComparer.OrdinalIgnoreCase);
    keysList = new List<string>();
    for (int i = 0; i < numberOfObjectsToCreate; i++)
    {
        string key = GenerateRandomString(lengthOfRandomString);
        if (!testData.ContainsKey(key))
        {
            string value = GenerateRandomString(lengthOfRandomString);
            testData[key] = value;
            keysList.Add(key);
        }
    }
    return testData;
}


string GenerateRandomString(int length)
{
    Random random = new Random();
    const string chars = "ABCDEFGHIJKLMNOPQRSTUVWXYZabcdefghijklmnopqrstuvwxyz";
    return new string(Enumerable.Repeat(chars, length)
      .Select(s => s[random.Next(s.Length)]).ToArray());
}

List Constructor Solution

The second solution that we can use is to pass the dictionary keys or values to the list constructor.

List Constructor Example Code

We can directly call the list constructor and pass the keys directly to it. This makes a copy of the keys and since it's a copy, it isn't being modified.

async Task PrintDictionaryAsync()
{
    try
    {
        List<string> list = new List<string>(primaryDictionary.Keys);//Copy dictionary keys to a list through the list constructor
        foreach (string country in list)//Loop through list which is copy of the dictionary keys
        {
            await Task.Delay(400);//Add slight delay so that the writer will modify the dictionary while in the loop
            Console.WriteLine($"Reader Thread:key:{country}");//Pring dictionary key
        }
    }
    catch (Exception ex)
    {
        Console.WriteLine("Reader Thread:" + ex.Message);//Write exception to screen
    }
    return;
}

Code Output

Reader Thread:key:China
Writer Thread:Add Country To List:Mexico
Reader Thread:key:India
Writer Thread:Add Country To List:Japan
Reader Thread:key:United States
Writer Thread:Add Country To List:Philippines
Reader Thread:key:Indonesia
Writer Thread:Add Country To List:Egypt
Reader Thread:key:Pakistan
Reader Thread:key:Brazil
Writer Thread:Add Country To List:Ethiopia

List Constructor Speed Test

This is the same test as before and the test parameters are below.

List Constructor Speed Test Code

void TestMethod(Dictionary<string, string> dictionary)
{
    List<string> list = new List<string>(dictionary.Keys);//Create a new List using List Constructor based on the Keys
}

Code Output

Test 1:Function Calls:100, In 0m 1s 35ms
Test 2:Function Calls:100, In 0m 1s 13ms
Test 3:Function Calls:100, In 0m 1s 11ms
Test 4:Function Calls:100, In 0m 0s 999ms
Test 5:Function Calls:100, In 0m 0s 976ms
Test 6:Function Calls:100, In 0m 0s 873ms
Test 7:Function Calls:100, In 0m 1s 16ms
Test 8:Function Calls:100, In 0m 0s 982ms
Test 9:Function Calls:100, In 0m 0s 962ms
Test 10:Function Calls:100, In 0m 1s 28ms
List Constructor Average Speed:990ms, In 10 Tests

This speed is close to LINQ's ToList method that they are practically the same.

using System.Diagnostics;

int numberOfTests = 10;//Number of tests 
int numberOfFunctionCalls = 100;//Number of function calls made per test
int numberOfObjectsToCreate = 1000000;//Number test objects
int lengthOfRandomString = 50;
string testName = "List Constructor";//Test name to print to average
List<string> keysList = new List<string>();
void TestMethod(Dictionary<string, string> dictionary)
{
    List<string> list = new List<string>(dictionary.Keys);//Create a new List using List Constructor based on the Keys
}

List<double> testSpeedList = new List<double>();
for (int testIndex = 0; testIndex < numberOfTests; testIndex++)
{
    testSpeedList.Add(StartTest(testIndex));
}
Console.WriteLine($"{testName} Average Speed:{Math.Round(testSpeedList.Average())}ms, In {numberOfTests} Tests");

double StartTest(int testIndex)
{
    Stopwatch stopwatch = new Stopwatch();
    Dictionary<string, string> testData = GetDictionaryData();//Get intial random generated data
    for (int i = 0; i < numberOfFunctionCalls; i++)
    {
        stopwatch.Start();//Start the Stopwatch timer
        TestMethod(testData);//
        stopwatch.Stop();//Stop the Stopwatch timer
    }
    stopwatch.Stop();//Stop the Stopwatch timer
    Console.WriteLine($"Test {testIndex + 1}:Function Calls:{numberOfFunctionCalls}, In {stopwatch.Elapsed.Minutes}m {stopwatch.Elapsed.Seconds}s {stopwatch.Elapsed.Milliseconds}ms");
    return stopwatch.Elapsed.TotalMilliseconds;
}

Dictionary<string, string> GetDictionaryData()
{
    Dictionary<string, string> testData = new Dictionary<string, string>(StringComparer.OrdinalIgnoreCase);
    keysList = new List<string>();
    for (int i = 0; i < numberOfObjectsToCreate; i++)
    {
        string key = GenerateRandomString(lengthOfRandomString);
        if (!testData.ContainsKey(key))
        {
            string value = GenerateRandomString(lengthOfRandomString);
            testData[key] = value;
            keysList.Add(key);
        }
    }
    return testData;
}
string GenerateRandomString(int length)
{
    Random random = new Random();
    const string chars = "ABCDEFGHIJKLMNOPQRSTUVWXYZabcdefghijklmnopqrstuvwxyz";
    return new string(Enumerable.Repeat(chars, length)
      .Select(s => s[random.Next(s.Length)]).ToArray());
}

Dictionary Constructor Solution

But what if I want to loop through the keys and values at the same time? Need we'll need to copy the whole dictionary itself. We can do this using the dictionary constructor.

Dictionary Constructor Example Code

This code shows that we can pass my primary dictionary into the dictionary constructor to make a copy while in the reader thread.

async Task PrintDictionaryAsync()
{
    try
    {
        Dictionary<string, double> copyDictionary = new Dictionary<string,double>(primaryDictionary);//Copy dictionary keys and values to another dictionary
        foreach (KeyValuePair<string,double> countryKeyValuePair in copyDictionary)//Loop through the dictionary which is a copy of the dictionary keys and values
        {
            await Task.Delay(400);//Add slight delay so that the writer will modify the dictionary while in the loop
            Console.WriteLine($"Reader Thread:key:{countryKeyValuePair.Key}, value:{countryKeyValuePair.Value}");//Pring dictionary key and value
        }
    }
    catch (Exception ex)
    {
        Console.WriteLine("Reader Thread:" + ex.Message);//Write exception to screen
    }
    return;
}

Code Output

Reader Thread:key:China, value:1393.78
Writer Thread:Add Country To List:Mexico
Reader Thread:key:India, value:1369.87
Writer Thread:Add Country To List:Japan
Reader Thread:key:United States, value:329.74
Writer Thread:Add Country To List:Philippines
Reader Thread:key:Indonesia, value:269.54
Writer Thread:Add Country To List:Egypt
Reader Thread:key:Pakistan, value:220.89
Reader Thread:key:Brazil, value:213.99
Writer Thread:Add Country To List:Ethiopia

Dictionary Constructor Speed Test

This is the same test as before.

Dictionary Constructor Speed Test Code

void TestMethod(Dictionary<string, string> dictionary)
{
    Dictionary<string, string> dictionaryCopy = new Dictionary<string, string>(dictionary);//Create a new Dictionary copy using the dictionary constructor based on the keys and values
}

Code Output

Test 1:Function Calls:100, In 0m 8s 637ms
Test 2:Function Calls:100, In 0m 8s 476ms
Test 3:Function Calls:100, In 0m 9s 80ms
Test 4:Function Calls:100, In 0m 8s 933ms
Test 5:Function Calls:100, In 0m 9s 279ms
Test 6:Function Calls:100, In 0m 9s 317ms
Test 7:Function Calls:100, In 0m 9s 49ms
Test 8:Function Calls:100, In 0m 9s 148ms
Test 9:Function Calls:100, In 0m 8s 843ms
Test 10:Function Calls:100, In 0m 8s 996ms
Dictionary Constructor Average Speed:8976ms, In 10 Tests

As you can see this is a more expensive operation because both the keys and values are being copied.

using System.Diagnostics;

int numberOfTests = 10;//Number of tests 
int numberOfFunctionCalls = 100;//Number of function calls made per test
int numberOfObjectsToCreate = 1000000;//Number test objects
int lengthOfRandomString = 50;
string testName = "Dictionary Constructor";//Test name to print to average
List<string> keysList = new List<string>();
void TestMethod(Dictionary<string, string> dictionary)
{
    Dictionary<string, string> dictionaryCopy = new Dictionary<string, string>(dictionary);//Create a new Dictionary copy using the dictionary constructor based on the keys and values
}

List<double> testSpeedList = new List<double>();
for (int testIndex = 0; testIndex < numberOfTests; testIndex++)
{
    testSpeedList.Add(StartTest(testIndex));
}
Console.WriteLine($"{testName} Average Speed:{Math.Round(testSpeedList.Average())}ms, In {numberOfTests} Tests");

double StartTest(int testIndex)
{
    Stopwatch stopwatch = new Stopwatch();
    Dictionary<string, string> testData = GetDictionaryData();//Get intial random generated data
    for (int i = 0; i < numberOfFunctionCalls; i++)
    {
        stopwatch.Start();//Start the Stopwatch timer
        TestMethod(testData);//
        stopwatch.Stop();//Stop the Stopwatch timer
    }
    stopwatch.Stop();//Stop the Stopwatch timer
    Console.WriteLine($"Test {testIndex + 1}:Function Calls:{numberOfFunctionCalls}, In {stopwatch.Elapsed.Minutes}m {stopwatch.Elapsed.Seconds}s {stopwatch.Elapsed.Milliseconds}ms");
    return stopwatch.Elapsed.TotalMilliseconds;
}

Dictionary<string, string> GetDictionaryData()
{
    Dictionary<string, string> testData = new Dictionary<string, string>(StringComparer.OrdinalIgnoreCase);
    keysList = new List<string>();
    for (int i = 0; i < numberOfObjectsToCreate; i++)
    {
        string key = GenerateRandomString(lengthOfRandomString);
        if (!testData.ContainsKey(key))
        {
            string value = GenerateRandomString(lengthOfRandomString);
            testData[key] = value;
            keysList.Add(key);
        }
    }
    return testData;
}
string GenerateRandomString(int length)
{
    Random random = new Random();
    const string chars = "ABCDEFGHIJKLMNOPQRSTUVWXYZabcdefghijklmnopqrstuvwxyz";
    return new string(Enumerable.Repeat(chars, length)
      .Select(s => s[random.Next(s.Length)]).ToArray());
}

Conclusion

LINQ ToList has a compact syntax and it is easier to read and it is fast. I wouldn't consider it faster than the List constructor since they are so close in time. So if you prefer the list constructor you can use that as well it is still fast. These work for both the keys and values but separately.

If you need to loop through the key and values at the same time then you'll have no choice but use the dictionary constructor. But be aware that it is a more expensive operation than just looping through the keys and or values.

Know any other ways to fix this collection was modified exception? Let me know in the comments below.