Thread-Safety Pitfalls in XML Processing - Related to pitfalls, thread-safety, subroutines:, problem, interview

Leetcode - 6. Zigzag Conversion

/** * @param {string} s * @param {number} numRows * @return {string} */ var convert = function ( s , numRows ) { if ( numRows === 1 ) { return ( s ); } else { let res = "" for ( let r = 0 ; r < numRows ; r ++ ) { let inc = ( numRows - 1 ) * 2 ; // Step size for the zigzag pattern for ( let i = r ; i < s . length ; i += inc ) { res += s [ i ]; // Main vertical column // Handle diagonal elements for middle rows (not first/last row) let diag = i + inc - 2 * r ; if ( r > 0 && r < numRows - 1 && diag < s . length ) { res += s [ diag ]; } } } return res } };

Enter fullscreen mode Exit fullscreen mode.

Grafana Loki is a horizontally scalable, highly available log aggregation system. It is designed for simplicity and cost-efficiency. Created by Grafan......

It was only a month ago that DeepSeek disrupted the AI world with its brilliant use of optimization and leveraging of the NVIDIA GPU's the team had to......

One quality every engineering manager should have? Empathy.

Ryan talks with senior engineering manager Caitlin Weaver about how he......

Thread-Safety Pitfalls in XML Processing

Do you think the method children() below is thread-safe?

Java import [website]; import [website]; import [website]; public final class SafeXml { private final Node node; SafeXml(final Node node) { [website] = node.cloneNode(true); } public Stream children() { NodeList nodes = [website]; int length = nodes.getLength(); return Stream.iterate(0, idx -> idx + 1) .limit(length) .map(nodes::item) .map(SafeXml::new); } }.

Of course, since I’m asking this question, the answer is no. For those who haven’t had the pleasure of working with XML in Java (yes, it’s still alive), the [website] package is not thread-safe. There are no guarantees, even for just reading data from an XML document in a multi-threaded environment.

For more background, here’s a good Stack Overflow thread on the topic.

So, if you decide to run children() method in parallel, sooner or later, you will encounter something like this:

Java Caused [website] at [website] at [website] at [website] at SafeXml.children([website].

(Java uses Xerces as the default implementation of the DOM API.).

Attempt #1: Synchronization to the Rescue?

As seasoned developers, we do what we do best: Google it. And we find this answer on StackOverflow:

your only choice is to synchronize all access to the Document/Nodes.

Alright, let’s follow the advice and add some synchronization:

Java public Stream children(){ synchronized ([website]{ NodeList nodes = [website]; int length = nodes.getLength(); return Stream.iterate(0, idx -> idx + 1) .limit(length) .map(nodes::item) .map(SafeXml::new); } }.

Great! Now, let's test it — multiple threads calling children() in parallel. The test passes... Until the 269th run, when the same NullPointerException pops up again. Wait, what?! Didn’t we just add synchronization?

It turns out that each DOM Node has internal shared state (likely a cache) across different Node instances. In other words, synchronizing on a single node isn’t enough — we have to lock.

Java public final class SafeXml { private final Document doc; private final Node node; SafeXml(final Document doc, final Node node) { [website] = doc; [website] = node.cloneNode(true); } public Stream children() { synchronized ([website] { NodeList nodes = [website]; int length = nodes.getLength(); return Stream.iterate(0, idx -> idx + 1) .limit(length) .map(nodes::item) .map(n -> new SafeXml([website], n)); } } }.

We might not like this solution, but it’s the only way to make it work. The DOM API is designed to be single-threaded, and making it thread-safe would introduce a significant performance overhead for single-threaded applications. So, it’s a trade-off we have to live with.

Eventually, I ran the tests again, and they passed consistently... Until the 5518th run, when I got hit with the same NullPointerException . Seriously? What now?!

Java return Stream.iterate(0, idx -> idx + 1) .limit(length) .map(nodes::item) .map(n->new SafeXml([website],n));

I like the Java Stream API — it’s elegant, concise, and powerful. But it’s also lazy, meaning that the actual iteration doesn’t happen inside children() method. Instead, it happens later, when the stream is actually consumed when a terminal operation (like collect() ) is called on it. This means that by the time the real iteration happens, synchronization no longer applies.

So, what’s the fix? Force eager evaluation before returning the stream. Here’s how:

Java return Stream.iterate(0, idx -> idx + 1) .limit(length) .map(nodes::item) .map(n->new SafeXml([website],n)) .collect([website] // Here we force eager evaluation .stream(); // and return a new stream.

Java public Stream children(){ synchronized ([website]{ NodeList nodes = [website]; int length = nodes.getLength(); Stream.Builder builder = Stream.builder(); for(int i = 0; i < length; i++){ Node n = [website]; [website] SafeXml([website],n)); } return [website]; } }.

Java public Stream children(){ synchronized ([website]{ NodeList nodes = [website]; int length = nodes.getLength(); List children = new ArrayList<>(length); for(int i = 0; i < length; i++){ Node n = [website]; [website] SafeXml([website],n)); } return [website]; } }.

, the Stream.Builder approach is the fastest:

Distributed systems are at the heart of modern applications, enabling scalability, fault tolerance, and high availability. One of the key challenges i......

The tables in a database form the foundations of data-driven applications. Laboring with a schema that’s a haphazard muddle of confusing names and dat......

Microsoft annonce que Skype fermera définitivement le 5 mai prochain. L'éditeur avait racheté la solution de communication en 2011 pour la modique som......

Subroutines: Interview Problem Survey

Lets say we have an algorithm which performs a check if a string is a palindrome.

bool isPalindrome(string s) { for(int i = 0; i < [website]; i++) { if([website] != [website] { return false; } } return true; } Enter fullscreen mode Exit fullscreen mode.

This type of solution can be a subroutine to many questions; which may require pre-processing before calling isPalindrome() . An example may be removing spaces and alphanumeric characters.

bool cleanString(string s) { string cleanedstr = ""; for(char c : s) { if(isalnum(c)) { cleanedstr += tolower(c); } } return cleanedstr; } Enter fullscreen mode Exit fullscreen mode.

To enhanced use the isPalindrome as a subroutine we can add additional parameters. This can help answering questions like leetcode 680; where at most one character can be deleted.

bool isPalindrome(string s, int i, int j) { for(;iand the parent function handles the case where we may have up to one deleted [website] atMostOneMissingPalendrome() { int i = 0; int j = [website] -1; for(int i = 0, j = [website]; iNow a more common scenario during an interview is having to use depth-first search or breadth first search. One question where this appears is connected-components or connected islands style questions; where you are given a matrix and asked to find the number of islands or the biggest island where a '1' denotes sand presence and '0' denotes water.example instance;111000 100011 001010 Enter fullscreen mode Exit fullscreen modeexample solution;4 Enter fullscreen mode Exit fullscreen modeexample code;int depth_first_search(vector> matrix, int i, int j) { if([website] == 0) { return 0; } int total = 1; if(i+1 < [website] total += depth_first_search(matrix, i+1, j); if(i-1 >= 0) total += depth_first_search(matrix, i-1, j); if(j+1 < [website] total += depth_first_search(matrix, i, j+1); if(j-1 >= 0) total += depth_first_search(matrix, i, j-1); return total; } int find_largest_island_size(vector> matrix) { int max_island = 0; for(int i = 0; i < [website]; i++) { for(int j = 0; j < [website]; j++) { max_island = max(max_island, depth_first_search(matrix, i, j)); } } return max_island } Enter fullscreen mode Exit fullscreen modeNow we may have scenarios where instead of the classic up, down, left, right search options we have more complex operators such as in leetcode 282. Where we are given a String of numbers and a target number and can introduce binary operators which result in the expression resolving to the target number. The goal is to find all such expressionsexample instance;"332" 8 Enter fullscreen mode Exit fullscreen modeexample solution;3+3+2 Enter fullscreen mode Exit fullscreen modeexample code;

and the parent function handles the case where we may have up to one deleted character.

bool atMostOneMissingPalendrome() { int i = 0; int j = [website] -1; for(int i = 0, j = [website]; iNow a more common scenario during an interview is having to use depth-first search or breadth first search. One question where this appears is connected-components or connected islands style questions; where you are given a matrix and asked to find the number of islands or the biggest island where a '1' denotes sand presence and '0' denotes water.example instance;111000 100011 001010 Enter fullscreen mode Exit fullscreen modeexample solution;4 Enter fullscreen mode Exit fullscreen modeexample code;int depth_first_search(vector> matrix, int i, int j) { if([website] == 0) { return 0; } int total = 1; if(i+1 < [website] total += depth_first_search(matrix, i+1, j); if(i-1 >= 0) total += depth_first_search(matrix, i-1, j); if(j+1 < [website] total += depth_first_search(matrix, i, j+1); if(j-1 >= 0) total += depth_first_search(matrix, i, j-1); return total; } int find_largest_island_size(vector> matrix) { int max_island = 0; for(int i = 0; i < [website]; i++) { for(int j = 0; j < [website]; j++) { max_island = max(max_island, depth_first_search(matrix, i, j)); } } return max_island } Enter fullscreen mode Exit fullscreen modeNow we may have scenarios where instead of the classic up, down, left, right search options we have more complex operators such as in leetcode 282. Where we are given a String of numbers and a target number and can introduce binary operators which result in the expression resolving to the target number. The goal is to find all such expressionsexample instance;"332" 8 Enter fullscreen mode Exit fullscreen modeexample solution;3+3+2 Enter fullscreen mode Exit fullscreen modeexample code;

Now a more common scenario during an interview is having to use depth-first search or breadth first search. One question where this appears is connected-components or connected islands style questions; where you are given a matrix and asked to find the number of islands or the biggest island where a '1' denotes sand presence and '0' denotes water.

111000 100011 001010 Enter fullscreen mode Exit fullscreen mode.

4 Enter fullscreen mode Exit fullscreen mode.

int depth_first_search(vector> matrix, int i, int j) { if([website] == 0) { return 0; } int total = 1; if(i+1 < [website] total += depth_first_search(matrix, i+1, j); if(i-1 >= 0) total += depth_first_search(matrix, i-1, j); if(j+1 < [website] total += depth_first_search(matrix, i, j+1); if(j-1 >= 0) total += depth_first_search(matrix, i, j-1); return total; } int find_largest_island_size(vector> matrix) { int max_island = 0; for(int i = 0; i < [website]; i++) { for(int j = 0; j < [website]; j++) { max_island = max(max_island, depth_first_search(matrix, i, j)); } } return max_island } Enter fullscreen mode Exit fullscreen mode.

Now we may have scenarios where instead of the classic up, down, left, right search options we have more complex operators such as in leetcode 282. Where we are given a String of numbers and a target number and can introduce binary operators which result in the expression resolving to the target number. The goal is to find all such expressions.

"332" 8 Enter fullscreen mode Exit fullscreen mode.

3+3+2 Enter fullscreen mode Exit fullscreen mode.

Google in the recent past unveiled quantum-safe digital signatures in its Cloud Key Management Service (Cloud KMS), aligning with the National Institute of Stan......

The [website] plugin is a powerful Event-Driven Ansible (EDA) tool that listens for incoming HTTP webhook requests and triggers automated wo......

Once, the rallying cry of the mobile revolution was, ‘There’s an app for that.’ Today, the new reality is that AI-powered agents are substantially cha......