Ye Olde Mirror Effect!

I use a Mac normally. I have to go back to Windows from time to time. After Mac OS X Lion reverse-scrolling, switching between Mac and Windows became a pain. I started to use AutoHotKey on my Windows machines to simulate the reverse-scrolling from Lion. All was well. Until I remoted from one of my Windows machines into another Windows machine which also had AutoHotKey installed with reverse-scroll script enabled. Guess what, it went back to the original scroll direction. I couldn't figure out for a while what was happening, until I suddenly realized what had happened. I couldn't stop laughing at myself for a long time :)

(Of course the solution was to suspend AHK on the target machine when I was remoted in)

Windbg + SOS Cheat Sheet

This is a useful link for people who use windbg/cdb for debugging Managed code.

Cheat Sheet

Visual Studio debugging might not always be helpful or it might not always be available.
Sometimes when you want to take a quick look at what's wrong, saying c:\debuggers\cdb -pn helps a lot.

Reporting issues with Windows Azure

For those of you out there who have tried Windows Azure and face problems, here is how you are supposed to report issues at the Windows Azure forum:

http://social.msdn.microsoft.com/Forums/en-US/windowsazure/threads/

Steps you need to follow:

1. Sign into Windows Azure portal.
2. Go to your Hosted Service Project which is causing issues.
3. Wait for the project info to load.
4. Copy the private deployment id.
5. Write a brief description of what you did, what you are seeing and add your private deployment id.

Here is a screen shot of my service showing the private deployment id.

ThreadStaticAttribute

Reminder to self:

ThreadStaticAttribute only works with static fields!

Here is why:

public class TestingTLS
{
    [ThreadStatic]
    public static int value = 0;
    [ThreadStatic]
    public int value2 = 0;   
}

public class MyClass
{
    public static void TLSTest()
    {
        TestingTLS tls = new TestingTLS();
        Dictionary<int, Thread> threads = new Dictionary<int, Thread>();
        Semaphore sem = new Semaphore(100, 100);
        long count = 0;
        for ( int i = 0 ; i < 100 ; i ++ )
        {
            Thread t = new Thread(state => {
                int tid = Thread.CurrentThread.ManagedThreadId;
                TestingTLS.value = tid; 
                Thread.Sleep(0);
                if ( tid != TestingTLS.value )
                {
                    Console.Error.WriteLine( "Value 1: {0} read {1}", tid, TestingTLS.value );
                    Interlocked.Increment(ref count);
                }
                tls.value2 = tid;
                Thread.Sleep(0);
                if ( tid != tls.value2 )
                {
                    Console.Error.WriteLine( "Value 2: {0} read {1}", tid, tls.value2);
                    Interlocked.Increment(ref count);

                }
                threads.Remove(tid);
                if ( threads.Count == 0 )
                    sem.Release();
            });
            threads.Add(t.ManagedThreadId, t);
            sem.WaitOne();
            t.Start();   
        }
        Console.WriteLine("Waiting for all threads to finish...");
        sem.WaitOne();
        Console.WriteLine("Total races: " + count);
    }
}

You will see races for value2 but not for value.

Update: The code snippet was updated, since the previous example had a race where the Total Race count would reach before all threads have finished reporting the wrong number of total races in cases where there were large number of races.

First Azure Service is live!

This is my first Azure service that I wrote a couple of weeks ago. It only uses the front end roles and no storage. I will add other applications to this portal, which will use other services offered by Azure.

XmlTextReader fiasco

I hit a snag a few days ago while doing some Xml parsing in one of my applications.

I am using XDocument for Xml parsing(XLinq is beautiful (K)).

Doing XDocument.Load(urlString) or XmlTextReader() would make my application choke with an Xml parsing error (invalid character). Using StreamReader to wrap the actual stream (WebResponse.GetResponseStream) and passing it to XDocument works like a charm. I wrote a sample app to figure out the issue and it turns out that XmlTextReader is the culprit. Maybe it fails to correctly recognize UTF-8 markers, or maybe it's stricter than other Readers, it didn't work for me. So beware. :)

Code Snippet (fails):
WebRequest req = WebRequest.Create("");
WebResponse res = req.GetResponse();
XmlTextReader reader = new XmlTextReader(res.GetResponseStream());
XDocument xdoc = XDocument.Load(reader, LoadOptions.SetLineInfo); //LoadOptions.SetLineInfo helps with debugging.

Code Snippet (works):
WebRequest req = WebRequest.Create("");
WebResponse res = req.GetResponse();
StreamReader reader = new StreamReader(res.GetResponseStream());
XDocument xdoc = XDocument.Load(reader, LoadOptions.SetLineInfo); //LoadOptions.SetLineInfo helps with debugging.


Happy XLinq-ing.

A Custom SpinLock based CriticalSection Implementation

Just something I implemented for fun :)

#include "stdafx.h"
#include <intrin.h>
#include <process.h>
#include <vector>
#include <utility>
#include <algorithm>
#include <iostream>
#pragma intrinsic (_InterlockedCompareExchange, _InterlockedExchange)
namespace DotFermion
{
    class CCriticalSection
    {
        enum { LOCK_IS_FREE = 0, LOCK_IS_TAKEN = 1 };
    private:
        long lock;
        long ownerThreadId;
        long acquireCount;
    public:
        CCriticalSection():
          lock(LOCK_IS_FREE),ownerThreadId(-1),acquireCount(0){}
          ~CCriticalSection()
          {
              Release();
          }
          //Non-recursive Acquire. Once its called, all threads, even the one owning the lock block on this lock.
          //Even a subsequent AcquireRecursive call after this one blocks.
          void Acquire()
          {
              while (_InterlockedExchange(&lock, LOCK_IS_TAKEN) == LOCK_IS_TAKEN);
          }
          //Recursive Acquire. It doesn't block if called multiple times from the same thread.
          void AcquireRecursive(long threadId)
          {
              if(ownerThreadId != -1 && threadId == ownerThreadId)
              {
                  _InterlockedIncrement(&acquireCount);
                  return;
              }
              Acquire ();
              _InterlockedExchange(&ownerThreadId, threadId );
              _InterlockedIncrement(&acquireCount);
          }

          void Release()
          {
              if(acquireCount > 1)
              {
                  _InterlockedDecrement(&acquireCount);
                  return;
              }
              _InterlockedExchange(&acquireCount, 0);
              _InterlockedExchange(&ownerThreadId, -1);
              _InterlockedExchange(&lock, LOCK_IS_FREE);
          }
    };

}

#define LIST_POS = 4;
#define INDEX_POS = 8;

typedef void (*LPFNTHREADPROC) (void*);
typedef std::vector<uintptr_t> THREADLIST;
typedef THREADLIST* PTHREADLIST;
typedef struct _args_t
{
    DotFermion::CCriticalSection* pCs;
    PTHREADLIST pThreadList;
    int index;
} args_t;

THREADLIST g_ThreadList;
int g_ThreadCount = 0;
int g_SharedCounter = 0;

void ThreadProc(void* argList)
{
    args_t* args = (args_t*)argList;
    args->pCs->AcquireRecursive((long)args->pThreadList->at(args->index));
    g_SharedCounter++;
    args->pCs->AcquireRecursive((long)args->pThreadList->at(args->index));
    g_SharedCounter++;
    args->pCs->Release();
    args->pThreadList->at(args->index) = -1L;
    g_ThreadCount--;
    args->pCs->Release();
    delete args;
}

int StartThreads ( int n, PTHREADLIST pThreadList, DotFermion::CCriticalSection* cs )
{
    pThreadList->resize(n);
    for ( int i = 0 ; i < n ; i ++ )
    {
        args_t* args = new args_t;
        if ( args == NULL ) continue;
        args->pCs = cs;
        args->pThreadList = pThreadList;
        args->index = i;
        uintptr_t threadId = _beginthread(ThreadProc, 0, (void*)args);
        if ( threadId != -1L )
        {
            cs->Acquire();
            pThreadList->at(i) = threadId;
            g_ThreadCount ++;
            cs->Release();
        }
    }
    return g_ThreadCount;
}


void Join (int* threadCount)
{
    while ( *threadCount );
}

int _tmain(int argc, _TCHAR* argv[])
{
    DotFermion::CCriticalSection cs;
    StartThreads(100, &g_ThreadList, &cs);
    Join(&g_ThreadCount);
    std::cout << g_SharedCounter << std::endl;
    return 0;
}

My own GetWsdl tool

I finally figured it out. It was there right in front of my eyes. I stumbled upon it so many times, but I always missed it.

I have been doing web service development for quite some time now. These services are part of pretty complex systems that require a long time to deploy. Imagine, every time there was a build number change and I wanted to modify one of the services. I wrote the code, built it, deployed the services (this step was the longest and most painful), got the wsdl, generated the proxy and updated and built again. The reason it required full redeployment on build number changes is that the assemblies are bound by strong names. It was one heck of a time consuming job. All this work, just to update the proxy; WSDL generation required a live service. :)

Though I still have to do this to verify that the code that I wrote actually works, proxy generation is not that much of a hassle, especially when I am writing code that I know works already and there is minimal need to try it out.

This time has been drastically reduced. Thanks to System.Web.Services.Description.
ServiceDescriptionReflector.

 

The GetWsdl tool is here. I will publish it soon. Stay tuned.

-UG