Archive for July 2011
There’s no doubt that in order to reasonably localize .Net applications in Persian, sooner or later the localizer should consider tampering .Net assemblies. The reason goes back to unreasonably poor implementation of Persian Calendar in .Net and applications in .Net. Providing Persian calendar is a must in most localization projects and there’s no way other than tampering assemblies to bring about that support.
For instance local users in Iran cannot live without Persian Calendar in their SharePoint sites, and you have to play with codes in SharePoint most important assembly (SharePoint.dll) to support Persian Calendar. Part of it is because that Windows in general does not provide for third party calendar systems to be added to the operating system.
The main trend in providing Persian Calendar in SharePoint is substituting one SharePoint standard calendars, such as Hijri with Persian Calendar. This way one have to somehow replace methods of an internal class namely HijriCalendarImplementation . For instance this class has a static method JulianDayToDate that will do conversion of a Julian day to a SharePoint SimpleDate structure. Obviously this should be changed if one plans to substitute Persian Calendar in place of Hijri. Expert guys here in Iran have used already available tools such as Reflector, to disassemble IL codes of SharePoint.dll and then replacing their codes and rebuilding the assembly back. They replace original assemblies with these modifies version. This way they’ve succeeded the mission.
When I first got this mission in Gostareh Negar, I actually didn’t know much about .Net programming. I was a C++ programmer, already expert in native code tracing and DLL overriding. Back to my experience in native code, I knew that sooner or later, rebuilding binary DLLs would show its disadvantages. So I just put my efforts to come up without a solution that does not require replacing the original libraries on persisted storage (hard disk). This leaded me to .Net Profiling API.
.Net Profiling API (see here) is originally devised for profiling tasks, i.e. performance measurements. Using this way one could instrument assemblies with specific calls to measure code metrics such as speed. For instance it can insert calls in method entry and exit points so that the total execution time of a method can be recorded. In effect, Profiling API provides methods to inject codes at run time when the CLR executes an application.
CLR (Common Language Runtime) includes a cross-CPU instruction format (Intermediate Language, or IL), and a JIT compiler to turn the IL into code executable by the target CPU. When it starts executing an assembly, it first Just In Time compile the IL codes into native machine code instructions on the target CPU. Within this process CLR may be asked to call a registered profiler and let it do profiling tasks and instrumentations including replacements of the IL code. In effect this will open a way to change IL codes at run time and an elegant way to do our localization mission.
While traditional profilers focus on instrumenting methods with measuring and logging calls, I focused on redirecting methods. Finally I came up with a Redirector. This could redirect method calls to another assembly by replacing method body with a call to injected method. Now I was able to inject my Persian Calendar codes directly into SharePoint.dll without touching the original assembly on the disk.
This method of code redirection based on profiling has many advantages including:
- It’s switchable: Many users fear that messing with binary codes may have side effects and causes errors. Since profilers can be easily switched off by server config, in case of suspicious behavior one may easily switch the redirection off and check if the problem is with the injected code.
- Does not require rebuild on new versions: When original provider releases a new version of the assembly, there is a good chance that changes are not in the redirected code. In case of SharePoint for instance the code for HijirCalendar didn’t change across the service packs and in SharePoint 2010. Therefore the redirector may still work on newer version of the assembly while others should rebuild it. In fact, Gostareh Negar clients installed SharePoint service packs without asking for an update.
- Does not interfere with code signing: Since original assemblies are normally signed, rebuilding them requires resigning which is normally a head-ache. Redirecting occurs in JIT compilation phase, and does not encounters signing issues.
There are also disadvantages:
- Speed: .Net code runs with lower performance while being profiled. CLR have to do profiling notifications in addition to normal tasks. This performance decrease is actually in load phase, when the program is completely JIT compiled, the effect vanishes. For web application it happens when the w3p process restarts.
Redirecting method based on .Net Profiling can be reasonably be a good plan for Persian Localization at least for web applications.
I often find the differences between State Machine and Sequential workflows much like those of event-driven programming with old control flows. Back to old days when we programmed with FORTRAN, we often thought of how to control the flow of program by branch instructions to do a job. The flow of control often had only one or at most a few of predefined paths. We could plan for these paths with the IF THEN branching instructions. Event driven programs are not so. Who can imagine the paths of instructions when someone uses Word?
State Machines are type of event-driven workflows most suitable for situations when it is hard to draw all of possible paths of a process. This is why State Machines provide a more flexible approach in programming Business Processes. It’s a pit SharePoint focuses on Sequential workflows.