Let's perform a simple benchmark testing the random and sequential access to arrays and lists. I would like to compare:

• static array with predefined size (allocated on stack)
• TList container class (allocated on heap)
• dynamic array (allocated on heap)
• generic Delphi TList or Free Pascal TFPGList (allocated on heap)

I use an array/list of 10M elements of integer and varaint data type. You can change the type definition in "type" section if need to test other cases.

The namespaces conception in C# seems exhaustive and insufficient at the same time in comparison with strong module notation like in Oberon-2 and even with modular programming like in Free Pascal/Delphi.

Why insufficient? You cannot declare constants and functions within namespaces but you should add a class with constant field or static method for it. It's really sad...

Why exhaustive? The "using" directive with such introduced classes is the source of unexpected errors. Here is an example below.

As I'm working constantly with both DUnit and FPCUnit, the differences of shortcuts and ergonomic are significants when running in GUI mode. I.e.

This bug is reproduced on Delphi 7 and C++Builder 6 at least (probably in later versions too) when starting Borland make utility. Indeed, the message "Fatal: Command arguments too long" and the real problem are as like as chalk and cheese.

Check the length of your PATH variable. In my case the length was > 400 characters. Deleting some paths reduced the length down to 300-350 characters and solve the problem.

I know not much feedbacks about real tests on hyper-threading in OLAP world. Here is a SQL Server hyper-threading from Linchi Shea.

So, using hyper-threading doesn't mean that you will get the advantages always. At least some performance tests are required to ensure this technical choice.