• #ponce (18/18) Aug 06 2009 For the low-level programmer, sometimes you must use switch because virt...
• ponce (1/1) Aug 06 2009 In fact final switch can be an improved computed GOTO :)
• Walter Bright (2/2) Aug 07 2009 You're right that final switch could drop the range check. You should

#ponce <aliloko gmail.com> writes:

For the low-level programmer, sometimes you must use switch because virtual
functions are more expensive (cache usage...) and pointers to functions too.

The problem is that even if the optimizer turn the switch into a jump-table, it
has generally no clue about whether the value will be in the right range. Only
static analysis could do that and in C/C++ well...

So in C++ production code I'm currently forced to use something like this :


---------------------------------------------
#pragma_assert(x >= 0)
#pragma_assert(x <= MAX)

switch(x)
{
  case 0: doSomething0(); break;
  case 1: doSomething1(); break;

...

  case MAX: doSomethingMAX(); break;
}
---------------------------------------------

to make the optimizer know that i'm in the good range and drop the runtime
check (such conditionnals breaks pipeline).

My proposition is that final switch does a runtime check and eventually throw
an exception in debug mode, but does not check anything in release mode. It
would be consistent with how ArrayBoundsException works. 

It may seems a stupid request since I don't know how final switch works
currently. 

But my feeling is that it would make the fastest possible dynamic dispatch,
without loosing reliability. This could be very valuable for people interested
in performance.

ponce <aliloko gmail.com> writes:

In fact final switch can be an improved computed GOTO :)

Walter Bright <newshound1 digitalmars.com> writes:

You're right that final switch could drop the range check. You should 
post this as an enhancement request to bugzilla!