Re: Exception Handling

Bernhard Tschirren <tschirre@cs.curtin.edu.au> wrote:
>Could somebody please explain "zero runtime overhead" exception handling.
>Im only familiar with the traditional setjmp() / longjmp() methods.
>
>[I thought the code was instrumented with a table that the exception
>handler consulted to figure out which exception handler to use. -John]


Right, thought that may be too short for a newcomer to understand.


Imagine a table that is index by PC ranges that contains a handler to
jump to, should the PC be within that range. When compiling code, you
arrange for all code blocks to be bracketed by handlers. For a
function body, its handler unwinds the function call. When selecting
a handler, you pick the innermost handler (the tightest range).


Advantage, no inline code to manage EH variable stack, or
activations/deactivations. It is managed solely by the PC contents.


If you want to see this in action, get g++ (egcs.cygnus.com), and
compile something up. -fsjlj-exceptions will select the setjmp style
cost you will recognize, and -fno-sjlj-exceptions will select the zero
overhead scheme.


There are tons more details and various optimizations that can be
done, but you get the idea.
--

Return to the comp.compilers page.
Search the comp.compilers archives again.