[Dwarf-discuss] optimized code debugging

Chris Quenelle Chris.Quenelle
Tue May 24 17:41:21 PDT 2005


Subject was: Issue 050222.1

Using an address to point to something
that *used* to be there seems like a fragile
thing in the face of compiler optimizations.
I doubt that particular address will mean
very much after the code around it has been scheduled
by the compiler.  Also, you will need the compiler
to keep an internal pointer to that location
in the intermediate code, and if it's not pointing
at anything, it will just get attached to
the following piece of code or fall through
to the end of the block, or some other nonsense.

I've been thinking about how to implement
statement markers and line markers in the
face of sophisticated optimization, and I'm
not sure such markers will mean anything unless you
also tell the compiler to treat those boundaries as
hard boundaries to most (or all) optimizations.

--chris




Bishop, John E wrote:
> Issue is
> 
>    Location list entries should allow ending
>    address to be same as starting address
> 
> I agree that the restriction should be lifted.
> 
> That's partly because we may want to use zero-
> length ranges as well, but also because there
> doesn't seem to be a strong reason not to allow
> it, and it'd be a handy way to put a kind of
> bookmark down for a variable.
> 
> 	-John



More information about the Dwarf-Discuss mailing list