Re: Live range splitting in new allocator

[Daniel Berlin <dberlin at redhat dot com>]

On Tue, 30 Jan 2001, Michael Meissner wrote:

> On Mon, Jan 29, 2001 at 04:05:51PM -0500, Daniel Berlin wrote:
> > I am, that is what the allocate_reg_info does.
>
> Yep.
>
> > There is some stupid table somewhere that is being allocated at the
> > beginning of a function, and doesn't ever sxpect to be extended.
>
> If you have the swap space, this is a case where the electric fence debug
> library helps.  If you are not familar with efence, you link cc1 with it, and
> it provides its own malloc/free which allocates every request so that the last
> element is at the end of a page, and then makes sure that the next page is not
> accessible.  Run it in the debugger, and where it traps with a segmentation
> violation, is the access beyond the end of the array.  In the past, with heavy
> use of obstacks, efence was less effective in debugging the compiler.

You actually don't need to even link it anymore, there is an "ef" script
that adds the shared libefence to LD_PRELOAD and runs the program.
Works great.

Now i have to figure out how to fix this stupidity.

--Dan