Sibcalls and exceptions
Richard Henderson
rth@redhat.com
Mon Nov 26 13:12:00 GMT 2001
On Mon, Nov 26, 2001 at 11:15:59AM -0800, Mark Mitchell wrote:
> I'm pretty sure that we agreed that there was an endemic problem here
> that could not be easily solved, probably involving the exception
> regions that are generated during the double argument evaluation.
That rings a bell.
Perhaps a temporary whose lifetime begins before argument evaluation,
and whose lifetime ends after argument evaluation, but before the
function call itself? In such a case, with the multiple evalutions
for normal and sibcalls, we do not have a nice nesting of exception
regions, and will confuse the code that attempts to assign region
numbers to instructions.
There is some code in calls.c that attempts to detect such a case,
but it's not implausable that it can't deal with whatever we came
across.
r~
More information about the Gcc-patches
mailing list