Re: RFA: ia64 portion of libunwind patch

[Andrew Cagney <ac131313@redhat.com>]

> On Fri, 07 Nov 2003 16:46:59 -0500, Andrew Cagney <ac131313@redhat.com> said:
> 
> 
>   Andrew> Would it be possible to define the interface so that, when
>   Andrew> dip-> u.ti.table_data is NULL, the code just fetches values
>   Andrew> from memory using the memory callbacks?  i.e., don't require
>   Andrew> GDB to fetch the entire table but instead fetch the bits
>   Andrew> that are needed.
> 
> Well, it's software, so anything is _possible_, but I'd rather not do
> that, because it creates artificial differences between the
> (speed-critical) local unwind case and the remote unwind case.
> Furthermore, the unwind table is of a known size, relatively small,
> and accessed fairly randomly (via binary search), so it is normally
> preferable to read the table all at once and I'd rather design the API
> for this common case.

Your comparing the self unwind case (i.e., what happens when a program 
does a "throw") vs the external unwind case (i.e., like what GDB has to do)?

Is fetching the table elements via a function, rather than a direct 
access, really that significant an overhead?

Andrew