> I have also started looking at converting the register_to_value
> gdbarch method to using values instead of buffer and a couple of
> parameters. Not very difficult, but not trivial either. That makes
> me worried about introducing bugs during the conversion, and
> I won't be able to test most configurations.
> 
> I'll send a patch as a follow-up email to this one.  It's not complete,
> but gives us an idea, and we can decide whether we want to continue
> or not (FYI: I will have very little time for this within the next
> couple of weeks).

So, here is the patch. I only did the conversion for alpha and
i386/i387. It shows the benefits, but it also reveals that the
implementation can be a little awkward, because the support routines
that are used of course where implemented using buffers rather than
values.  And in the end, the transformations need to be performed
on the contents of those buffers, so in their context, it might
even make some kind of sense.

Overall, this still seems like a step in the right direction, but
I am definitely concerned about introducing bugs on platforms that
I cannot test...

So, what do you think we should do?

-- 
Joel