Potentially merging the transactional-memory branch into mainline.

[Aldy Hernandez]

> Aldy, what folks are asking for is reasonably contained patches from
> the branch that can be reviewed.  Ideally they could be installed
> independently, but it's not strictly necessary.

I already did so:

compiler/
libitm/
libstdc++-v3/
misc/
toplevel/

The ChangeLog's are at the top.  The testsuite changes are at the bottom.

What Richi is complaining about is something that got fixed yesterday 
and is no longer in the branch, and can easily be fixed by hitting 
reload on his browser.

> So you might have the runtime as a patch, the testsuite as a patch,
> the front-end changes as a patch, the expander bits as a patch, the
> optimization bits as a patch, any backend bits as a patch.

I can continue splitting up the patch into smaller and smaller pieces, 
but I'd rather spend my time elsewhere.  Really, the patch isn't that 
hard to look at, and is fairly self-contained (or mechanical in other 
palces).  However, if after looking at the above patches, you deem them 
not independent enough, I am willing to oblige.

What I'd like to know is why I am constantly being asked for more stuff 
on the transactional memory branch, whereas the memory model stuff got 
approved without even asking for a complete diff (and I know what I'm 
being asked for, cause I'm doing both merges).  Similarly for when I 
merged the tuples branch and that stuff touched everything!