With the LLVM patch, it seems like this works.

Fix isAllocCall check

If we allocate a callee-save register that the parent function had
not also allocated, change its value, and then call a function
that then unwinds, the unwinder would pass the new (incorrect)
value of the callee save register to the next frame.

We either need to
- make sure callee-save registers are restored before any potentially-throwing
  callsite, or
- make the unwinder able to restore these registers for us, potentially
  by writing our own exception unwinder.

For now, the easiest thing to do is to disallow allocation of those registers.
(I'm not even sure how much we allocate them at all at the moment.)

Closes PR #188

Will get rid of it once it's merged upstream

Just let the rewrite continue but stop emitting any assembly,
and then at the end abort the rewrite.

fixes #169 - allow a negative step in xrange

Previously it would always link against the system libstdc++; not sure how
that was working.

Also, put that in its own included makefile so that we can track the dependence
on the configuration parameters (otherwise you would have to run a configuration
manually to get this change).

Make some things owned by the cf rather than using global registries,
so that 1) we can eventually free the right things when we start freeing
cf's, and 2) we can have a single ip->cf mapping instead of lots of
ip->other things mappings.

Also added some helper classes for making stack crawling easier.

fix header ordering in codegen/entry.cpp

Had to improve some of our rewriting; looks like there
must have been an LLVM commit that changes the live-out registers
we see.

Also, temporarily disabled format-checking: the rebase
updated clang-format, which now is formatting a few files
differently.  Will re-enable and re-format in the next commit.

Bunch of fairly-simple API changes to contend with.

Main problem is we're still running into those debuginfo issues.

Add tp_call handling, a couple random functions, and
allow GC-aware objects (though we don't do much with them).

ie "[a, b] = 1, 2"

The issue was that if we transformed the AST nodes corresponding to scopes,
we wouldn't be able to match the initial analysis with the subsequent queries
to the transformed AST nodes.

We had run into that before, but worked around it by just modifying the AST
nodes in place.  For generator expressions that wasn't a possibility,
so now we explicitly registers when we replace scope-related AST nodes.

looks like the memory leak in go.py is not coming from our heap (strings?)

Add some more slightly-modified benchmarks, but also make it easier to run them from the pypy directory

which would eventually cause stack space exhaustion and segfault.