Which are quite hard to reproduce because they depend on the details
of the C stack.

Anyway, we should probably start using the recursion-depth-tracking functions
in more place, rather than getting segfaults.  There are a lot of CPython tests
that test for recursion-depth-checking.

Fix some corner cases in the irgen refcounter, and fix a refcount
issue in the bz2 module itself.

Switch to CPython way to build standard C extensions.

Merge AST/BST work into master

AST: deallocate nodes

Use a bumpptr allocator to keep track of the AST nodes.
I could not use the LLVM one because it does not call the destructor on allocated elements.

AST & BST: remove unnecessary nodes and fields + little less memory usage

Disable some extensions, part of because they were handled by
from_cpython/CMakefile.txt, part of beacuase we need to enable it in
seperated PR. Such as tk extension.

And also add some modifications to let extension can find the source
code in correct path.

Some minor improvements from the metaserver branch

We mostly did a good job of picking flags that CPython doesn't use (or at least, that
its manpage doesn't advertise), but there are a couple that need to be changed.

One is for TRAP which is a feature I don't even remember anymore, so that's removed.

The other is for stats -- '-s' is now changed to '-T'.  This will be annoying to change
on the perf-tracking side.

This reverts commit 111ca525.

BST: a bytecode-like AST

And make BST_arguments a bit safer by removing misleading fields.

And remove some extraneous things from bst

fix a bug where compare could ignore CAPI exceptions and add _heapq

A bunch of function metadata refactorings

since this is where it's supposed to live.  The function ctor
copies these fields from the code object.

Instead, merge it into BoxedCode.  This way we'll get refcounting
of our codegen datastructures.

I didn't do anything more than inline all of FunctionMetadata's fields
into BoxedCode, and then delete the FunctionMetadata object.

Revert "Merge pull request #1349 from kmod/scoping"

This reverts commit 2783ab4e, reversing
changes made to ea8890e8.

Let Pyston compatible with GCC 5.x

I think we don't need libunwind to generate doc. And libunwind will call
latex2man and pdflatex to generate its doc. And this can not be disabled
through command line. Please see here: https://bugzilla.mozilla.org/show_bug.cgi?id=738302

When update to gcc 5.4, it complain there has no
has_trivial_copy_constructor, please see here: http://www.chpc-tech.com/EN/NewDevelopments/2015/15.04.23.gcc5.1.html
So use standard c++ 11 type traits.

GCC 5.4 will report those type castings are not legal.