GitLab

http://codespeak.net/pipermail/pypy-dev/2003q3/001012.html

for in-depth discussion and description of what is new.

replace it with the branch in the next checkin.
This way we keep the history of the changes during
the builtinrefactor branch.

files on your working copy that have inconsistent line endings

Now we give enough static clues about the types to allow for
a complete list of what-to-try-in-which-order to be built
for each input argument types. This cleanly separates method
selection from method call, as wished by mwh and Samuele.
It should also speed up pypy a bit but not dramatically
enough to notice I'm afraid.

methods

to trigger the new TestSkip mechanism.

skipped when testing other objspaces)

interpreter traceback through SApplicationExceptions.

make the interactive option -i
(more like CPython)

There are two conceptual changes, but they are two textually entangled
to be easily checked in separately :-(

The first change is allowing a test to skip, by raising a
pypy.tool.test.TestSkip exception.

test.objspace() raises TestSkip if it is asked for a specific flavour
of object space and that differs from the specific flavour of object
space that is being tested (as controlled by command-line arguments).

This change is all of the changes to interpreter/unittest_w.py and
some of the work in tool/test.py.

The other change is a new option (-k, for no good reason other than -i
already being taken) that drops you into a interactive session when a
test run has failures.  From this you can list the errors ('l'), print
tracebacks ('tb') and drop into a pdb session for each failure ('d').
Blank lines and 'q' quit.

I've also added a sketal PyPy-specialized subclass of pdb.Pdb.

make it work on Mac OS X (don't know what was wrong before, sorry, not that
interested in finding out)
remove directories that are inside the pypy tree from sys.path (reduces problems
with 2.3 which expects to find a module called opcode in the std library)

fix to get test_assign_local_w_flow_control properly working while keeping the other tests in pypy/objspace/ann/test/test_objspace.py OK.

the point is not to consider a fix-point reached as long as there's a W_ConstantIterator which is being consumed.

  actually running the tests with the respective objspace.
  (i.e. objspace.ann* tests are skipped for StdObjSpace and
   objspace.std* for AnnObjSpace)

- removed some cruft from tests

M    interpreter/test/test_interpreter.py
     "test_print" now redirects sys.stdout to some class to
     check if "print 42" works correctly

M    objspace/std/stringobject.py
     removed print junk

_M   objspace/std/register_all.py
     fixeol

M    tool/test.py
     improved hack to skip tests when running wrong objspace tests.

better, but still, many tests hang when run with -A, and there's at
least one (in test_interpreter.py) that causes an infinite loop.

M    interpreter/pyframe.py
	Move the clone() method and the clonecells() function to
	objspace/cloningcontext.py.
	Commented out extracting __builtins__; it seems not needed
	and confused the Ann space.
M    objspace/ann/test/test_objspace.py
	Add another test; disable one that no longer works.
M    objspace/ann/cloningcontext.py
	Many changes; moved cloning here from pyframe.py.
	App helpers called from the interpreter are now run in a
	special subclass of CloningExecutionContext, which doesn't
	unify frames (because we need the real results from these
	helpers).
M    objspace/ann/wrapper.py
	Add constant iterators; add some APIs; other misc changes.
M    objspace/ann/objspace.py
	Change the wrapper cache to contain the key objects too;
	before, wrap(1) would return self.w_True.
	Get rid of unused clone_locals() method.
	Add gethelperspace() to return a HelperObjSpace() instance.
	Implement some more operations correctly on constants, for the
	appspace helpers.
	All operations now catch exceptions and call self.reraise()
	when appropriate.
	Slight optimizations in generic_operator().

Ann space code to work (these are used by some appspace helpers called
from the interpreter).

passed to AppHelper.  It defaults to self.

Re-enabled them (I cannot see the problem mentioned by Holger?).
Removed the default fall-backs in default.py.
The purpose of this fix is to make the special methods visible at the
application level (e.g. 'None.__ne__').

Re-enabled the test.

the last hours of the sprint nobody really ran all the tests?!
Maybe the fact that stringobject now uses the list object to
implement itself brought up some boostrapping issues regarding
strings <-> lists. For me, all tests pass now for all object
spaces.


M    interpreter/executioncontext.py
     fixed exception printing problem (probably a bootstrap issue
     regarding string ops)

M    objspace/std/test/test_multimethod.py
     rearranged to more finer grained unit tests. obvious delegation
     doesn't work as expected with the FakeObjSpace.

     I had to disable one delegation-test! Please somebody look into it!
     I couldn't figure out why exactly but it's probably due to
     Armin/Samuele's latest changes in combination with other's
     changes.

M    objspace/std/register_all.py
     disabled the add_extra_comparison stuff because it also seems to
     have bootrapping issues. Probably because it tries to iterate over
     all types on every register_all invocation of each module.

dict.__new__ when I translated into dict.__init__).

One thing added is properly creating function objects from code.

M      objspace.py
        support calling python functions
        support newslice
        reraise exception on add of incompatible arguments
M      wrapper.py
        add W_PythonFunction
M      test\test_objspace.py
        two new tests that helped discover the problem with cloning fast locals

Reorganized constructors of all object types accordingly.