This patch is implementation of PEP-3102, keyword-only
parameter.

Important changes include

• code object now has co_kwonlyargcount to keep the
  number of keyword only argument - this is analogous to
  co_argcount.
• function object now has func_kwdefaults (dictionary)
  to keep the mapping between keyword only arguments and
  defaults for them. Only kwonly argument with default
  values are in the dictionary - this is analogous to
  func_defaults.
• APIs for code object changed - both C API and Python
  Api. PyCode_New now takes number of keyword only
  arguments, and new.code also takes number of keyword
  only arguments.
• MAKE_FUNCTION now takes another oparg, which is
  number of default keyword only arguments - and the name
  of keyword only argument and its default value are in
  the value stack - it is similar to oparg of CALL_FUNCTION.
• MAGIC in import.c changed, since bytecode is changed.

That's pretty much everything that's important, and the
rest is in the code itself.

And my patch passes all regression tests excepts
test_frozen.py, which depends on the hard-coded mashal
value, which needs to be regenerated when bytecode
changes. However, freeze.py is broken - specifically,
modulefinder.py is broken as Guido said. So, currently,
I commented it out.

Logged In: YES
user_id=33168

import.c: comment has double word 'added'
ast.c:

• why is static commented out for ast_for_arguments?
• in handle_keywordonly_args, doesn't the default case need
  to return -1 after setting the error?
• need to change the // comments to /* */
• it looks like a bunch of lines had whitespace, but no
  other changes. this makes it hard to see the real changes.

compile.c:

• i think there is a bug in the arglength calc if there are
  more than 255 positional args and kw_default_count since
  there is |= kw_default_count << 8 (2 places)
• return should go on its own line like the rest of the
  surrounding code.
• why kwonlyargs and kw_args? the underscore seems
  inconsistent

in the compiler package, I didn't see a change to
MAKE_FUNCTION similar to the one in compile.c for the opcode
stack effect.

the change to regrtest.py doesn't seem necessary (just an
added blank line)

there should be a lot more tests added for both positive and
negative conditions (syntax errors). Include variants with
func(**kwargs) func(*args), etc in calls.

it's not good to comment out the tests in test_frozen.
otherwise, we will probably forget about the problems.

when you say it passes all tests, did you run with -u all?
the compiler doesn't do all the tests without it (and one or
two other tests too).

in the new test, i think you need to pass a name for the
filename param to compile. I think there was some platform
(windows?) that crapped out with an empty filename. (if you
copied from an existing test, then i must be remembering
something different.)

In testRaiseErrorWhenFuncall(), you can use
self.assertRaises, at least for most of the cases.

you need a test_main for this test to be run from regrtest.

Logged In: YES
user_id=595483

> * it looks like a bunch of lines had whitespace, but no
> other changes. this makes it hard to see the real changes.

I changed some tabs in ast_for_argument function to spaces -
tabs and spaces are mixed, and it's hard to edit. It would
have been nice if I separate that white space conversion
into another patch, but since it's p3yk branch I thought it
can be allowed.

> when you say it passes all tests, did you run with -u all?
> the compiler doesn't do all the tests without it (and one
> or
> two other tests too).

Yup, it passes all the tests except test_frozen and
test_linuxaudiodev.py(maybe because my box doesn't have
audio device..?) and some skipped tests including
test_bsddb3.py, etc

> in the new test, i think you need to pass a name for the
> filename param to compile. I think there was some platform
> (windows?) that crapped out with an empty filename. (if you
> copied from an existing test, then i must be remembering
> something different.)

Yeah, I copied it from test_with.py, but now I'm passing a
filename.

Logged In: YES
user_id=595483

Maybe I should mention that I've uploaded new patch that
fixed what Neal mentioned.

Logged In: YES
user_id=595483

The original patch crashes when a function has keyword only
arguments but no default values for those arguments, and
calls the function without enough arguments. Fixed the bug,
and added testcases for that.

Logged In: YES
user_id=6380

General comment: I think you should update or add a comment
explaining the contents of the code object as it has changed
(unless that's all described in the PEP).

There are a number of places where the code is wider than 79
characters or where the indentatiln style seems to be off.
You may have to set your tabstops to 8 spaces to see these.

E.g.:

ceval.c:

• line 2624 should be indented with next line
• line 2680 is too long
• @@ -2694,13 +2715,35 @@ several lines too long
• line 3618 should be indented with the next line

ast.c: @@ -591,6 +591,63 @@ please follow the surrounding
4-space indents; don't use tabs here

marshal.c: line 871 too long

codeobject.c: several lines too long

funcobject.c: too long at "non-dict keyword only default args");

Lib/compiler/*.py: several lines too long

Lib/test/test_new.py

Modules/pyexpat.c: @@ -279,6 +279,7 @@ indentation error

Modules/parsermodule.c: several lines too long; the loop at
"while (res && i+1 < nch)" is indented one tabstop too many
(you seem to have edited this file with ts=4?)

Logged In: YES
user_id=595483

The codeobject has a kwonlyargcount meaning the number of
keyword only arguments. Thus, the api for creating new code
object is changed: PyCode_New now takes kwonlyargcount.

So, the signature changes from this
PyCode_New(argcount, nlocals, stacksize, ...)

to following.
PyCode_New(argcount, kwonlyargcount, nlocals, stacksize, ...)

From python, you can access this with name
"co_kwonlyargcount" such as, co.co_kwonlyargcount just like
you access argcount with co.co_argcount.

Logged In: YES
user_id=595483

I made a big mistake with the grammar - currently, the
grammar doesn't allow keyword only argument after *vararg.
At some point of the development process, I just assumed
as such. Underlying logic does not assume anything about
that, so I'll try to fix it as soon as possible

Logged In: YES
user_id=6380

I see. The Grammar should simply make the NAME after the '*'
optional. Shouldn't be too hard to fix should it?

Logged In: YES
user_id=595483

The new patch that I am uploading fixes the Grammar
problem - originally I made mistake not allowing *vararg
after keyword only argument. The python compiler package
doesn't work now, but everything else should work now.

Logged In: YES
user_id=6380

Checked in as r52491.

I'm keeping this open awaiting a patch for the compiler package.

Now the compiler package works. The patch passes regrtest with "-u all" option and "-u compiler" option (except test_frozen.py). The patch is against latest p3yk branch HEAD. The last patch accidentally included some test code in test_compiler.py (and which is committed to the svn) so this patch includes fix for that.
Ah, Guido, when you committed the patch (previous version) last time, you did not add Lib/test/test_keywordonlyarg.py so, this time, don't forget to add it :) Also, last time when we talked, I think you said that the patch makes some test case fail on your laptop machine. You might want to check that.

Submitted.

Jiwon, can you mail me another copy of test_keywordinoyarg.py? I believe I lost it.

All submitted. (test_keywordonlyarg.py too)