| msg52224 - (view) |
Author: Guido van Rossum (gvanrossum) *  |
Date: 2007-03-15 00:09 |
| A patch for PEP 3115 (metaclass syntax for Py3k). This enables full [arglist] syntax in the class header, e.g. class C(B1, B2, metaclass=MC, *more_bases, **kwds): pass |
|
|
| msg52225 - (view) |
Author: Guido van Rossum (gvanrossum) * |
Date: 2007-03-15 21:06 |
| Here's a new patch, which generates correct code for all variants of the class header syntax. For now, I've implemented __build_class__ in Python; the built-in function by that name imports the build_class module and calls its build_class function. (I'll later redo this in C.) This is still rough; I didn't even delete the old BUILD_CLASS opcode. But only a few unit tests fail (especially the ast and compiler tests). There are no unit tests for the new functionality. There is no documentation. Failing tests: test_ast test_compiler test_ctypes test_descr test_grp test_pep292 File Added: meta.patch |
|
|
| msg52226 - (view) |
Author: Guido van Rossum (gvanrossum) * |
Date: 2007-03-15 21:53 |
| New version. Fixes problems with bsddb, removes dead declaration from bltinmodule.c, and updates outdated docs for MAKE_FUNCTION and MAKE_CLOSURE in opcode.h while I was at it. File Added: meta.patch |
|
|
| msg52227 - (view) |
Author: Guido van Rossum (gvanrossum) * |
Date: 2007-03-16 04:11 |
| Here's a new version that implements __build_class__ in C. It's still an ordinary built-in function loaded using LOAD_GLOBAL so you can play games with this; however I'd rather keep that an implementation detail and not a feature (unlike overriding __import__, which *is* a feature). One thing that's missing is support for __metaclass__, either in the module, or in the class. I don't want to support those (maybe the PEP should mention this), but we need a fixer for 2to3 to warn about this, at least (and to convert class C: __metaclass__ = M; ... into class C(metaclass=M): ...). BTW, the test_grp failure mentioned earlier is bogus; that's an artefact of Google's LDAP. Some more tests probably fail, but I'll worry about those later. File Added: meta.patch |
|
|
| msg52228 - (view) |
Author: Guido van Rossum (gvanrossum) * |
Date: 2007-03-16 15:53 |
| New patch, gets rid of LOAD_LOCALS and BUILD_CLASS opcodes and adds LOAD_BUILD_CLASS opcode which avoids looking for __build_class__ in the globals; it only searches the builtins (like the IMPORT statement). Though this is debatable. File Added: meta.patch |
|
|
| msg52229 - (view) |
Author: Guido van Rossum (gvanrossum) * |
Date: 2007-03-16 18:50 |
| And another. This adds a unittest for the new build_class functionality, and fixes one bug (in the way __prepare__ was called) found this way. BTW doctest rocks for this purpose. File Added: meta.patch |
|
|
| msg52230 - (view) |
Author: Guido van Rossum (gvanrossum) * |
Date: 2007-03-16 20:07 |
| Fixed two bugs, with the help of (improved) test_metaclass.py: - bug in symtable handling of keyword args that triggered an assert in debug mode - bug in build_class that mutated the keyword args dict in place I've got the feeling that there are leaks. File Added: meta.patch |
|
|
| msg52231 - (view) |
Author: Guido van Rossum (gvanrossum) * |
Date: 2007-03-16 20:27 |
| Fixed all library and test code I could find that used __metaclass__. The only two remaining tests (without -uall) failing are test_ast and test_compiler. File Added: meta.patch |
|
|
| msg52232 - (view) |
Author: Guido van Rossum (gvanrossum) * |
Date: 2007-03-16 23:12 |
| Fix syntax error in pickletester.py File Added: meta.patch |
|
|
| msg52233 - (view) |
Author: Guido van Rossum (gvanrossum) * |
Date: 2007-03-18 15:49 |
| Checked in as revision 54428. The PEP is accepted. (And yes, I realize this has been quite the monologue. :-) |
|
|