[Python-Dev] Stabilizing the C API of 2.6 and 3.0

[M.-A. Lemburg]

On 2008-05-26 23:34, Christian Heimes wrote:
> M.-A. Lemburg schrieb:
>> Isn't that an awefuly confusing approach ?
>>
>> Wouldn't it be better to keep PyString APIs and definitions in
>> stringobject.c|h
>>
>> and only add a new bytesobject.h header file that #defines the
>> PyBytes APIs in terms of PyString APIs ? That maintains
>> backwards compatibility and allows Python internals to use the
>> new API names.
>>
>> With your approach, you've basically backported the confusing
>> notion in Py3k that str() maps PyUnicode, only that in Py2
>> str() will now map to PyBytes.
> 
> The last time I brought up the topic, I had a lengthy discussion with
> Guido. At first I wanted to rename the API in Python 3.0 only. Guido
> argued that it's going to cause too much merge conflicts. He then
> suggested the approach I implemented today.

That's the same argument that came up in the module renaming
discussion.

I have a feeling that we should be looking for better merge
tools, rather than implement code changes that cause more trouble
than do good, just because our existing tools aren't smart
enough.

Wouldn't it be possible to have a 2to3.py converter
take the 2.x code (including the C code), convert it and then
apply any changes to the 3.x branch ?

This wouldn't be merging in the classical sense, it would be
automated forward porting.

> I find the approach less confusing than your suggestion and my initial
> idea.

I disagree on that.

Renaming old APIs to use the new names by adding a header file with
#define <oldname> <newname> is standard practice.

Renaming the old APIs in the source code and undoing the renaming
with a header file is not.

> The internal API names are consistent for Python 2.6 and 3.0. The
> byte string C API is prefixed PyBytes and the unicode C API is prefixed
> PyUnicode. A core developer has just to remember that 'str' is a byte
> string in 2.x but an unicode object in 3.0.

So you've solved part of the problem for 3.x by moving the naming mixup
back to 2.x.

> Extension developers don't have to worry at all. The ABI and external
> API is mostly the same and still exposes the 'str' functions as PyString.

Well, yes, but only due to a preprocessor hack that turns the
names used in bytesobject.c back into names you'd normally look
for in stringobject.c.

And all this, just because Subversion can't handle merging of
symbol renaming.

>> You'd have to add an aliase bytes -> str to the builtins to
>> at least reduce the confusion a bit.
> 
> Python 2.6 already has an alias bytes -> str
> 
>> Yes, but please let's first discuss this some more. I don't think
>> that the timing was right.... you started this thread just yesterday
>> and the patches are already checked in.
> 
> I'm sorry if I was too hasty for you. I got +1 from a couple of
> developers and it's basically Guido's suggestion.

Please discuss any changes of the 2.x code base on python-dev.

Such major changes do need more discussion and possibly a PEP as well.

Thanks,
-- 
Marc-Andre Lemburg
eGenix.com

Professional Python Services directly from the Source  (#1, May 27 2008)
 >>> Python/Zope Consulting and Support ...        http://www.egenix.com/
 >>> mxODBC.Zope.Database.Adapter ...             http://zope.egenix.com/
 >>> mxODBC, mxDateTime, mxTextTools ...        http://python.egenix.com/
________________________________________________________________________
2008-07-07: EuroPython 2008, Vilnius, Lithuania            40 days to go

:::: Try mxODBC.Zope.DA for Windows,Linux,Solaris,MacOSX for free ! ::::


    eGenix.com Software, Skills and Services GmbH  Pastor-Loeh-Str.48
     D-40764 Langenfeld, Germany. CEO Dipl.-Math. Marc-Andre Lemburg
            Registered at Amtsgericht Duesseldorf: HRB 46611