On Sat, Jan 24, 2009 at 2:05 AM, Greg Stein <gstein_at_gmail.com> wrote:
> On Sat, Jan 24, 2009 at 00:17, Bert Huijben <bert_at_vmoo.com> wrote:
>>> -----Original Message-----
>>> From: Hyrum K. Wright [mailto:hyrum_at_hyrumwright.org]
>>> Sent: Friday, January 23, 2009 10:26 PM
>>> To: svn_at_subversion.tigris.org
>>> Subject: svn commit: r35439 - in trunk/subversion/bindings/swig: .
>>> python/libsvn_swig_py ruby/libsvn_swig_ruby
>>> Author: hwright
>>> Date: Fri Jan 23 13:26:16 2009
>>> New Revision: 35439
>>> Fix the swig bindings after r35424.
>>> * subversion/bindings/swig/core.i,
>>> Include apr_md5.h in place of svn_md5.h.
>> If this breaks our build, it probably breaks third party tools too?
>> I like the idea of the public header cleanup, but I'm not sure if we can do
>> this without breaking third party code.
>> While it is not part of the ABI, the header files are certainly part of our
>> public API.
> I thought about that, and am somewhat torn. You're right in that a
> source change is now needed ("add another #include"). But I'm also
> thinking, "well... you used something from a header, and didn't
> include it. fix the bug in your code."
> All of our APIs and ABIs are the same. All that is changed is our
> delivery of them to developers.
> (yes, it is a grey area)
For the record (and yes, this is very delayed, and I don't actually
expect us to go back and change this now that it's been released in
Had I been paying more attention back in January, I would have
objected to this. The recursive includes of a header are part of its
API. It is a huge pain to have to go back and add #includes for
svn_props.h all over the place because it is no longer included in
svn_error_codes.h (which ~everything includes). In the future, I
think we should avoid doing this again.
glasser_at_davidglasser.net | langtonlabs.org | flickr.com/photos/glasser/
Received on 2009-06-05 01:10:13 CEST