Re: State of new abi

[David Chisnall]

Hi,

Sorry for the delayed response - I was at a workshop last week and was a bit 
behind on email.

On 13 May 2018, at 22:08, Ivan Vučica <address@hidden> wrote:
> 
> Hi,
> 
> to confirm, is current libobjc2 master incompatible with current gnustep-base 
> master?
> 
> Stjepan has been building GNUstep using the following instructions:
>   http://wiki.gnustep.org/index.php/GNUstep_under_Ubuntu_Linux
> and I remembered that the failure could be due to the development of the new 
> abi.

This is a bug in libobjc2, which I have now fixed.  I forgot that objc_slot was 
part of the public API and so renamed it to objc_slot_v1, which then broke 
things.  I have now reverted its name to objc_slot and introduced a new 
objc_slot2, which fixes the issue.

> I powered on the laptop during my vacation, took a quick look, and told him 
> to git checkout 1.9. This seems to have worked.
> 
> Is this intentional?

No, it was a bug (though the master branch of libobjc2 may contain experimental 
code so may be less reliable than a release).

On a related note, I have had no feedback at all from requests to test the 1.9 
and master branches in preparation for the 1.9 and 2.0 releases.

> Presumably -base newabi branch is not ready to be merged yet — but should 
> libobjc2’s master branch be using the new abi then?

I have some other changes in the -base newabi branch.  It’s probably fine to 
merge the changes so support the support for the new NSConstantString parts, 
but unfortunately the new constant string exposes some bugs in -base’s unicode 
handling (in particular, incompatibilities with NSString and GSString), which 
are probably exposed by other NSString subclasses.

> I think that’s fine, apart from breaking scripts that build GS from head. 
> Travis CI is also erroring out:
>   https://travis-ci.org/gnustep/libs-base

The -base Travis script should be building libobjc2 without tests.  The 
libobjc2 Travis script builds it with tests and runs them.

David