[gdal-dev] GDAL/OGR 1.4.3 ABI incompatability with 1.4.2

[Even Rouault]

Yes, we must announce it clearly and withdraw it from download site.

More important, as we human can get wrong easily, it would be great to use an 
automated system to detect such accidental ABI break, to avoid that in the 
future. I'm not aware of such a means unfortunately, but it's worth looking 
for.

Le Wednesday 21 November 2007 19:05:59 Frank Warmerdam, vous avez écrit :
> Folks,
>
> It appears this change:
>
>    http://trac.osgeo.org/gdal/changeset/12517
>
> broke ABI compatability between GDAL 1.4.3 and GDAL 1.4.2.  In particular
> we absolutely cannot alter the virtual tables (or size - ie. layout of
> private data fields) of public classes such as OGRLinearRing.
>
> This came to light when I attempted to provide plugins for MrSID and
> ECW built against GDAL 1.4.3 DLLs to those working with GDAL 1.4.2
> DLLs.
>
> I think we will need to retract the GDAL 1.4.3 release, announcing it as
> dangerous to propagate and attempt to issue a corrected GDAL 1.4.4 release
> as soon as is practical.
>
> Does anyone have any thoughts on this?  Am I overreacting?
>
> I think this helps hightlight the importance of making ABI compatability
> requirements for stable branches very clear, and furthermore introducing
> some mechanism to test the theory.
>
> /me swears a blue streak.
>
> Best regards,