[geos-devel] Problem with building 3.0.0rc

[Sean Gillies]

Mark Cave-Ayland wrote:
> On Fri, 2007-12-28 at 11:45 -0700, Sean Gillies wrote:
> 
>>>> The mkinstalldirs script appears to be buggy on this particular
>>>> platform. One shift too many. Strangely, the 3.0rc builds fine on
>>>> another Ubuntu 6.10 i686 of mine. Same bash version even.
>>>>
>>>> I suppose I could try to fix mkinstalldirs, but it seems like it should
>>>> be simpler and more foolproof for the build to ignore swig/python
>>>> alltogether unless it is explicitly enabled.
>>> It would be good to know what is going wrong.  Strk and I did (and Mark
>>> a bit) did spend some time trying to make sure this works, and it seems
>>> like you've found some bug in the code or our logic.
>>>
>>> Charlie
>> I'd like to implement the solution at
>>
>> http://www.gnu.org/software/automake/manual/html_node/Conditional-Subdirectories.html
>>
>> It appears trivial. Was there really a 3.0 final made on the 21st as
>> appears in the downloads directory? If not, I'd really like to make this
>> fix ASAP.
>>
>> Sean
> 
> 
> Hi Sean,
> 
> Having looked at Makefile.am in the SWIG directory, the mkinstalldir
> target appears to be generated automatically by autotools. So while I
> don't object to applying a conditional subdirectory fix, the fact that
> it works on another box of yours with the same configuration means that
> there is something else somewhere on your box which is causing the build
> to fail. In other words, applying a patch for this is just masquerading
> another problem on the box in question.
> 
> (FWIW it builds fine here on 32-bit Ubuntu 6.06)
> 
> 
> Kind regards,
> 
> Mark.
> 

Maybe it does in this case, but making the builds conditional also heads
off any number of future problems of this sort. If you look in
swig/python/Makefile.am, you'll see that conditional building has
already been implemented, but in a less robust fashion.

Cheers,
Sean