[Gdal-dev] Re: CVS

[Charlie Savage]

I agree with Frank, I'd leave them in CVS.  Otherwise you are requiring 
windows users to get SWIG, install it, and run it. That might be ok, but 
SWIG is changing alot and that means you really have to use the *right* 
version of SWIG - i.e., the ones that the wrappers were built with.

Anyway, I've been using 1.3.29 since there was a lot of work fixing 
backwards compatibility issues.

As far as GEOS, my understanding from strk is that it will likely 
include the SWIG generated bindings for the 3.0 release (I've been 
pushing for it).  However, it does make sense to generate a "final" 
version right before tagging a branch....

Charlie


Frank Warmerdam wrote:
> Howard Butler wrote:
>> Some reasons for keeping them in CVS:
>>
>> 1)  GDAL has history of keeping generated code in the CVS tree.  The 
>> old-style Python wrappers, which are generated from swig 1.1, pretty 
>> much *have* to be pushed into the tree because Frank is the only 
>> person I know of with swig 1.1 installed anywhere.
>>
>> 2) SWIG is in severe flux right now.  1.3.28 was a *major* release, 
>> complete with incompatibilities, bugs, and API changes.  The distros 
>> all vary in the swig that they provide, and this can have a great 
>> impact on what people see when they generate bindings themselves. 
>> 1.3.29 came quickly (and still has some broken Python things).
>>
>> 3) Windows
>>
>> Some reasons for dropping them from CVS:
>>
>> On the flip side, I agree that having the generated bindings in CVS is 
>> a pain.  They almost always seem to generate CVS conflicts and 
>> sync'ing between the various flavors is not always consistent.  I 
>> almost always generate my own bindings with the swig I have installed 
>> when doing any development.  I agree that a 'make dist' target should 
>> be generating *all* of the bindings for each language flavor anyway, 
>> regardless of whether or not we keep the bindings actually stored in CVS.
> 
> Howard, Ari,
> 
> At least for now it still seems quite tricky to produce working swig
> wrappers for the next gen stuff, so I would prefer that they live in
> CVS.  I'm especially concerned with how we can make sure generated wrappers
> actually work for all the languages before a release.  One approach is for
> me to use whatever is in CVS.  The files could lag a bit, but would
> presumably be in a working state.
> 
> Do either of you feel confident enough in your own swig version to generate
> all the bindings and commit them shortly before a release?
> 
> Best regards,