Re: distribution issues - moving zlib into HDF5...

Mike Folk <mfolk@xxxxxxxxxxxxx> writes:

> Having seen the arguments from the HDF side and the netCDF side, I
> still agree, wholeheartedly, with Ed.  If we say HDF5 supports a
> certain kind of compression, there should be no if's about it.  It has
> to be complete support, not 95% support.
>
> Mike

That's funny, because Albert and Quincey have started to change my
mind!

Certainly Albert makes a good point - if zlib is built into HDF5, and
then a user wants to link a program with zlib (say, because he wants
to use zlib for something else), then everything will break, because
the zlib functions will already be built into HDF5, and there will be
clashes.

(This could be solved, however, by the user just using the hdf5
versions, and rebuilding HDF5 with a new zlib if that is required).

So now I don't know what to do. 

I am going to cycle home and think about it. Any consensus from the
HDF5 team on this would be very interesting.

One other problem with just accepting what the user has installed is
the version. I don't just need HDF5, I need version 1.8 or
better. Otherwise my compile will break, because I use functions from
the (upcoming) 1.8 release which are not in previous versions.

This problem is removed if I distribute HDF5 with netCDF, because of
course then I could distribute a HDF5 version that I know to work with
netCDF-4.

I would like to find a way to make this easier for the end users, but
it is not clear what the best solution is. I have posted a general
question about this in the autoconf mailing list, to see if there is
any community consensus there. I'll let you know if there is any
useful feedback.

Thanks,

Ed
-- 
Ed Hartnett  -- ed@xxxxxxxxxxxxxxxx