[SciPy-user] HDF5 vs FITS

[Francesc Altet]

El dg 22 de 04 del 2007 a les 16:40 -0400, en/na James Turner va
escriure:
> Hi Francesc,
> 
> Thanks for the Pytables history. I wasn't suggesting that there is a
> problem with the similarity of the names -- I did wonder at first if
> there is a connection with FITS tables, but that can always be solved
> with a comment in the documentation :-).

Yes, that would be enough (at least on those fortunate cases where
people do actually read docs ;-).

>  > It depends on what the author would mean by 'compatible'. I think
>  > that HDF5 is not meant to read FITS directly (nor will be in the
>  > future), but through a conversor.
> 
> Yes, I understood that.
> 
>  > There is a RFC about this subject going on:
>  > http://www.hdfgroup.uiuc.edu/RFC/HDF5/fits2h5/fits2h5.htm
> 
> That IS interesting to know. Defining a standard mapping between the
> two formats seems like a good idea, allowing at least some level of
> interoperability at the end-user-program level (as opposed to NumPy).
> Since the convertor is only one-way, I infer an expectation that HDF5
> would supersede FITS, which I wouldn't really agree with, for similar
> reasons to Perry's comments about archiving. There is a LOT of
> existing astronomy software that does not handle HDF5, so a
> conversion back to FITS from the "FITS within HDF5" structure would
> be needed before the latter is really useful. Of course sticking to
> "FITS within HDF5" sort-of defeats the point, but at least it could
> allow processing FITS data with HDF5 software or vice-versa, assuming
> the HDF5 software is flexible enough in its expectations regarding
> data structures.

Well, I don't know the plans for HDF5 and FITS for the future at all, so
I'm not the most adequate person to talk about this, but my impression
is that fits2h5 is a try to bring to the astronomers a conversion tool
that would let them to use HDF5 aware tools for coping with their data.
That's all. In particular, I doubt that the goal would be to make
astronomers to change their format, because, among many other reasons,
there should be a *vast* set of libraries that already works against
FITS, and changing all of this to use HDF5 would simply be a no go.

Although, perhaps having a bidirectional "FITS within HDF5" solution as
you are suggesting can be a great idea, who knows.

> Just an off-the-top-of-my-head reaction; I'm sure Perry et al. are
> familiar with such issues in more detail.

Mine too indeed.

Cheers,

-- 
Francesc Altet    |  Be careful about using the following code --
Carabos Coop. V.  |  I've only proven that it works, 
www.carabos.com   |  I haven't tested it. -- Donald Knuth