At 17:04 22/04/2002 +0200, mbanck@... wrote:
> > At 12:31 21/04/2002 -0500, Geoff Hutchison wrote:
> > >I don't think we're quite at the stage for a 1.99.1 release,
>Why not? It's not like the whole world waits for us to release and then
>says: "ugh, they didn't even implement <insert popular format here>"
The opensource philosophy is "release early, release often" - and I think
this makes sense. I guess 80+% of likely babel users only want an exe, and
10% want an api/library. I know that when I wanted babel about 6 months ago
I simply needed a download and go:
> > >much less at 2.0 release.
> > >However, I think we have some nice bugfixes and new
> > >features, esp. including Peter's CML code to make a new "development
> > I'd be grateful to know what the philosophy of these three concepts is.
> > Presumably at 2.00 the code is stable enough to provide (hopefully) robust
> > babel.exe. Is that true of 1.99? Would a snapshot have a distributable
> > *.exe or is that too risky?
>Hmm, I'd say we should make a code snapshot without executable and make
>babel.exe available seperate. I guess we just need somebody to build it.
>Peter, could do it, that would be fine.
Unfortunately for the next 9 days I am overwhelmed so would be happy if
someone else did. The CML needs very minor bugfixes already discussed to
allow compilation. Of course there may be other bugs - I haven't tested the
internal coords stuff and have a little work there but...
>Otherwise we could ask Bryan Herger,
>who already contributed to the win32-build-stuff and distributes .exe's of
>his xdrawchem package. For the 2.00 there should probably be a nice setup.exe
>with help files and all.
>For what should be in 2.00, I'd say that we should get the library part of
>openbabel finished, so that we have a clean interface, namespaces and libtool
>building (I dunno how much of this has already be done by Geoff) . I don't
>know how this would work as a .dll in Windows though. I already know of one
>person (Martin Kroeker) who decided to parse the usage output and then
>file types through pipes. C-Bindings would probably help for this kind of
>IMO, we should keep releasing 1.99.x stuff until we're satisfied with the
>internal layout. Adding file-types can be done post 2.00, too, I guess.
>The downside of this is that I don't really have much experience in
>C++-library-coding, so apart from the libtool stuff, it would probably be
>Geoff to do it.
My experience with Java libraries is that it is highly worthwhile,
time-consuming and boring. I have written ca. 200 classes for CML in Java,
each with perhaps 20 methods. A large part of that is making sure the
interfaces are consistent and documented. Unlike writing algorithms I
don't rush home from a hard day's real-life work and start coding
libraries. IMO we need experience from babel.exe to find out what the
demand is, what the chemical bugs are, etc.
One aspect of all this is testing. Again highly boring, but necessary. I
have done some of this with CML-DOM in Java - putting molecules in and
seeing if you get the same molecule out :-) If it helps I have transferred
my library of CML test molecules to the CVS repository at
http://www.sourceforge.net/projects/cml - these are a slightly annotated
collection of nonCML filetypes which might be useful. NO guarantees that
they conform to *their* specs, of course :-) Any main release should have a
collection of molecules (including some pathological ones - I remember
writing a reader for CCDC CON format many moons ago and it ran for days
only to barf on K+ CH3-
which had "-1" bonds IIRC.
Peter Murray-Rust, pm286 AT cam.ac.uk
Unilever Centre for Molecular Informatics, Chemistry Department
Lensfield Road, Cambridge, CB2 1EW, UK