All,
First of all, congratulations to Jennifer and Brian on the milestone! I am
very much looking forward to making v2 the focus for the opengrads
activities.
I was able to download Jennifer's sources last night and relatively easily
build it with the new supplibs. For convenience I built --without-hdf as I
wanted to see v2 in action rather than tinker with baselibs. Was able to
start it from the command classic line as well as through the pygrads
interpreter. (PyGrADS has limit functionality as I need the IPC extension to
exchange variables with python, something *not* used in pytests.) Yes, the
"-u" switch works out of the box. This means that the pytests should work
pretty much out of the box except for small adjustments for new executable
names, etc. However, I had to go to bed before I could play it.
My approach was as follows. The autoconf scripts in the v1.9 code base
appear in better shape than in v2.0.a1 (they are supplibs v2 ready, include
Pat's dynamic supplibs, and have other cleanup and more robust discovery
macros). Therefore, I started with our configure.ac and Makefile.am,
replaced all the files in src/, doc/, lib/ and etc/ with the
v2.0.a1versions. Then I merged the src/Makefile.am with the what we
had from
v1.9 and voila it built grads and gradsdap. I still need to do a bit clean
as I cut a couple of corners to get it done quickly. But the message is: we
should be able to v2.0.a1 building with the supplibs, Pat's dyn supplibs
feature, and pytests in a relatively short amount of time. This will allow
builds on all the platforms supported by the new supplibs, including
windows. Here is my proposed roadmap:
1. Place all current opengrads development under a GRADS1_DEV_BRANCH
2. Import GrADS v2.0.a1 into our CVS repo, in the usual COLA vendor branch,
release tag v2_0_a1
3. Patch the hdf4 in the current supplibs (I expect this to be a single line
change in the GNUmakefile) and rebuild them. Since I still have my sandboxes
around I may only have to rebuild hdf4 and re-tar.
4. Clean up what I did last night, making sure I have a full build
--with-hdf, and make sure "make check" and the pytests work. I'll need to
iterate with Pat to make sure the dynamic supplibs also work. I'll then
check this into the trunk, merging it with the COLA vendor branch. From that
point on, checking out module "Grads" will produce the v2 codebase.
At this point we can submit it to COLA as a patch or simply give them a tag
as Jennifer has repository privileges. If COLA is interested, I can help out
with builds at the platforms we have supplibs for; hoop is also setup to
build on solaris. My next priority would be to add the hooks for dynamically
linked extensions as this is the bread-and-butter of opengrads.
Brian/Jennifer: I could have a first crack at adding the dynamic UDFs/UDFs
hooks, and go over with you in a face-to-face meeting at COLA. Or else, I
can be available for questions if you would prefer to have a first crack
yourselves. Let me know.
Porting the current extensions to v2 deserves its own roadmap. Let's
postpone this discussion.
As always, any comments on this proposal is greatly appreciated.
Cheers,
Arlindo
--
Arlindo da Silva
da...@al...
|
