[Plplot-devel] The next step toward a new release of PLplot

[Alan W. I. <ir...@be...>]

Our next release date (January 2003 or later) is still up in the air because
the new linking scheme only works on Linux, and considerable effort is going
to be required to get it to work for all Unix platforms. (There are other
issues as well; please have a look at plplot/PROBLEMS.) However, I intend to
generalize the new linking scheme across Unix platforms shortly; my recent
astrophysical research effort was quite successful, but it is now winding
down, and I expect to have a lot more time for PLplot starting about a week
from now.

Here is the current status.  The current Linux linking has been set up in a
way that is largely compatible with what libtool expects.  The "long
shopping list of linking all possible libraries" scheme is gone and instead
each application or library is only linked against the specific libraries
that it needs in a hierarchical way with object code split amongst the
libaries in such a way that no cross-linking occurs between libraries.  This
scheme includes the linking of the dynamic devices which are themselves
(mini-) dynamically loaded libraries. The result of this new linking scheme,
as Maurice has noted, is much faster linking.  Also, this scheme allows for
the first time to have the tkwin, tk, and xwin devices treated as dynamic
devices.

So far this new linking scheme has only been implemented by hand for the
combination of Linux shared libraries and dynamic devices without use of
libtool at all.  We could continue this "by hand" approach to generalize the
linking scheme for some (but not all) of the other Unix platforms.  But this
would be a lot of work, and I would much rather spend my efforts making our
configuration work with automake and libtool.

These configuration tools have been adopted by a number of large (KDE, for
example) Linux projects to greatly simplify their configuration effort.
Also, there is a derived benefit that once we get the scheme to work on
Linux, it should automatically work on all Unix platforms which neatly
solves all our Unix cross-platform issues.  Thus, PLplot would greatly
benefit from using automake/libtool.  Rafael has long since come to the same
conclusion, and in fact pioneered a branch of an old version of PLplot which
used those tools for configuration.  Unfortunately, nobody else from the
development group got involved with that effort and eventually it came to a
standstill when Rafael had a new job to deal with.

What I propose to do starting late this week is to make a concentrated
effort to understand libtool, automake, and Rafael's branch with the goal of
(a) getting AM/LT configuration of our current CVS HEAD source code to work
on Linux, and (b) immediately after that bringing the AM/LT configuration
scheme to CVS HEAD for fuller group evaluation and participation in the new
scheme.  My judgement of when to bring it to CVS HEAD will follow what
happened with Geoffrey's dynamic driver effort; he brought it to CVS HEAD
once it was basically working on Linux.  There was a lot of group effort
afterward in polishing it which went quite well, and I hope something
similar happens in the AM/LT case once I bring it to CVS HEAD.

Through a concentrated effort I plan to get AM/LT configuration of PLplot to
the Linux working stage on a time scale of several weeks to a month or so.
***If anyone else wants to participate in this effort before I bring it to
CVS HEAD please get in touch with me so we can arrange how we will share the
effort.*** In particular, this effort will obviously go faster the more
Rafael is involved (in fact I would be glad if he would lead the project),
but I understand he may still have some heavy research commitments that
preclude this. Regardless of that question, PLplot configuration needs
(IMHO) to use the AM/LT approach, and I also have the time to deal with
changing to that approach now.  Thus, I plan to just go ahead with the
effort if Rafael cannot participate much at this time, but I hope he can.

Alan

email: ir...@be...
phone: 250-727-2902	FAX: 250-721-7715
snail-mail:
Dr. Alan W. Irwin
Department of Physics and Astronomy,
University of Victoria, P.O. Box 3055,
Victoria, British Columbia, Canada, V8W 3P6
__________________________

Linux-powered astrophysics
__________________________