[Plplot-devel] Installed version of PLplot working again!

[Alan W. I. <Ala...@gm...>]

As of commit e298aa70a, the installed version of PLplot works again
after a long time (since early April) where it did not work.  This
commit completes a project to implement a highly recommended
modernization of our build system (i.e., to use a namespaced prefix
(of "PLPLOT::") for every target that is installed.  This commit
also solves some further "modern software" PLplot bugs that occur
for Debian Buster versions of free software libraries.

Preliminary work on this project started in March.  Although the
build-tree version of PLPlot worked throughout this time, my apologies
for the long down-time for the installed version of PLplot caused by
(1) the distraction of replacing my computer right in the middle of
this project and (2) dealing with the quite large number of "new
software" version issues caused by upgrading my software from Debian
Jessie to Debian Buster right in the middle of this project.

I got an unconstrained (i.e., virtually complete with both interactive
and noninteractive tests for both the core build tree and installed
examples trees for our three principal library/driver configurations)
comprehensive test of PLplot to work for me for Debian Buster so
I encourage you to make such a comprehensive test yourselves (likely with
"--do_test_interactive no" initially to save yourselves quite a bit of
"baby-sitting" to get through the test) and let me know of any issues
you find.

This good comprehensive test result for Debian Buster has some caveats.

* intermittent (UGH) plmeta error

There was an error concerning an attempted *interactive* test on the
(confirmed) disabled plmeta device for the CMake-based build system
for the installed examples.  Such tests should not occur because the
device is disabled and especially not for interactive tests since it
is a noninteractive device!  However, since that error was a
showstopper for that component of the overall comprehensive test, I
have confirmed all our device numbers were unique (see the above
commit message for details) to make sure no other device was being
misidentified as plmeta.  Furthermore, I have been completely unable
to trigger that error again with two similar comprehensive tests (one
parallel, one not) focussing on just that one component.  And also for one
exactly identical comprehensive test (parallel and for all
components).  So my plan is to wait until this nasty error triggers
again (if it ever does) and look deeper at it then.  So if you see
any "plmeta" testing issues in the near future for
the default case where plmeta is disabled, please let me know!

* Modern (Debian Buster) software bugs

   + Modern Ada gnatmake (version 7.3.0-2) has a bug where parallel
     builds don't work properly.  There is no way I want to give up the
     huge speed advantage of parallel builds so I have worked around
     this bug by disabling Ada for the above test.  I have made an
     unoffical query about this return of parallel gnatmake issues to
     Debian with no response so far so it is time to escalate this to
     an official Debian bug report and likely an official upstream
     gnatmake bug report as well.

   + Modern Lua 5.3.3 has a bug where if more than a modest number of
     arrays are initialized all arithmetic starts failing!  Only
     example 23 in the set of Lua examples exposes this bug so I have
     worked around the issue by temporarily dropping that example from
     the Lua tests.  I have reported this issue with a simple
     (completely independent of PLplot) test Lua script demonstrating
     this Lua issue as an official Debian bug report, but there has
     been no response so it is past time to take this upstream to the
     Lua developers.

   + Our extXdrawable_demo application is written for the ancient GTK+
     version 2 case and needs to be rewritten for the (modern) GTK+
     version 3 case.  Jonathan Woithe has kindly volunteered to do that
     rewrite, but until he completes his work, I have worked around the
     issue by disabling this example.

* Modern (Debian Buster) software warnings

   + OCaml build warnings

     These seem fairly innocuous, but this is a Debian Buster versus
     Debian Jessie regression that I need to investigate further.

   + I am now getting undefined symbol warnings from "ldd -r" tests
     done by the above script run.  These issues did not cause any
     obvious errors, but this is a Debian Buster versus Debian Jessie
     regression that I need to investigate further.

Now that the installed version of PLplot is largely working again and
both build-tree and install-tree PLplot more or less working for a
system (Debian Buster) with near cutting-edge software versions, I am
beginning to consider making a PLplot release.  My current plan is to
push quite a few further topics (all of them much shorter than the
present topic that was just finished) before that release which would
likely delay the release until at least December or possibly into
early 2019.  However, if someone here has a strong preference for an
earlier release than that sort of time scale, I would certainly be
willing to consider that possibility by dropping some of the topics
I want to work on before the release.

Alan
__________________________
Alan W. Irwin

Programming affiliations with the FreeEOS equation-of-state
implementation for stellar interiors (freeeos.sf.net); the Time
Ephemerides project (timeephem.sf.net); PLplot scientific plotting
software package (plplot.sf.net); the libLASi project
(unifont.org/lasi); the Loads of Linux Links project (loll.sf.net);
and the Linux Brochure Project (lbproject.sf.net).
__________________________

Linux-powered Science
__________________________