plplot-devel — Technical discussion about PLplot development

[Plplot-devel] May I remove the undocumented page advance from plssub?

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

You have to be careful with plssub in a series of plots (such as
tcldemos.tcl or pythondemos.py) where plinit is called once before the
string of examples and plend called once afterward.  If the examples contain
a call to plssub you would like to return the plot to default state by
calling plssub(1,1) afterward (in the same example so the various examples
can be combined in any order).

Unfortunately, this idea is messed up because buried in the plssub code is
an undocumented page advance which generates extra empty pages (and also
perhaps non-standard postscript, because gv seems to get into a frozen state
if I attempt to view those empty pages.)  I don't really see the utility of
combining the subwindowing with a page advance. For example, you might want
to combine two kinds of sub-windowing on the same page. If our users really
do need a page advance after the call to plssub, then shouldn't they be
calling pladv themselves?

I have taken out the page advance buried in c_plssub, and all the ordinary
examples using plssub seem to be fine without it, and it solves the
problem of the empty pages.  Does anybody have an objection if
I commit that change to CVS?

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
__________________________

[Plplot-devel] png driver no longer works for python

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

With plplot tree freshly updated from CVS, I configure like I normally do, e.g.,

./configure --prefix=/usr/local/plplot --with-double
--enable-dynamic-drivers --enable-java --enable-gnome > & ! configure.out

Then from tmp I executed
./pythondemos.py -dev png
Unable to load driver: gd.drv.
Reason: /usr/local/plplot/lib/drivers/gd.drv: undefined symbol: plP_setpxl
Segmentation fault

First, why should a test in tmp refer to some previously installed version
in /usr/local/plplot?
Second, when I removed /usr/local/plplot/* I got a new error message:

./pythondemos.py -dev png -o temp.png
Unable to load driver: gd.drv.
Reason: /usr/local/plplot/lib/drivers/gd.drv: cannot open shared object
file: No such file or directory

So somehow /usr/local/plplot is hardcoded in the python case.  With
/usr/local/plplot removed, I have no trouble with the png driver for the
x01c example, and the psc driver works both for the x01c and pythondemos.py
examples.

I am virtually positive I checked the png driver before when I was first
testing pythondemos.py.  Since then, I have changed to python-2.1 and Joao
has made some driver changes so I don't know which is causing the trouble.
Joao, can you replicate this problem on your (presumably) python-1.5 system?

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
__________________________

[Plplot-devel] Win one, lose one

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

First the win:  Thanks Andrew for your fix on x19c.  It works like a charm
now!

Now the loss: xw09.py is segfaulting again.  The story behind this is there
have been some huge python and libqt changes on my research computer where I
test plplot.  Debian woody python has just migrated from python-1.5 to
python-2.1, and for the first time I am using the stock QT (2.3.1) and KDE
(2.2.1) on woody rather than versions which I compiled myself.  These two
changes meant I could also build and install (with some extra effort to
accomodate the new woody python) the woody versions of libsip, sip, and
python-pyqt packages rather than compiling my own tarballs.  Thus, I now
believe I have a fully integrated python-2.1/python-pyqt-2.5 environment,
and I intend to stick with it from now on for my tests. (Which reminds
me.... Allesandro, you still haven't answered me on the qplplot.py hack to
comment out self.__attr(*args) that I had to do for the python-1.5
environment.  Shall I put that back?  It seems to make no difference so I am
wondering if qplplot.py can be simplified by removing unused functions).

pythondemos.py with xw09 commented out shows that all xw??.py demos work
fine with that one exception which is segfaulting again just like it did for
the experimental 2.0 python environment.  I am particular sensitive to bugs
that screw up demo programmes so I will get to the bottom of this before the
plplot release I hope to do in December.  I also plan to make some further
changes to the xw??.py demos to bring them more in line with the x??c.c
demos, and I still want to complete the java demos (aside from the required
API additions).

The pyqt GUI (aside from xw09) works well for the new python/python-pyqt
test environment.  I encourage Rafael to give this a try since he also has
access to woody.  (Contact me off list for the tricks I played to get Debian
woody sip, etc. to play nice with the new python.  I have a wishlist bug
report in for sip so the tricks may not be required for too long.)

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
__________________________

Re: [Plplot-devel] Plplot embedded in pyqt + plimage function

[Joao C. <jca...@in...>]

On Thursday 22 November 2001 17:16, Alessandro Mirone wrote:
| Hi Jo=E3o,
|  it's great that you got plimage working!!
| I will certainly send a description of plimage.
|
| I agree that plmoveimage is not necessary and
| we can discard it by the moment.

OK.

| Yes it is a good idea to collapse all those PLESC.
|
| I would like to keep the PLESC that disactivate the pixmap
| and metabuffer.=20

The PLESC_DOUBLEBUFFERING already does "write_to_window =3D 0/1"

|That is because I am planning to
| write things like xhairs ( or whatever shape )
| from python.

xhairs already exists, as you might know, use plGetCursor()

| That lets you the freedom to ( just imagine)
| select a polygonal shape from an image for an integral
| or whatever having at the same time a visual representetion
| of the shape. The shape changes continuosly and might be
| very general so it's good to manage that from python.

You could as well just use plGetCursor() to retrieve the shape, then plot=
 it=20
in xor mode, try "./x01c -dev xwin -xor".

| About the call to plimage we can discuss the arguments..
| I have no particular  preference so may be we could generate an
| analogous of
| pgimage.

I used pgplot years ago, so I don't remember how pgimage() works.
Why not use the standard plplot way,  "PLFLT **z"? It is a bit more diffi=
cult=20
to dinamicaly allocate memory, but one could write a utility function for=
 it=20
(as well as an utility to return the maximum and minimum of such arrays,=20
which I believe is a very frequent operation on plplot and particularly i=
n=20
plimage related functions.

| What do you plan todo? You commit a working plimage
| and the we refine the plesc? Or you want me to refine the plescs?

I will commit my changes (in a couple of days from now), and we will latt=
er=20
refine some details. OK?

| As you prefer. By the moment I'll send a description of plimage
| to generate the html

Nice. No job is finish till the paper work is done :) (aarrrghhhh--agains=
t me=20
I speek!)

| Ciao
|    Alessandro

Bye,
Joao

Re: [Plplot-devel] Plplot embedded in pyqt + plimage function

[Alessandro M. <mi...@es...>]

Hi Jo=E3o,
 it's great that you got plimage working!!
I will certainly send a description of plimage.

I agree that plmoveimage is not necessary and =

we can discard it by the moment.

Yes it is a good idea to collapse all those PLESC.

I would like to keep the PLESC that disactivate the pixmap
and metabuffer. That is because I am planning to =

write things like xhairs ( or whatever shape )
from python. That lets you the freedom to ( just imagine)
select a polygonal shape from an image for an integral
or whatever having at the same time a visual representetion
of the shape. The shape changes continuosly and might be
very general so it's good to manage that from python.


About the call to plimage we can discuss the arguments..
I have no particular  preference so may be we could generate an
analogous of
pgimage.


What do you plan todo? You commit a working plimage
and the we refine the plesc? Or you want me to refine the plescs?

As you prefer. By the moment I'll send a description of plimage =

to generate the html

Ciao
   Alessandro



Jo=E3o Cardoso wrote:
> =

> On Thursday 01 November 2001 22:06, Alessandro Mirone wrote:
> | Hello Alan,
> ...
> | If you agree I would like to submit to  your kind evaluation, in
> | the next weeks, small bunches of code ( as you suggested)  in order
> | to implement step by step a plimage that fits our needs and
> | hopefully those of the plplot community.
> =

> I don't know if there has been some further particular conversations
> with Alan about plimage. I have applied your original patch and,
> after some modifications, got plimage to work (I created a x20c.c
> demo). But I need some clarifications and further work from you,
> before I commit the changes to cvs.
> =

> 1-We need a description of usage, including the arguments
> descriptions. See for example
> http://plplot.sourceforge.net/resources/docbook-manual/plplotdoc-html-0=
=2E4.2/plshade1.html
> (but plain text is OK).
> =

> 2-Your patch includes a plmoveimage() and associated functions and
> PLESC driver functions. These are only used in qplplot.py. Are these
> functions really necessary? For one thing, one is introducing a
> function in plcore.c that is only meaningfull for plimage and a
> particular driver, and for the other side there are lots of PLESCs
> (if necessary, those could be colapsed just in one
> PLESC_PLIMAGEOPS(arg)). Also, the "move" operation is not very usual
> in plots. What is its purpose?
> =

> 3-In plimage() you use a row vector as argument for the image, but
> plplot generally uses a 2D matrix for similar calls. Is there any
> reason for this?
> =

> I am willing to commit my changes to cvs, without plmoveimage(), and
> we could latter solve some other issues. What do you thing? Please
> anwser to the list.
> =

> Thanks,
> Joao
> =

> | Best regards
> |
> |      Alessandro Mirone
> |
> |
> | _______________________________________________
> | Plplot-devel mailing list
> | Plp...@li...
> | https://lists.sourceforge.net/lists/listinfo/plplot-devel

Re: [Plplot-devel] Plplot embedded in pyqt + plimage function

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

On Thu, 22 Nov 2001, [iso-8859-1] Jo=E3o Cardoso wrote:

> On Thursday 01 November 2001 22:06, Alessandro Mirone wrote:
> | Hello Alan,
> ...
> | If you agree I would like to submit to  your kind evaluation, in
> | the next weeks, small bunches of code ( as you suggested)  in order
> | to implement step by step a plimage that fits our needs and
> | hopefully those of the plplot community.
>
> I don't know if there has been some further particular conversations
> with Alan about plimage.

No because I don't have the qualifications to deal with it.  So it is great
that Jo=E3o has been able to make so much progress with plimage.

> 1-We need a description of usage, including the arguments
> descriptions. See for example
> http://plplot.sourceforge.net/resources/docbook-manual/plplotdoc-html-0.4=
=2E2/plshade1.html
> (but plain text is OK).

Here, I can help.  That html file was created from docbook source, see
plplot/doc/docbook/src/api.xml.  So if Allesandro creates a description of
of plimage using any function from that file such as plshade1 as a template
that would be great.  If the description is submitted as ascii, I would als=
o
be willing to change it to the proper form for inclusion in api.xml.

Alan

email: ir...@be...
phone: 250-727-2902=09FAX: 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
__________________________

Re: [Plplot-devel] Plplot embedded in pyqt + plimage function

[<jca...@in...>]

On Thursday 01 November 2001 22:06, Alessandro Mirone wrote:
| Hello Alan,
=2E..
| If you agree I would like to submit to  your kind evaluation, in
| the next weeks, small bunches of code ( as you suggested)  in order
| to implement step by step a plimage that fits our needs and
| hopefully those of the plplot community.

I don't know if there has been some further particular conversations=20
with Alan about plimage. I have applied your original patch and,=20
after some modifications, got plimage to work (I created a x20c.c=20
demo). But I need some clarifications and further work from you,=20
before I commit the changes to cvs.

1-We need a description of usage, including the arguments=20
descriptions. See for example  =20
http://plplot.sourceforge.net/resources/docbook-manual/plplotdoc-html-0.4=
=2E2/plshade1.html
(but plain text is OK).

2-Your patch includes a plmoveimage() and associated functions and=20
PLESC driver functions. These are only used in qplplot.py. Are these=20
functions really necessary? For one thing, one is introducing a=20
function in plcore.c that is only meaningfull for plimage and a=20
particular driver, and for the other side there are lots of PLESCs=20
(if necessary, those could be colapsed just in one=20
PLESC_PLIMAGEOPS(arg)). Also, the "move" operation is not very usual=20
in plots. What is its purpose?

3-In plimage() you use a row vector as argument for the image, but=20
plplot generally uses a 2D matrix for similar calls. Is there any=20
reason for this?

I am willing to commit my changes to cvs, without plmoveimage(), and=20
we could latter solve some other issues. What do you thing? Please=20
anwser to the list.

Thanks,
Joao

| Best regards
|
|      Alessandro Mirone
|
|
| _______________________________________________
| Plplot-devel mailing list
| Plp...@li...
| https://lists.sourceforge.net/lists/listinfo/plplot-devel

Re: [Plplot-devel] prova.py

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

On Fri, 16 Nov 2001, Alessandro Mirone wrote:

> Hallo,
>  I have fixed the behaviour of prova.py
> concerning the loading of xw??.py.
> In particular the exec statement now runs properly
>       ( for python expert I passed the globals() name space
>       to exec so that the functions defined in xw??.py
>        dont get fooled by different scopes)

Now in cvs with one change.  xw??.py --> *.py (which allows access to all
the python scripts in the current directory which I believe is a better
approach).  I too noticed that you could type import xw08 (for example)
before and run it, but you could not open xw08 with the widget and run it.
Now you can.  Thanks very much for this fix to the problem.

> I also added few lines for python2.1 in the configure files.

Now in CVS (although I also hope you try out python-1.5).

>
> I also had a problem compiling with --with-double=yes (necessary
> for Numeric) :
>  I had to copy manually
>          cp tmp/strutil.o tmp/shared/
> or libplplotd.so did'nt get compiled
>

Discussed elsewhere with question for you about whether you still see this
problem.

> Alan
> , you wish I could fix the page control for the qt widget.
>  If I understand well you mean the interacting behaviour, isn't it?
> When you click, or press a key a new example page is plotted, right?
> Concerning, this point,  interactivity, I think that a lot of thing may be
> done, bypassing plplot and using python.

For an example of the problem, try xw08.py.  (This problem also occurs in
xw01.py, but it goes by so fast you literally cannot see it.) How do you get
access to the various pages of xw08 (and other multi-page examples)? For a
KDE solution of a similar problem, look at how kview allows you to go
forward and back between various image files using arrow keys on the
toolbar.

./pythondemos.py -fam -fflen 2 -dev png -o test.png

kview test.png.*

(Be sure to enable the toolbar so you can see the arrows used for moving
between files.) It would be nice to have something similar for the pyqt
widget to navigate between various pages of an image.  Viewing various pages
of an image is, of course, different than viewing various image files, but the
widget part of it is similar.

Yes, please use the pyqt facilities to provide page access if that is possible.
(I am assuming here that the pyqt widget has this possibility.)

If this is difficult and not obvious, then I don't mind if you put it off
until after the release.  Probably a higher priority at the moment is getting
everything working cleanly under python-1.5 (note my request for your
comments on my hack to qplplot.py so it would work under python-1.5).

Thanks once again for your help.

Alan

Re: [Plplot-devel] prova.py

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

On Fri, 16 Nov 2001, Alessandro Mirone wrote:

> I also had a problem compiling with --with-double=yes (necessary
> for Numeric) :
>  I had to copy manually
>          cp tmp/strutil.o tmp/shared/
> or libplplotd.so did'nt get compiled

Yes, I had that exact same problem and fixed it in CVS with a new objs.in.
(I may not have commented, here, but you should subscribe to plplot_cvs to
get the check-in messages where I did comment.)

With clean checkout of current CVS do you still have the same problem? If
so, perhaps the make depend trick I used to generate the objs.in in the
current CVS is covering up some fundamental problem on my machine but not
yours, that we should fix.

I will answer your other questions/comments, later, after I have tried your
patch.  (BTW, the future of sourceforge is a little iffy so conceivably the
patches you post using the SF patch facility will be lost in the future.  As
a backup to the SF plplot_devel mail archives (for the same reason of an
iffy SF future) I am saving most plplot_devel, plplot_general, and
plplot_cvs posts.  So if you simply attach your patch to your plplot_devel
post, there is a good chance it will be saved even if SF folds.

Alan

[Plplot-devel] prova.py

[Alessandro M. <mi...@es...>]

Hallo,
 I have fixed the behaviour of prova.py 
concerning the loading of xw??.py.
In particular the exec statement now runs properly
      ( for python expert I passed the globals() name space 
      to exec so that the functions defined in xw??.py 
       dont get fooled by different scopes)

You can find the patch at sf.net in the patch database.

I also added few lines for python2.1 in the configure files.

I also had a problem compiling with --with-double=yes (necessary 
for Numeric) :
 I had to copy manually 
         cp tmp/strutil.o tmp/shared/
or libplplotd.so did'nt get compiled

Alan
, you wish I could fix the page control for the qt widget.
 If I understand well you mean the interacting behaviour, isn't it?
When you click, or press a key a new example page is plotted, right?
Concerning, this point,  interactivity, I think that a lot of thing may be
done, bypassing plplot and using python. 
 If you agree I can  apport my original solution, may be after next
release,

Best regards

               Alessandro Mirone
 

[Plplot-devel] reverting to python-1.5 which solves xw09 problem but which requires associated pyqt GUI change

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

I did a lot of printing out of variables in plmodule2.c, and finally pinned
down the xw09 segfault there to the line:

if ( o && PyArray_Check(o) )

(xw09.py calls plcont which in turn exercises this logic).

If I commented out the call to PyArray_Check, the plot was fine, and there
was no segfault.  I have no idea why this code is not working under
python-2.0, but that is an experimental version under Debian which I presume
is not maintained at all since the python-2 future for Debian is taking a
quite different path. Thus, I don't really trust the "orphaned" Debian
python-2.0 package, and I decided to revert to python-1.5 until python-2
becomes the standard (as opposed to experimental) Debian version of python.
At that point, the xw09.py segfault may come back, but I think it is fairly
likely that it won't.  Anyhow, we'll cross that bridge if/when we come to
it.

I rebuilt sip and pyqt using python-1.5, and similarly rebuilt plplot
for python-1.5.  pythondemos.py now works fine with all examples (i.e.
the PyArray_Check logic above works fine when called by plcont which
in turn is called by xw09.py.)

However, under python-1.5, prova.py quit working with the following error
message:

./prova.py
Traceback (innermost last):
  File "./prova.py", line 20, in ?
    import qplplot
  File "/home/software/plplot_cvs/HEAD/plplot/tmp/qplplot.py", line 39
    self.__attr(*args)
                ^
SyntaxError: invalid syntax

As a brute-force workaround, I commented out this line in qplplot.py (this
change now in CVS), and the pyqt GUI for plplot now seems to work fine
again.  So, Allesandro, is this code (which appears inside __wrap) not
actually used?  Let me know what you think is the best solution here since
we probably don't want this temporary workaround to become permanent.  Also,
from now on since I am reverting to python-1.5, I would very much appreciate
it if you tested all your coming changes to the pyqt GUI for plplot in a
python-1.5 environment.  I don't plan to support python-1.5 indefinitely,
but it still is the usual environment you find for most Linux distributions
so I think we should support it for now.

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
__________________________

Re: [Plplot-devel] How can java be told permanently about where to find libplplot(d)?

[Geoffrey F. <fu...@ga...>]

Alan W. Irwin writes:
 > On Tue, 13 Nov 2001, [iso-8859-1] Jo=E3o Cardoso wrote:
 > > I don't know if this (LD_RUN_PATH) will work for the java demos, b=
ut I guess that
 > > it will.
 >=20
 > Thanks for that suggestion, but it did not work.  I suspect LD_RUN_P=
ATH is
 > specific to ld (or ld called implicitly by gcc).  man ld is the only=
 place I
 > can find documentation for it.  It is indeed that sort of functional=
ity (or
 > the sys.path.insert() functionality in python) that I am needing for=
 java,
 > but I haven't found such functionality, yet.

It would have to be applied ot the jvm in order for it to be active
when the jvm goes to look to dlopen libplplot.so.  We can't control
how the jvm is formed.  I'd say we're sunk in this regard.

I suppose this is as good a place as any to just point out that I
don't actually share your aversion to setting LD_LIBRARY_PATH.  You
have to set CLASSPATH to find .class files or .jar files, and you have
to set LD_LIBRARY_PATH to find .so's.  Doesn't seem like much of an
issue to me, for the benefit of loading JNI code bases.

 > Your suggestion did point me toward gcj which is a java compiler tha=
t is a
 > variation on gcc.  Once I can get that compiler to work with our cod=
e, it
 > may well be that LD_RUN_PATH is automatically honored since there is=
 an
 > implicit call to ld from gcc.  I like the idea that all of our C fro=
nt end,
 > our c++ front end and our java front end can potentially be compiled=
 under
 > gcc. So that is an additional reason to get gcj to work with our jav=
a code.
 >=20
 > However, I am having trouble.  Currently gcj generates tons of undef=
ined
 > symbols messages, but I suspect I am just not using the correct opti=
ons
 > since the symbols I have looked at are the plplot ones that should b=
e there
 > if it is finding our library at link time (deja vu all over again, s=
ort
 > of!).  Geoffrey, have you tried out gcj, and if so, how do you get i=
t to
 > work with our code?

I have not yet tried gcj, or jikes.  It would be nice to work with
both of these, but I haven't had time to investigate such matters, and
cannot promise to ever do it.  Work is getting ever more demanding, so
my ability to do this kind of exploratory investigation is very
limited.=20

RE: [Plplot-devel] How can java be told permanently about where to find libplplot(d)?

[Geoffrey F. <fu...@ga...>]

Alan W. Irwin writes:
 > On Mon, 12 Nov 2001, Geoffrey Furnish wrote:
 > 
 > > Alan W. Irwin writes:
 > >  > I agree there is nothing we can do with libplplot.  What I am asking is
 > >  > whether there is a mechanism *under java* (some additional command in
 > >  > PLStream.java or some javac option) which allows java to find libplplot (or
 > >  > any other C extension library) when it is installed in a non-standard place.
 > >  > I would be surprised and disappointed if java relied solely on
 > >  > LD_LIBRARY_PATH to solve this problem.
 > >
 > > Well, it's not anything special to java.  We're just talking about the
 > > semantics of dlopen(), which all programs use to load .so's.
 > 
 > >From what you have found out about dlopen, it looks like you can do anything
 > with it.  Thus, looking at this from the java level, java is free to do
 > anything before the low-level call to dlopen.  So it really becomes a
 > question of how java integrates the specification of library names using
 > System.loadLibrary with their treatment (if any) of the library path.
 > 
 > I did try your suggestion of using an absolute path as part of the library
 > name. (That is quite easy to do in a configurable way by making the
 > appropriate modification to config.java which in turn is accessed by the
 > System.loadLibrary call in PLStream.java.) Alas, this idea did not work. The
 > run time error was as follows:
 > 
 > Exception in thread "main" java.lang.UnsatisfiedLinkError: Directory
 > separator should not appear in library name:
 > /home/software/plplot_cvs/HEAD/plplot/tmp/plplotd

Yikes.

 > So apparently there is some name interpretation going on before the dlopen
 > call, and as part of that interpretation java insists the path must not be
 > part of the name specified to System.loadLibrary.

Well, that's pretty much a death knell (sp?) for your goal.

 > Of course, this still leaves wide open the question about whether there
 > might be some alternative mechanism in java to specify the path part of the
 > name since java can do anything it likes before the dlopen call. Thus, what
 > I am looking for is the java analog of the python sys.path.insert() function
 > that we use in pythondemos.py to manipulate the path where python looks for

I don't know of anything like that.

 > plmodules.so. Unfortunately, I am now stuck because I don't know how to
 > search for such functionality in the enormous number of java functions that
 > are available. Geoffrey, are you aware of a general site where I could
 > specify say "library path" as a search string and it would find all the java
 > functions that have something to do with the library path?  The java

All I have used is the java.sun.com site, which I suppose is the one
you're referring to.

 > tutorial does have a search engine so I tried that, but all it came up with
 > was the LD_LIBRARY_PATH idea.  I suppose it is possible that java has no way
 > to specify the library path other than LD_LIBRARY_PATH, but that is quite
 > limited compared to the sys.path.insert() function available for python.

I think this is where we say, "Welcome to the wonderful world of
Java."

Re: [Plplot-devel] How can java be told permanently about where to find libplplot(d)?

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

On Tue, 13 Nov 2001, [iso-8859-1] Jo=E3o Cardoso wrote:
> I don't know if this (LD_RUN_PATH) will work for the java demos, but I gu=
ess that
> it will.

Thanks for that suggestion, but it did not work.  I suspect LD_RUN_PATH is
specific to ld (or ld called implicitly by gcc).  man ld is the only place =
I
can find documentation for it.  It is indeed that sort of functionality (or
the sys.path.insert() functionality in python) that I am needing for java,
but I haven't found such functionality, yet.

Your suggestion did point me toward gcj which is a java compiler that is a
variation on gcc.  Once I can get that compiler to work with our code, it
may well be that LD_RUN_PATH is automatically honored since there is an
implicit call to ld from gcc.  I like the idea that all of our C front end,
our c++ front end and our java front end can potentially be compiled under
gcc. So that is an additional reason to get gcj to work with our java code.

However, I am having trouble.  Currently gcj generates tons of undefined
symbols messages, but I suspect I am just not using the correct options
since the symbols I have looked at are the plplot ones that should be there
if it is finding our library at link time (deja vu all over again, sort
of!).  Geoffrey, have you tried out gcj, and if so, how do you get it to
work with our code?

Alan

Re: [Plplot-devel] How can java be told permanently about where to find libplplot(d)?

[<jca...@in...>]

On Sunday 11 November 2001 23:25, Alan W. Irwin wrote:
| Our java examples work only if LD_LIBRARY_PATH is set, but I
| *really* dislike the necessity of having to set this variable. (I
| believe Joao has also commented on this problem.)

I don't recall it.

The Octave bindings can't use (in a straighforward way) "-rpath", and=20
so I use LD_RUN_PATH instead of LD_LIBRARY_PATH. The difference is=20
that LD_RUN_PATH is stored in the executable. In cf/pkg_octave.in:

# 'mkoctfile' does not accept -rpath, use LD_RUN_PATH

plplot_octave.oct plplot_octave.o: plplot_octave.cc
   LD_RUN_PATH=3D`pwd` $(MKOCTFILE) -v -I. -L. plplot_octave.cc ...

install_octave: ...
=2E...
   LD_RUN_PATH=3D$(LIB_DIR) $(MKOCTFILE) -v -I. -L. plplot_...

I don't know if this will work for the java demos, but I guess that=20
it will,

Joao

| Essentially, we
| need some java equivalent to the -rpath option so that libplplot(d)
| is always found by java. From the description I thought the
| -extdirs option on javac command might do the job, but it doesn't
| (according to my tests). Perhaps we need a command inside
| PLStream.java?
|
| Once somebody informs me of the proper way to tell java permanently
| how to find our library, I am willing to do the configuration work
| so that this happens automatically both for the tmp directory and
| the installed directory similarly to the way we handle rpath now.
|
| Alan
|
| email: ir...@be...
| phone: 250-727-2902=09FAX: 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
| __________________________
|
|
| _______________________________________________
| Plplot-devel mailing list
| Plp...@li...
| https://lists.sourceforge.net/lists/listinfo/plplot-devel

RE: [Plplot-devel] How can java be told permanently about where to find libplplot(d)?

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

On Mon, 12 Nov 2001, Geoffrey Furnish wrote:

> Alan W. Irwin writes:
>  > I agree there is nothing we can do with libplplot.  What I am asking is
>  > whether there is a mechanism *under java* (some additional command in
>  > PLStream.java or some javac option) which allows java to find libplplot (or
>  > any other C extension library) when it is installed in a non-standard place.
>  > I would be surprised and disappointed if java relied solely on
>  > LD_LIBRARY_PATH to solve this problem.
>
> Well, it's not anything special to java.  We're just talking about the
> semantics of dlopen(), which all programs use to load .so's.

From what you have found out about dlopen, it looks like you can do anything
with it.  Thus, looking at this from the java level, java is free to do
anything before the low-level call to dlopen.  So it really becomes a
question of how java integrates the specification of library names using
System.loadLibrary with their treatment (if any) of the library path.

I did try your suggestion of using an absolute path as part of the library
name. (That is quite easy to do in a configurable way by making the
appropriate modification to config.java which in turn is accessed by the
System.loadLibrary call in PLStream.java.) Alas, this idea did not work. The
run time error was as follows:

Exception in thread "main" java.lang.UnsatisfiedLinkError: Directory
separator should not appear in library name:
/home/software/plplot_cvs/HEAD/plplot/tmp/plplotd

So apparently there is some name interpretation going on before the dlopen
call, and as part of that interpretation java insists the path must not be
part of the name specified to System.loadLibrary.

Of course, this still leaves wide open the question about whether there
might be some alternative mechanism in java to specify the path part of the
name since java can do anything it likes before the dlopen call. Thus, what
I am looking for is the java analog of the python sys.path.insert() function
that we use in pythondemos.py to manipulate the path where python looks for
plmodules.so. Unfortunately, I am now stuck because I don't know how to
search for such functionality in the enormous number of java functions that
are available. Geoffrey, are you aware of a general site where I could
specify say "library path" as a search string and it would find all the java
functions that have something to do with the library path?  The java
tutorial does have a search engine so I tried that, but all it came up with
was the LD_LIBRARY_PATH idea.  I suppose it is possible that java has no way
to specify the library path other than LD_LIBRARY_PATH, but that is quite
limited compared to the sys.path.insert() function available for python.

Alan

RE: [Plplot-devel] How can java be told permanently about where to find libplplot(d)?

[Geoffrey F. <fu...@ga...>]

Alan W. Irwin writes:
 > On Mon, 12 Nov 2001, Geoffrey Furnish wrote:
 > 
 > > The jvm has already been linked by the time we get it, there isn't
 > > much we can do about it.  When you dlopen something in the jvm, it's
 > > gonna look wherever it looks.  You can influence this by
 > > LD_LIBARY_PATH, or by editng ld.so.conf or whatever.  That's to /find/
 > > libplplot.so.
 > 
 > Yes, that is the crux of the issue.  I am interested in non-LD_LIBRARY_PATH
 > options (if they exist) for solving this problem under java.
 > 
 > >
 > > ....In other words, I think you are expressing a beef with the JVM, not
 > > with plplot.  If I've misunderstood, please clarify.
 > 
 > I agree there is nothing we can do with libplplot.  What I am asking is
 > whether there is a mechanism *under java* (some additional command in
 > PLStream.java or some javac option) which allows java to find libplplot (or
 > any other C extension library) when it is installed in a non-standard place.
 > I would be surprised and disappointed if java relied solely on
 > LD_LIBRARY_PATH to solve this problem.

Well, it's not anything special to java.  We're just talking about the
semantics of dlopen(), which all programs use to load .so's.  At the
top of man dlopen, I find:

       dlopen loads a dynamic library from the file named by  the
       null  terminated  string  filename  and  returns an opaque
       "handle" for the dynamic library.  If filename is  not  an
       absolute  path  (i.e., it does not begin with a "/"), then
       the file is searched for in the following locations:

              A colon-separated list of directories in the user's
              LD_LIBRARY path environment variable.

              The     list     of    libraries    specified    in
              /etc/ld.so.cache.

              /usr/lib, followed by /lib.

I take this to mean you /can/ give it an absolute path.  Although I
have never actually tried that myself.  So, I guess it would be
possible for PLplot's install target to sed the plconfig.java file so
that it traded loadLibrary( "plplot" ); for something totally specific
like System.loadLibrary( "$prefix/lib/libplplot.so" );.  Maybe that
would work.

Note, however, that the java user will still have to add
$prefix/lib/java to CLASSPATH.  But if that much is done, then perhaps
the plconfig that is loaded there, would be able to find libplplot.so
using the full path spec sed'd in at install time.

If you wanted to try this out, note that you could just edit
plconfig.java by hand after installation, just to see.  But remember
to recompile plconfig.java afterwards.  If this approach does work,
we'd have to think a bit to figure out exactly how to feed this
approach into an improved install target.  We wouldn't want to sed the
plconfig.java in the tmp dir.  So we'd have to set up a separate build
tree for all the java stuff, sed that, then compile it, then copy it
all into place (or use a .jar file).

Re: [Plplot-devel] PLPlot on MS Windows

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

On Mon, 12 Nov 2001, Alan W. Irwin wrote:

> Thanks in advance for your win32 efforts with the latest cvs release.  I
> hope they go smoothly so we can get them into the next release.

OOPS.  That first line should read
Thanks in advance for your win32 efforts with the latest cvs.

Alan

RE: [Plplot-devel] How can java be told permanently about where to find libplplot(d)?

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

On Mon, 12 Nov 2001, Geoffrey Furnish wrote:

> The jvm has already been linked by the time we get it, there isn't
> much we can do about it.  When you dlopen something in the jvm, it's
> gonna look wherever it looks.  You can influence this by
> LD_LIBARY_PATH, or by editng ld.so.conf or whatever.  That's to /find/
> libplplot.so.

Yes, that is the crux of the issue.  I am interested in non-LD_LIBRARY_PATH
options (if they exist) for solving this problem under java.

>
> ....In other words, I think you are expressing a beef with the JVM, not
> with plplot.  If I've misunderstood, please clarify.

I agree there is nothing we can do with libplplot.  What I am asking is
whether there is a mechanism *under java* (some additional command in
PLStream.java or some javac option) which allows java to find libplplot (or
any other C extension library) when it is installed in a non-standard place.
I would be surprised and disappointed if java relied solely on
LD_LIBRARY_PATH to solve this problem.

Alan

Re: [Plplot-devel] PLPlot on MS Windows

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

On Mon, 12 Nov 2001, Olof Svensson wrote:

> Hi Alan,
>
> I'm happy to try to get the latest cvs plplot version working under
> win32. When are you planning to release a new stable release of plplot?
>
> Regards,
>
> Olof

Thanks in advance for your win32 efforts with the latest cvs release.  I
hope they go smoothly so we can get them into the next release.

I want to emphasize that we are a small developer group with good e-mail
communications between us so that allows a rather informal release process
(which I think is a good thing). For example, the release timing is usually
not fixed until the last several days depending on last-minute decisions
about what people want to get into the release, and I tend to declare a
freeze to cvs changes only for the few hours it takes me to test and get out
the release.

Let's aim for a month from now for the release.  Mid-December should keep us
away from Christmas travel disruptions and should give us plenty of time to
fix the only two bugs (segfault for xw09 and half the map missing for the
3rd page of the world map demo, x19c) in the latest cvs version that were
not in 5.0.4. The 4-5 weeks until the release gives developers a chance to
add some additional features as well.  For example, I hope to add more java
examples and extend the python examples, and certainly improvements to the
pyqt GUI and getting plplot to work on w2K would be most welcome for the
release.

Alan

RE: [Plplot-devel] How can java be told permanently about where to find libplplot(d)?

[Geoffrey F. <fu...@ga...>]

Alan W. Irwin writes:
 > Our java examples work only if LD_LIBRARY_PATH is set, but I *really*
 > dislike the necessity of having to set this variable. (I believe Joao has
 > also commented on this problem.) Essentially, we need some java equivalent
 > to the -rpath option so that libplplot(d) is always found by java. From the
 > description I thought the -extdirs option on javac command might do the job,
 > but it doesn't (according to my tests). Perhaps we need a command inside
 > PLStream.java?
 > 
 > Once somebody informs me of the proper way to tell java permanently how to
 > find our library, I am willing to do the configuration work so that this
 > happens automatically both for the tmp directory and the installed directory
 > similarly to the way we handle rpath now.

The jvm has already been linked by the time we get it, there isn't
much we can do about it.  When you dlopen something in the jvm, it's
gonna look wherever it looks.  You can influence this by
LD_LIBARY_PATH, or by editng ld.so.conf or whatever.  That's to /find/
libplplot.so.  Adjusting how libplplot.so is linked, will determine
whether we need to set LD_LIBRARY_PATH just to find whatever
libplplot.so is linked to (or dlopen's itself), but I don't see how
any action taken in the preparation of libplplot.so itself, will ever
even conceivably influence how the jvm finds things it loads.

In other words, I think you are expressing a beef with the JVM, not
with plplot.  If I've misunderstood, please clarify.

-- 
Geoffrey Furnish         fu...@ga...

Re: [Plplot-devel] PLPlot on MS Windows

[Olof S. <sve...@es...>]

Hi Alan,

I'm happy to try to get the latest cvs plplot version working under
win32. When are you planning to release a new stable release of plplot?

Regards,

Olof


"Alan W. Irwin" wrote:
> 
> On Fri, 9 Nov 2001, Alan W. Irwin wrote:
> 
> > Hello Olof:
> >
> > Our windows expertise is limited so we welcome any help we can get with the
> > windows platform for plplot.  So thanks very much for your patches to the
> > two files. They are isolated to the win32 directory and only a few lines so
> > unless any expert here (which I don't pretend to be) sees a problem, I will
> > put them into cvs.
> 
> I have thought better of immediately putting this patch into CVS.  The
> current cvs version of PLplot is different from 5.0.4 in that a mechanism
> has been introduced to allow dynamically loaded drivers.  The old static
> method is still available as well, but I am not sure whether or not anything
> extra will have to be done in the win32 tree to continue to use the static
> method. Anyhow, Olof, would you be willing to try out the plplot cvs version
> under w2k?  If the required changes to get it working are as simple as in
> the 5.0.4 case, I will be happy to put those changes into the CVS version.
> Such effort would be well worthwhile since I am planning to bring out a new
> stable release of plplot in the near future based on the current CVS
> version, and it would be nice for me to be able to state that you have
> introduced changes to make it work under w2k.
> 
> To get the latest cvs version of plplot follow the anonymous cvs
> instructions at http://sourceforge.net/cvs/?group_id=2915.  For the second
> cvs command, "modulename" should be plplot.  The result will be a directory
> tree called plplot in your current directory so make sure nothing in your
> current directory is called plplot before you invoke that second command.
> 
> I am also most interested (for both 5.0.4 and the present cvs version) whether
> you can build all the test plots using plplot-test.sh under w2k.
> 
> Alan

[Plplot-devel] How can java be told permanently about where to find libplplot(d)?

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

Our java examples work only if LD_LIBRARY_PATH is set, but I *really*
dislike the necessity of having to set this variable. (I believe Joao has
also commented on this problem.) Essentially, we need some java equivalent
to the -rpath option so that libplplot(d) is always found by java. From the
description I thought the -extdirs option on javac command might do the job,
but it doesn't (according to my tests). Perhaps we need a command inside
PLStream.java?

Once somebody informs me of the proper way to tell java permanently how to
find our library, I am willing to do the configuration work so that this
happens automatically both for the tmp directory and the installed directory
similarly to the way we handle rpath now.

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
__________________________

Re: [Plplot-devel] Plplot embedded in pyqt

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

On Sat, 10 Nov 2001, Alan W. Irwin wrote:
> (6) cd tmp; make pyqt_plmodule.so
>
> This makes the additional module that you need for the pyqt GUI. This module
> will be automatically made under configuration control in near future, i.e.,
> the "make" command in (5) will soon make this module if pyqt is available.

I have now commited changes to cvs so pyqt_plmodule.so is built
automatically. (I also fixed a couple of bugs in the configuration that I
had managed to introduce.)

My overall impression is the pyqt GUI for plplot lacks features (such as
page control) at the moment, but it is a good start, and also it has
astounding speed compared to any of our other GUI's.

So if you want to see this speed for yourself, build and install plplot with
python support like you normally would.  Then simply run the example
prova.py from either the tmp directory or the installed examples directory
(/usr/local/plplot/lib/plplot5.0.4/examples/python on my system).  If you
have pyqt installed, prova.py will work, otherwise it won't be able to find
the qt module and will quit with an error message.  If you get an error
message having to do with libqt, then set LD_LIBRARY_PATH to point to the
directory where the libqt library is located (this is due to a rough spot in
the way pyqt is linked but I am preparing a wishlist bug report on that so
hopefully it will go away with later versions). Once you have the GUI
running, type in import xw01 (or whatever xw?? example you want to try).

Feedback, please!  (Especially on features that you would like this GUI to
have.)

Alan

[Plplot-devel] Plplot embedded in pyqt

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

To Allesandro:

Please let me know if/when you are subscribed to the plplot_devel list (see
http://sourceforge.net/mail/?group_id=2915) so I don't end up putting double
e-mails into your mail box.

I have just put the relevant pyqt GUI front end files into cvs.  There are a
number of changes from your original files so could you please check that
the cvs version works for you?  I believe you were on the mailing list when
I previously today gave instructions about how to access the latest cvs
version, but I will send that separately if you didn't get that e-mail.

Here is the quick cookbook for the pyqt GUI frontend for plplot (which I
urge other plplot developers to try for themselves).

(1) install libqt (devel version of binary package if you are not building
it yourself).

(2) install sip (Allesandro has given details before).

(3) install pyqt (Allesandro has given details before).

(4) clean checkout of plplot from cvs.  Make sure there is no plplot dir
in the current directory when you do this since the cvs checkout
will create a new plplot directory in the current directory.

(5)  cd plplot; make configure; \
./configure --prefix=/usr/local/plplot --with-double ; \
make

(6) cd tmp; make pyqt_plmodule.so

This makes the additional module that you need for the pyqt GUI. This module
will be automatically made under configuration control in near future, i.e.,
the "make" command in (5) will soon make this module if pyqt is available.

(7) Set your LD_LIBRARY_PATH environment variable to the directory where
libqt resides on your system.  On my system I compiled my own in a user
account so I specify

setenv LD_LIBRARY_PATH \
/home/software/kde2_download/debian_athlon_20011013/qt-copy/lib/

but you will want to do something different.  I believe I could have avoided
this if I had specified LD_RUN_PATH before building pyqt (or if that build
had used -rpath).  I recall I had to specify my qt location in any case for
that build to work so specifying LD_RUN_PATH (see man ld) in addition would
not be that difficult.

(8) ./prova.py #Note this (and all other files you need) are in the CVS
version so do not copy any other files into plplot tree for this test. This
brings up the pyqt GUI front end for plplot.  Click anywhere in the white
box and simply type

import xw08

(or any other of the xw?? examples except for xw09) and then click on
file-->run to get many nice looking plot examples.

Allesandro, once you have confirmed the cvs version works for you, could you
please put some page control in the GUI so each page of xw08 (or any other
of the multi-page examples) can be browsed?  IMHO, this should be a high
priority although I am sure there are plenty of other things you want to do
with this GUI now that it is working and in the CVS version of plplot.

Things I still need to do:

automatically make pyqt_plmodule.so (or not) under configuration control
depending on availability of pyqt.

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
__________________________