plplot-devel — Technical discussion about PLplot development

Re: [Plplot-devel] The infamous "missing" strikes again

[Joao C. <jc...@fe...>]

On Tuesday 04 March 2003 01:13, Joao Cardoso wrote:
> On Monday 03 March 2003 23:52, Alan W. Irwin wrote:
> > On Mon, 3 Mar 2003, Rafael Laboissiere wrote:
> > > * Joao Cardoso <jc...@fe...> [2003-03-03 20:53]:
> > > > After a fresh cvs checkout, running bootstrap.sh
=2E..
> I don't want to make a distro, just be able to quickly test on another
> architecture (I'm concerned with NaN stuff in nn/csa/plgridd.c).

As I was afraid, under OSF1-V4.0 libcsa has trouble because of NaNs. libn=
n=20
must have the same problem.

I ended up trying libcsa with its own test programs, but they fail, and I=
=20
trace the problem down to NaNs handling

It is increadible how ieee-754 is still "vendor dependent"! Cray is know =
to be=20
not conformant, but OSF...

Joao

Re: [Plplot-devel] The infamous "missing" strikes again

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

On Tue, 4 Mar 2003, Joao Cardoso wrote:

> Alan receipt, "touch plConfig.h.in  plDevs.h.in", is enough for my needs.

In general, yes, but in specific you have to be careful of what you do and
the order you do it.

Clean tree, then bootstrap.sh, then touch, then create tarball.

Of course, when you use the tarball, you will have to manually turn off tcl
and python (java is already disabled by default) if they are on the system
at all.

I still think it is easier to do

clean tree then script then create tarball

with no worries about turning off anything afterward when you use the
tarball, but that is up to you.

Alan

__________________________
Alan W. Irwin
email: ir...@be...
phone: 250-727-2902

Astronomical research affiliation with Department of Physics and Astronomy,
University of Victoria (astrowww.phys.uvic.ca).

Programming affiliations with the Canadian Centre for Climate Modelling and
Analysis (www.cccma.bc.ec.gc.ca) and the PLplot scientific plotting software
package (plplot.org).

__________________________

Linux-powered Science
__________________________

Re: [Plplot-devel] The infamous "missing" strikes again

[Joao C. <jc...@fe...>]

On Monday 03 March 2003 23:52, Alan W. Irwin wrote:
> On Mon, 3 Mar 2003, Rafael Laboissiere wrote:
> > * Joao Cardoso <jc...@fe...> [2003-03-03 20:53]:
> > > After a fresh cvs checkout, running bootstrap.sh
> > >
> > > [jcard@feup] ./bootstrap.sh
> > > Running aclocal (GNU automake) 1.7.2... done
> > > Running autoheader (GNU Autoconf) 2.57... done
> > > Running automake (GNU automake) 1.7.2... done
> > > Running libtoolize (GNU libtool) 1.4.3... done
> > > Running autoconf (GNU Autoconf) 2.57... done
> > >
> > > taring, ftp to an  alphaev6-dec-osf4.0f system, untaring and runnin=
g
> > > configure gave several errors:
> > >
> > > [...]
> >
> > This technique is error-prone.  The correct approach is to generate a
> > tarball with "make dist".  Unfortunately, make dist does not work
> > perfectly yet (but will for the next release, hopefully).
>
> I agree you do have to be careful with making a tarball, but on the oth=
er
> hand it is almost as straightforward as avoiding pitfalls when developi=
ng
> with cvs.  You must start with a clean tree, and you need the correct
> version of swig and autotools on your system where you are preparing th=
e
> tarball.  The only peculiar thing is two files need to be touched if yo=
u
> are not using my script.
>
> Note that scripts/make_tarball.sh touches two files so that autotools a=
re
> not invoked on the system where the tarball is being used.

I don't want to make a distro, just be able to quickly test on another=20
architecture (I'm concerned with NaN stuff in nn/csa/plgridd.c).

Alan receipt, "touch plConfig.h.in  plDevs.h.in", is enough for my needs.

Thanks,
Joao

Re: [Plplot-devel] The infamous "missing" strikes again

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

On Mon, 3 Mar 2003, Rafael Laboissiere wrote:

> * Joao Cardoso <jc...@fe...> [2003-03-03 20:53]:
>
> > After a fresh cvs checkout, running bootstrap.sh
> >
> > [jcard@feup] ./bootstrap.sh
> > Running aclocal (GNU automake) 1.7.2... done
> > Running autoheader (GNU Autoconf) 2.57... done
> > Running automake (GNU automake) 1.7.2... done
> > Running libtoolize (GNU libtool) 1.4.3... done
> > Running autoconf (GNU Autoconf) 2.57... done
> >
> > taring, ftp to an  alphaev6-dec-osf4.0f system, untaring and running configure
> > gave several errors:
> >
> > [...]
>
> This technique is error-prone.  The correct approach is to generate a
> tarball with "make dist".  Unfortunately, make dist does not work perfectly
> yet (but will for the next release, hopefully).

I agree you do have to be careful with making a tarball, but on the other
hand it is almost as straightforward as avoiding pitfalls when developing
with cvs.  You must start with a clean tree, and you need the correct
version of swig and autotools on your system where you are preparing the
tarball.  The only peculiar thing is two files need to be touched if you are
not using my script.

Note that scripts/make_tarball.sh touches two files so that autotools are
not invoked on the system where the tarball is being used. That script also
does the appropriate swig and perl preparation so you won't need swig or
perl on the system where the tarball is being used.

Until Rafael can get make dist to work, using this script is probably your
best bet for preparing a clean tree for a tarball.  Just now, I have
commented out getting access to our documentation results from the website
because those files are changing.  I will put them back in correct form
later once our documentation settles down (and if make dist continues not to
work), but for now, nobody needs those documentation results.

Let me know if you have any trouble generating a tarball with the script and
testing it on your Linux system.  If the tarball works on Linux (i.e., no
autotools or swig execution and build, install, and plplot-test.sh are fine)
but does not work on other platforms, that is a pretty clear indication that
cross-platform issues have crept into CVS again.

Alan
__________________________
Alan W. Irwin
email: ir...@be...
phone: 250-727-2902

Astronomical research affiliation with Department of Physics and Astronomy,
University of Victoria (astrowww.phys.uvic.ca).

Programming affiliations with the Canadian Centre for Climate Modelling and
Analysis (www.cccma.bc.ec.gc.ca) and the PLplot scientific plotting software
package (plplot.org).

__________________________

Linux-powered Science
__________________________

[Plplot-devel] New documentation including plgriddata

[Rafael L. <lab...@ps...>]

I uploaded the new PDF documentation file including the recent addition by
Joao.  The entry for plgriddata in the API chapter appears in page 103:

    http://people.debian.org/~rafael/plplot-5.2.0.cvs.20030303.pdf

-- 
Rafael

Re: [Plplot-devel] The infamous "missing" strikes again

[Rafael L. <lab...@ps...>]

* Joao Cardoso <jc...@fe...> [2003-03-03 20:53]:

> After a fresh cvs checkout, running bootstrap.sh
> 
> [jcard@feup] ./bootstrap.sh 
> Running aclocal (GNU automake) 1.7.2... done
> Running autoheader (GNU Autoconf) 2.57... done
> Running automake (GNU automake) 1.7.2... done
> Running libtoolize (GNU libtool) 1.4.3... done
> Running autoconf (GNU Autoconf) 2.57... done
> 
> taring, ftp to an  alphaev6-dec-osf4.0f system, untaring and running configure 
> gave several errors:
>
> [...]

This technique is error-prone.  The correct approach is to generate a
tarball with "make dist".  Unfortunately, make dist does not work perfectly
yet (but will for the next release, hopefully).

-- 
Rafael

[Plplot-devel] The infamous "missing" strikes again

[Joao C. <jc...@fe...>]

Hi,

After a fresh cvs checkout, running bootstrap.sh

[jcard@feup] ./bootstrap.sh=20
Running aclocal (GNU automake) 1.7.2... done
Running autoheader (GNU Autoconf) 2.57... done
Running automake (GNU automake) 1.7.2... done
Running libtoolize (GNU libtool) 1.4.3... done
Running autoconf (GNU Autoconf) 2.57... done

taring, ftp to an  alphaev6-dec-osf4.0f system, untaring and running conf=
igure=20
gave several errors:

checking for gcc option to accept ANSI C... none needed
checking build system type... ./config.guess: umask: `7077' is not an oct=
al=20
numb
er from 000 to 777
config.guess: cannot create /tmp/config-guess-23584
alphaev6-dec-osf4.0f
checking host system type... alphaev6-dec-osf4.0f

=2E..

checking for mkoctfile... yes
sed: Function s/--compile// ; t END ; d ; : END q cannot be parsed.
checking for octave... yes
sed: Function s/^.*version // ; q cannot be parsed.
warning: broken pipe
sed: Function s/^2.1//p ; d ; q cannot be parsed.
checking for Python.h... no
warning: can't find Python.h, setting enable_python to no
checking for qh_new_qhull in -lqhull... no
checking for KCC... no

=2E..


config.status: executing depfiles commands
configure: configuring in libltdl
configure: running /bin/bash './configure'=20
--prefix=3D/usr/users1/deec/jcard/plplo
t-alf/test  '--prefix=3D/usr/users1/deec/jcard/plplot-alf/test'=20
'--disable-static'
 --enable-ltdl-convenience --cache-file=3D/dev/null --srcdir=3D.
loading cache /dev/null
=2E/configure: .: /dev/null: not a regular file

=2E..

The summary gave

Optional libraries:
with_freetype:  no              with_qhull:     yes

although qhull was not detected!

Then, after removing all, untaring again and reconfiguring with=20
--prefix=3D<xxx> --disable-static --without-qhull

besides the above errors gave:

bash-2.01$ make
cd . && /bin/bash /usr/users1/deec/jcard/plplot-alf/missing --run autohea=
der
/usr/users1/deec/jcard/plplot-alf/missing: autoheader: command not found
WARNING: `autoheader' is missing on your system.  You should only need it=
 if
         you modified `acconfig.h' or `configure.ac'.  You might want
         to install the `Autoconf' and `GNU m4' packages.  Grab them
         from any GNU archive site.
touch ./config.h.in
cd . && /bin/bash ./config.status config.h
make: *** [stamp-h1] Error 139
bash-2.01$ ls -l configure*
-rwx------   1 jcard    docentes  626581 Mar  3 20:20 configure
-rw-------   1 jcard    docentes   27320 Mar  3 16:27 configure.ac

configuring a second time and making again worked, but latter

make[2]: Entering directory `/usr/users1/deec/jcard/plplot-alf/include'
cd .. && /bin/bash /usr/users1/deec/jcard/plplot-alf/missing --run autohe=
ader
/usr/users1/deec/jcard/plplot-alf/missing: autoheader: command not found

=2E..

Joao

Re: [Plplot-devel] Remove deprecated functions from API?

[Joao C. <jc...@fe...>]

On Monday 03 March 2003 17:02, Alan W. Irwin wrote:
> Here is a list of functions in our API that have been officially deprec=
ated
> (in comments and documentation) since 2000 (and probably for even longe=
r).
> I searched for these using "deprecated" or "Do <emphasis>not</emphasis>=
 use
> this" in api.xml.
> Alternatives that can be used instead are put on the right.
>
> "deprecated"
> plclr =3D=3D> pleop
> plcol =3D=3D> plcol0
> plpage =3D=3D> plbop
>
> "Do <emphasis>not</emphasis> use this"
> plhls =3D=3D> plHLS_RGB followed by plscol0
> plrgb =3D=3D> plscol0
> plrgb1 =3D=3D> plscol0
>
> The first 3 are implemented with #defines.  I guess there is no harm in
> continuing to keep them, but in the interests of keeping things as simp=
le
> as possible for our users, I would prefer to remove them from the API a=
nd
> documentation.  There has certainly been enough notice given!

I agree on removing them from the docs, but not from plplot.h.
As you say, there is no harm, and not being documented new users won't us=
e=20
them.
But I think that plotsh3d() and MinMax2dGrid() could be removed, because =
they=20
had such an ephemeral life. I don't know for how long the other defines h=
ave=20
been there.
To safely remove them all, after one or two more releases, I would do the=
=20
following:

#define    plclr=09plclr_replace_me_with_pleop

and users will get a linker error:
   undefined reference to `plclr_replace_me_with_pleop'

But I'm sure that most of you won't agree on this.

> The second group of three are actually coded rather than #defined,

In that case I would replace the code with something like plexit("plrgb: =
as I=20
could be doing Bad Things to your program, I was (ex)terminated") and wai=
t=20
one or two more releases to remove them all.

But again not everybody likes this kind of jokes when they are working.

Joao

> and are
> a more serious matter because they completely break are cmap0 paradigm.=
=20
> They use the cmap0 colour index of PL_RGB_COLOR=3D128 to store the colo=
ur.=20
> This is likely to cause heap corruption since by default we have only 1=
6
> cmap0 colours.  Even when we have allocated enough colours for this to
> work, these functions pound away on the 128th colour, regardless, which
> will certainly confuse users wanting to use that colour index for somet=
hing
> else.  I do not know who put in the "Do NOT use this" comment in the co=
de,
> but I completely agree with them.  Note, plhls calls plHLS_RGB, then pl=
rgb
> so has the same problems as plrgb (and plrgb1).
>
> I think it is definitely long past time to remove the second three (and
> possibly the first three) from the API.
>
> Comments, please?
>
> Alan
> __________________________
> Alan W. Irwin
> email: ir...@be...
> phone: 250-727-2902
>
> Astronomical research affiliation with Department of Physics and Astron=
omy,
> University of Victoria (astrowww.phys.uvic.ca).
>
> Programming affiliations with the Canadian Centre for Climate Modelling=
 and
> Analysis (www.cccma.bc.ec.gc.ca) and the PLplot scientific plotting
> software package (plplot.org).
>
> __________________________
>
> Linux-powered Science
> __________________________
>
>
>
> -------------------------------------------------------
> This sf.net email is sponsored by:ThinkGeek
> Welcome to geek heaven.
> http://thinkgeek.com/sf
> _______________________________________________
> Plplot-devel mailing list
> Plp...@li...
> https://lists.sourceforge.net/lists/listinfo/plplot-devel

[Plplot-devel] Remove deprecated functions from API?

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

Here is a list of functions in our API that have been officially deprecated
(in comments and documentation) since 2000 (and probably for even longer).
I searched for these using "deprecated" or "Do <emphasis>not</emphasis> use
this" in api.xml.
Alternatives that can be used instead are put on the right.

"deprecated"
plclr ==> pleop
plcol ==> plcol0
plpage ==> plbop

"Do <emphasis>not</emphasis> use this"
plhls ==> plHLS_RGB followed by plscol0
plrgb ==> plscol0
plrgb1 ==> plscol0

The first 3 are implemented with #defines.  I guess there is no harm in
continuing to keep them, but in the interests of keeping things as simple
as possible for our users, I would prefer to remove them from the API and
documentation.  There has certainly been enough notice given!

The second group of three are actually coded rather than #defined, and are a
more serious matter because they completely break are cmap0 paradigm.  They
use the cmap0 colour index of PL_RGB_COLOR=128 to store the colour.  This is
likely to cause heap corruption since by default we have only 16 cmap0
colours.  Even when we have allocated enough colours for this to work, these
functions pound away on the 128th colour, regardless, which will certainly
confuse users wanting to use that colour index for something else.  I do not
know who put in the "Do NOT use this" comment in the code, but I completely
agree with them.  Note, plhls calls plHLS_RGB, then plrgb so has the
same problems as plrgb (and plrgb1).

I think it is definitely long past time to remove the second three (and
possibly the first three) from the API.

Comments, please?

Alan
__________________________
Alan W. Irwin
email: ir...@be...
phone: 250-727-2902

Astronomical research affiliation with Department of Physics and Astronomy,
University of Victoria (astrowww.phys.uvic.ca).

Programming affiliations with the Canadian Centre for Climate Modelling and
Analysis (www.cccma.bc.ec.gc.ca) and the PLplot scientific plotting software
package (plplot.org).

__________________________

Linux-powered Science
__________________________

Re: [Plplot-devel] plgriddata builds & works

[Joao C. <jc...@fe...>]

On Monday 03 March 2003 15:59, Alan W. Irwin wrote:
> > Take a look at:
> >
> >     http://people.debian.org/~rafael/griddata.ps
>
> Wow!
>
> > Thanks very much for this valuable addition, Jo=E3o!
>
> Seconded!
>
> Alan

As I said when I committed x21c.c, don't be fooled by the default plots.
Try specially the --randn option, to simulate non-uniform sampling, and a=
lso=20
the -rosen function, that is much less smooth than the sombrero function.
Also, verify that some algorithms are not very sensitive to peaks, while=20
others are sensitive in excess.

Also, there are border effects on most algorithms, as they return NaNs wh=
ere=20
interpolation/approximation is not possible, and I have to circumvent the=
=20
fact that Plplot is not NaN/Inf aware(my next to come but always delayed=20
project). To solve the problem  I do local averaging, when possible.

The last cvs commits I have done will enable all you that (still) don't h=
ave=20
Qhull installed in your systems to try plgriddata(). Please try it before=
=20
installing qhull.

I'm still working on x12c.c, so that the plots show a fair comparison of =
the=20
several algorithms and do not give (harmless) errors at the same time.

Joao

> __________________________
> Alan W. Irwin
> email: ir...@be...
> phone: 250-727-2902
>
> Astronomical research affiliation with Department of Physics and Astron=
omy,
> University of Victoria (astrowww.phys.uvic.ca).
>
> Programming affiliations with the Canadian Centre for Climate Modelling=
 and
> Analysis (www.cccma.bc.ec.gc.ca) and the PLplot scientific plotting
> software package (plplot.org).
>
> __________________________
>
> Linux-powered Science
> __________________________
>
>
>
> -------------------------------------------------------
> This sf.net email is sponsored by:ThinkGeek
> Welcome to geek heaven.
> http://thinkgeek.com/sf
> _______________________________________________
> Plplot-devel mailing list
> Plp...@li...
> https://lists.sourceforge.net/lists/listinfo/plplot-devel

Re: [Plplot-devel] plgriddata builds & works

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

> Take a look at:
>
>     http://people.debian.org/~rafael/griddata.ps

Wow!

>
> Thanks very much for this valuable addition, Jo=E3o!

Seconded!

Alan
__________________________
Alan W. Irwin
email: ir...@be...
phone: 250-727-2902

Astronomical research affiliation with Department of Physics and Astronomy,
University of Victoria (astrowww.phys.uvic.ca).

Programming affiliations with the Canadian Centre for Climate Modelling and
Analysis (www.cccma.bc.ec.gc.ca) and the PLplot scientific plotting softwar=
e
package (plplot.org).

__________________________

Linux-powered Science
__________________________

[Plplot-devel] plgriddata builds & works

[Rafael L. <lab...@ps...>]

Joao's plgridd.c, as well ass the libnn and libcsa stuff, have been added
recently to the CVS tree.  All the configuration rough edges have been
polished and everything is building and works.  The results of the new API
function plgriddata looks great.  Take a look at:

    http://people.debian.org/~rafael/griddata.ps

Thanks very much for this valuable addition, João!

-- 
Rafael

[Plplot-devel] Problems in using Plplot-5.2.0 tcl/tk drivers on Linux

[Pankaj L. <pa...@da...>]

Hi All,

I have built Plplot-5.2.0 on RH7.1. I want to use tcl/tk driver. I have built
plplot twice. My observations are as following

1 Without defining CFLAG=-DUSE_TCL_STUBS _DUSE_TK_STUB
In this case I am able to build the library without any compilation error. But
plplot crashes when I do following on the wish prompt
%package require Plplotter
5.2.0
%plframe .p1
%plframe .p2
When I try to create the second plframe, I am getting errors from CreatePixmap
function from "\driver\tkwin.c" file. The errors thrown by some XCreatePixmap
function stating bad window parameter, bad window option.

2. With CFLAG=-DUSE_TCL_STUBS _DUSE_TK_STUB
2a In this case I get compilation error in file "\bindings\tcl\tclmain.c" file
for line 50, 51. I could build the package once I comment these lines. But the
my wish prompt does not return when I do the following
% package require Plplotter
5.2.0
% plframe .p1
% .p1 cmd plinit
After this command, I get list of all the available drivers like xwin tcl/tl
jpeg etc When I select tcl/tk as my driver wish prompt does not return.

Incidentally, I am trying to build Plplot-5.2.0 on Solaris also. In this case
also, I am facing the same problem.

I want to create multiple Plplot windows and use these to plot some real time
data plotting. I am able to do this on Windows. But I am struggling to make
Plplot working on Solaris and Linux.

Can you somebody  help me to sort out these problems on Linux and Solaris?

Thanks in advance for any help.
-Pankaj.

Re: [Plplot-devel] Configuration defaults

[Joao C. <jc...@fe...>]

On Monday 03 March 2003 00:49, Rafael Laboissiere wrote:
> * Rafael Laboissiere <lab...@ps...> [2003-03-03 01:05]:
> > For me two, provided that you the data passed back and forth are cast=
ed
> > with PLFLT.  Single precision people should know that they are going =
to
> > loose something (in this case memory space) if they want to use the g=
reat
> > plgriddata.  However, they should no be prevented to use it because t=
he
> > API only accepts double.

I have already made the changes.
I made the cast explicit, so that casual readers will understand that the=
=20
types might be different.

I think that most compilers will just ignore a cast from a variable that =
is=20
already in the final type, like in:

=09x =3D (double) y;

The cast should represent no overhead if both x and y are doubles. Do you=
=20
agree?

> Sorry, I have looked carefully into plgridd.c.  It seems that plgriddat=
a
> already uses PLFLT in a consistent way, so my comment above is not
> relevant. However, why don't use PLINT instead of int?

I never worried about it, except in structure definitions, for padding=20
purposes, and in drivers, where it can matter.

Joao

Re: [Plplot-devel] Configuration defaults

[Rafael L. <lab...@ps...>]

* Rafael Laboissiere <lab...@ps...> [2003-03-03 01:05]:

> For me two, provided that you the data passed back and forth are casted
> with PLFLT.  Single precision people should know that they are going to
> loose something (in this case memory space) if they want to use the great
> plgriddata.  However, they should no be prevented to use it because the API
> only accepts double.

Sorry, I have looked carefully into plgridd.c.  It seems that plgriddata
already uses PLFLT in a consistent way, so my comment above is not relevant.
However, why don't use PLINT instead of int?
 
-- 
Rafael

Re: [Plplot-devel] Configuration defaults

[Rafael L. <lab...@ps...>]

* Joao Cardoso <jc...@fe...> [2003-03-02 23:35]:

> That was my idea, but forget it, we have more important things to fix.

Okay, let us keep this project dormant.  I still think that regrouping the
options in .c/onfigure --help and organising them by sections may be a good
idea.

-- 
Rafael

Re: [Plplot-devel] Configuration defaults

[Rafael L. <lab...@ps...>]

* Joao Cardoso <jc...@fe...> [2003-03-02 23:27]:

> I understand that situation. How could it be solved?
> 
> 1-Adding another configure option --short-in-memory or perhaps 
> --with-huge-data? In that case, if --with-double=no --with-huge-data=yes were  
> specified, then nn/csa would not be built; if only --with-double=no, than 
> floats would be converted to doubles and vice versa.
> 
> 2-Even if --with-double=no, we always convert floats<->doubles, but rely that 
> users know what they are doing and would not call the nn/csa based gridding 
> algorithms? If they would they would do it just once -- learn by error.
> 
> 3-Or making --with-qhull=no by default? I vote against this last one :()
> 
> The second option seems reasonable to me.

For me two, provided that you the data passed back and forth are casted with
PLFLT.  Single precision people should know that they are going to loose
something (in this case memory space) if they want to use the great
plgriddata.  However, they should no be prevented to use it because the API
only accepts double.

> It could be run by configure, then users would just sent the result to us.
> Stupid woman!(1) don't you see that we already have this? It's called 
> config.summary!

I know, I was thinking about something more complete, including the
architecture and other things.  And yes, config.summary is a good starting
point.

-- 
Rafael

Re: [Plplot-devel] Configuration defaults

[Joao C. <jc...@fe...>]

On Sunday 02 March 2003 21:58, Rafael Laboissiere wrote:
> * Joao Cardoso <jc...@fe...> [2003-03-02 20:01]:
=2E..
>
> > Is it possible to categorize this with sub-help options? I guess not.
>
> I do not know exactly what you mean.  If you are suggesting that config=
ure
> should accept new sub-help options like --help-foo and --help-bar, I th=
ink
> we are out of hope.

That was my idea, but forget it, we have more important things to fix.
Joao

Re: [Plplot-devel] Configuration defaults

[Joao C. <jc...@fe...>]

On Sunday 02 March 2003 22:10, Rafael Laboissiere wrote:
> * Joao Cardoso <jc...@fe...> [2003-03-02 20:01]:
> > To be accurate, it is straightforward, as data structures have to be
> > filled forth and back, but I don't think that's acceptable if the use=
r
> > specified with-float at configure time.
>
> I understand your argument, and I would also defend it.  However, Alan
> pointed me out in a private discussion that there may be cases where us=
ers
> do not care about efficiency or memory occupation, but would like to
> configure --without-double, because they need to use some legacy code.

I understand that situation. How could it be solved?

1-Adding another configure option --short-in-memory or perhaps=20
--with-huge-data? In that case, if --with-double=3Dno --with-huge-data=3D=
yes were =20
specified, then nn/csa would not be built; if only --with-double=3Dno, th=
an=20
floats would be converted to doubles and vice versa.

2-Even if --with-double=3Dno, we always convert floats<->doubles, but rel=
y that=20
users know what they are doing and would not call the nn/csa based griddi=
ng=20
algorithms? If they would they would do it just once -- learn by error.

3-Or making --with-qhull=3Dno by default? I vote against this last one :(=
)

The second option seems reasonable to me.

> > > Another possibility would be to make a pool at our website.
> >
> > That approach relies on a positive action, deliberately filling a for=
m,
> > and as such it would undersample our users universe; on the contrary,
> > asking at the end of configure "send ... yes/no?" only requires a
> > keypress.
>
> To be frank, I would be pissed off if some project I configure & make a=
sked
> me such questions.  People are not used to that and I think that PLplot
> should not make an exception.  Besides that, the configure script must =
be
> non-interactive, at any case.
>
> What about including a command in the distribution (like a shell script=
,
> with text AC_SUBSTituted by configure and which email the informations =
to
> us) and we ask our users in the mailing lists to run it?

It could be run by configure, then users would just sent the result to us=
=2E
Stupid woman!(1) don't you see that we already have this? It's called=20
config.summary!

Joao

(1) from an old British comics series, "Hallo, Hallo", if I recollect wel=
l :)

Re: [Plplot-devel] Configuration defaults

[Rafael L. <lab...@ps...>]

* Joao Cardoso <jc...@fe...> [2003-03-02 20:01]:

> To be accurate, it is straightforward, as data structures have to be filled 
> forth and back, but I don't think that's acceptable if the user specified 
> with-float at configure time.

I understand your argument, and I would also defend it.  However, Alan
pointed me out in a private discussion that there may be cases where users
do not care about efficiency or memory occupation, but would like to
configure --without-double, because they need to use some legacy code.

> > Another possibility would be to make a pool at our website.
> 
> That approach relies on a positive action, deliberately filling a form, and
> as such it would undersample our users universe; on the contrary, asking at
> the end of configure "send ... yes/no?" only requires a keypress.

To be frank, I would be pissed off if some project I configure & make asked
me such questions.  People are not used to that and I think that PLplot
should not make an exception.  Besides that, the configure script must be
non-interactive, at any case.

What about including a command in the distribution (like a shell script,
with text AC_SUBSTituted by configure and which email the informations to
us) and we ask our users in the mailing lists to run it?  Okay, it relies
also on positive action, but is less hassle than filling forms in a web
site.

-- 
Rafael

Re: [Plplot-devel] Configuration defaults

[Rafael L. <lab...@ps...>]

* Joao Cardoso <jc...@fe...> [2003-03-02 20:01]:

> 1) --with-double does not appears when we do 
>  
>     ./configure --help
> 
> but it is honoured if we configure --with-double=no.

Fixed in my last cvs commit.  It is impressive that we have never included a
MY_ARG_WITH for the --with-double option.

> 3) ./configure --help | wc -l
> 
> reports 180 lines of help!!!, 7 screen pages!
> The options count are:
>    29  are --with, 
>    49 are --enable,
>    6  are  --disable
>    1  is --without

I do not see a problem with this.  I usually do "./configure --help | less".
What we have absolutely to do is to update and improve the documentation in
doc/docbook/src/os_notes.xml.

> Is it possible to categorize this with sub-help options? I guess not.

I do not know exactly what you mean.  If you are suggesting that configure
should accept new sub-help options like --help-foo and --help-bar, I think
we are out of hope.  If you are talking about a better presentation when
./configure --help is run, we could do something.

For instance, if I modify the MY_ARG_WITH(double,...) that I just committed
to read:

MY_ARG_WITH(double,[
Options that will affect the behavior of the library:
]AC_HELP_STRING([--with-double],
    [Compile the library with double precision floating point
     ($with_double)]))
 
then I have:

./configure --help
[...]
  --with-mkoctfile=file   Specify mkoctfile
  --with-octavex=file      Specify Octave

Options that will affect the behavior of the library:
  --with-double           Compile the library with double precision floating
                          point ($with_double)
  --with-www-user=NAME    User name at WWW host
[...]

Of course, this will only be meaninful if we do a reorganisation of
configure.ac and sysloc.in to regroup the calls to MY_ARG_WITH and
MY_ARG_ENABLE in a logical order.

-- 
Rafael

Re: [Plplot-devel] plgriddata() commited

[Rafael L. <lab...@ps...>]

* Rafael Laboissiere <lab...@ps...> [2003-03-02 17:22]:

> * Joao Cardoso <jc...@fe...> [2003-03-02 02:45]:
> > (1) After a fresh checkout, I got this:
> > 
> > jcard@home:~/tmp/plplot> ./bootstrap.sh 
> > Running aclocal (GNU automake) 1.7.2... done
> > Running autoheader (GNU Autoconf) 2.57... done
> > Running automake (GNU automake) 1.7.2...configure.ac: installing 
> > `./install-sh'
> > configure.ac: installing `./mkinstalldirs'
> > configure.ac: installing `./missing'
> > configure.ac:433: installing `./config.guess'
> > configure.ac:433: installing `./config.sub'
> > configure.ac:433: required file `./ltmain.sh' not found
> > Makefile.am:30: required directory ./libltdl does not exist
> > bindings/c++/Makefile.am: installing `./depcomp'
> > drivers/Makefile.am: installing `./compile'
> > Makefile.am:30: required directory ./libltdl does not exist
> >  done
> > Running libtoolize (GNU libtool) 1.4.3... done
> > Running autoconf (GNU Autoconf) 2.57... done
> > 
> > The second time I run it there where no problems.
> 
> I will try to fix this.

Actually, there is no fix, since there is no problem here.  These are only
harnless warning messages.  After the first run of bootstrap.sh, the source
tree should be ready for configure and make.

-- 
Rafael

Re: [Plplot-devel] plgriddata() commited

[Joao C. <jc...@fe...>]

On Sunday 02 March 2003 16:22, Rafael Laboissiere wrote:
> * Joao Cardoso <jc...@fe...> [2003-03-02 02:45]:
> > This all works with *hand tuned* makefiles. I tried to automate
> > everything but I wasn't able, sorry. Rafael and Alan, I count on you =
to
> > fix the configuration.
>
> I will take a look.

Thanks. Meanwhile I will keep trying.

>
> > I also added more capabilities to plenv(), allowing for square plots =
and
> > minor tick marks.
>
> Have you broken backward compatibility?

No.

Joao

Re: [Plplot-devel] Configuration defaults

[Joao C. <jc...@fe...>]

On Saturday 01 March 2003 09:22, Rafael Laboissiere wrote:
> * Joao Cardoso <jc...@fe...> [2003-03-01 03:30]:
> > [snip]
> >
> > If the user wants to use those techniques he will ponder and rebuild
> > plplot. So, making float the default is not that bad. With a proper
> > Announcement, users will be warned from the very beginning and will
> > configure accordingly.  (But with a proper Announcement we could also=
 say
> > that doubles are now the default.)
> >
> > Given all the above, I still think that doubles should be the default=
=2E
>
> I am pretty convinced by your arguments.  I thought that it would be
> straightforward (and acceptable) to cast back and forth all the data
> processed with nn/csa.

To be accurate, it is straightforward, as data structures have to be fill=
ed=20
forth and back, but I don't think that's acceptable if the user specified=
=20
with-float at configure time.
Another example where memory is importante is in plimage(), that works=20
internally with shorts and chars!

> > Pity we are not MS. We could make "configure" to send us the configur=
e
> > summary so we would collect statistics ;-)
>
> Well, you can always implement this and this would work for non-MS syst=
ems
> too.  Of course, we should not play the Insidious PLplot Big Brother, b=
ut
> letting users to choose if they wish to send the informations or not.=20

That was what I meant.
And on what service would you rely? sendmail? With most desktop users rel=
ying=20
in they mail-reader ability to send and receive mail directly from their=20
organization mailer, I think that sendmail is not correctly setup on most=
=20
systems.

> There is a package in Debian, called popularity-contest, that sends
> regularly reports about what is installed in the system to a specific e=
mail
> address. We are using this to collect statistics.
>
> Another possibility would be to make a pool at our website.

That approach relies on a positive action, deliberately filling a form, a=
nd as=20
such it would undersample our users universe; on the contrary, asking at =
the=20
end of configure "send ... yes/no?" only requires a keypress.

Re: [Plplot-devel] Configuration defaults

[Joao C. <jc...@fe...>]

Hi,

1) --with-double does not appears when we do=20
=20
    ./configure --help

but it is honoured if we configure --with-double=3Dno.

 Users that require floats might think that the option is no longer suppo=
rted.

2) Also, --disable-drivers seems to not sort  effect, as=20

    gd ljii ljiip ntk pbm ps pstex tk tkwin xfig xwin

are built.

3) ./configure --help | wc -l

reports 180 lines of help!!!, 7 screen pages!
The options count are:
   29  are --with,=20
   49 are --enable,
   6  are  --disable
   1  is --without

Is it possible to categorize this with sub-help options? I guess not.

Joao