9:20 p.m.
Hello,
I just got a notification of update for Gretl on my Fedora 26.
[image: Inline image 1]
Please excuse my Portuguese ;)
A big thanks to you, master developers, and rpm maintainers.
12:46 a.m.
On Mon, 20 Nov 2017, Hélio Guilherme wrote:
Hello,
I just got a notification of update for Gretl on my Fedora 26.
[image: Inline image 1]
Please excuse my Portuguese ;)
A big thanks to you, master developers, and rpm maintainers.
Thanks, Hélio!
For the rpm, it's Johannes Lips to be thanked -- and for deb
packages it's Dirk Eddelbuettel. Both of these guys have been
following our development for a long time, and have been very
helpful in (patiently) pointing out, and suggesting fixes for, gretl
build bugs.
Allin
6:30 a.m.
On Mon, 20 Nov 2017, Allin Cottrell wrote:
On Mon, 20 Nov 2017, Hélio Guilherme wrote:
> Hello,
>
> I just got a notification of update for Gretl on my Fedora 26.
>
> [image: Inline image 1]
>
> Please excuse my Portuguese ;)
>
> A big thanks to you, master developers, and rpm maintainers.
Thanks, Hélio!
For the rpm, it's Johannes Lips to be thanked -- and for deb packages it's
Dirk Eddelbuettel. Both of these guys have been following our development for
a long time, and have been very helpful in (patiently) pointing out, and
suggesting fixes for, gretl build bugs.
Speaking of which: I don't think mpi is enabled in the stock debian build.
I noticed that several programs on debian have an "-openmpi" deb package
variant. Perhaps we could have a gretl-openmpi package too? On my "buster"
box (current testing) once you have libopenmpi-dev and libopenmpi2
installed, all you need to do is give the --with-mpi option to the
configure script, and everything just works. I guess it'd just be a matter
of adjusting the build dependencies, no? Dirk?
-------------------------------------------------------
Riccardo (Jack) Lucchetti
Dipartimento di Scienze Economiche e Sociali (DiSES)
Università Politecnica delle Marche
(formerly known as Università di Ancona)
r.lucchetti(a)univpm.it
http://www2.econ.univpm.it/servizi/hpp/lucchetti
-------------------------------------------------------
8:03 a.m.
On 20.11.2017 22:20, Hélio Guilherme wrote:
Hello,
I just got a notification of update for Gretl on my Fedora 26.
Inline image 1
Please excuse my Portuguese ;)
A big thanks to you, master developers, and rpm maintainers.
Hey Hélio,
you are welcome! It's always great, when one's contributions are
appreciated and noticed. :-)
Cheers
Johannes
_______________________________________________
Gretl-devel mailing list
Gretl-devel(a)lists.wfu.edu
http://lists.wfu.edu/mailman/listinfo/gretl-devel
12:23 p.m.
On 21 November 2017 at 07:30, Riccardo (Jack) Lucchetti wrote:
| Speaking of which: I don't think mpi is enabled in the stock debian build.
| I noticed that several programs on debian have an "-openmpi" deb package
| variant. Perhaps we could have a gretl-openmpi package too? On my "buster"
| box (current testing) once you have libopenmpi-dev and libopenmpi2
| installed, all you need to do is give the --with-mpi option to the
| configure script, and everything just works. I guess it'd just be a matter
| of adjusting the build dependencies, no? Dirk?
Yes. But ... what we probably want is two builds, one serial (as before) and
one with mpi as you suggest.
That requires a rejigging / enhancing / rewriting of the the (Debian) build
file to ensure that any files needed by both end up in a -common package,
that the serial build ends up in a package of its own, that the parallel
build ends up in a package of its own and that all corner cases of upgrading
from (current) serial to (new serial), or new parallel etc pp are
covered. Plus all switches between and all possible transition. All doable,
just a lot of patient nitty gritty work.
And I just don't have the time. I am truly sorry.
I _could_ change the standard build to always use openmpi. May open yet
another can of worms, but at least it is less involved.
Dirk
--
http://dirk.eddelbuettel.com | @eddelbuettel | edd(a)debian.org
1:50 p.m.
Am 21.11.2017 um 13:23 schrieb Dirk Eddelbuettel:
I _could_ change the standard build to always use openmpi. May open
yet
another can of worms, but at least it is less involved.
Yes, why not? The only real downside would be additional dependencies,
right? Are there any cases (architectures etc.) where those dependencies
would fail which would then block installation of gretl? Or could
openmpi become a package suggestion instead of a dependency?
thanks,
sven
2:20 p.m.
On Tue, 21 Nov 2017, Sven Schreiber wrote:
Am 21.11.2017 um 13:23 schrieb Dirk Eddelbuettel:
> I _could_ change the standard build to always use openmpi. May open yet
> another can of worms, but at least it is less involved.
Yes, why not? The only real downside would be additional dependencies, right?
Are there any cases (architectures etc.) where those dependencies would fail
which would then block installation of gretl? Or could openmpi become a
package suggestion instead of a dependency?
The openmpi library is in the nature of an "optional dependency" for
gretl.
When gretl is built with MPI support an extra CLI executable is
compiled (gretlmpi) alongside the regular GUI and CLI binaries.
This extra binary won't run unless the user has openmpi installed, but
the regular ones work as usual with or without openmpi on the system.
So I don't think one loses anything by including MPI aupport in the
build by default (as we do for MS Windows). A user without any MPI
library just has an extra binary that won't work unless/until they
install MPI.
Allin
4:26 p.m.
On Tue, 21 Nov 2017, Dirk Eddelbuettel wrote:
I _could_ change the standard build to always use openmpi. May open
yet
another can of worms, but at least it is less involved.
Would gretl_x11 work on a system without openmpi installed, or would it
complain about the missing library? Not a big deal, mind you, I guess it
would just amount to make gretl dependent on libopempi2, but I was
wondering if that could be avoided.
-------------------------------------------------------
Riccardo (Jack) Lucchetti
Dipartimento di Scienze Economiche e Sociali (DiSES)
Università Politecnica delle Marche
(formerly known as Università di Ancona)
r.lucchetti(a)univpm.it
http://www2.econ.univpm.it/servizi/hpp/lucchetti
-------------------------------------------------------
5:06 p.m.
On Tue, 21 Nov 2017, Riccardo (Jack) Lucchetti wrote:
On Tue, 21 Nov 2017, Dirk Eddelbuettel wrote:
> I _could_ change the standard build to always use openmpi. May open yet
> another can of worms, but at least it is less involved.
Would gretl_x11 work on a system without openmpi installed, or would it
complain about the missing library?
It will work OK, as will gretlcli. It's just gretlmpi that won't work
without the MPI library.
We've taken some pains to arrange this: neither gretl_x11 nor gretlcli
are linked against the MPI library; we get the required symbols at
runtime via dlopen, if they're available.
Allin
5:53 p.m.
On 21 November 2017 at 12:06, Allin Cottrell wrote:
| On Tue, 21 Nov 2017, Riccardo (Jack) Lucchetti wrote:
|
| > On Tue, 21 Nov 2017, Dirk Eddelbuettel wrote:
| >
| >> I _could_ change the standard build to always use openmpi. May open yet
| >> another can of worms, but at least it is less involved.
| >
| > Would gretl_x11 work on a system without openmpi installed, or would it
| > complain about the missing library?
|
| It will work OK, as will gretlcli. It's just gretlmpi that won't work
| without the MPI library.
|
| We've taken some pains to arrange this: neither gretl_x11 nor gretlcli
| are linked against the MPI library; we get the required symbols at
| runtime via dlopen, if they're available.
That sounds good.
So ... one never needs one of the MPI frontend, ie orte_run? Or, conversely,
one gets this only when gretl is launched this way? Ie user opts in, no side
effects for others?
Can't stand in the way of such Pareto-improvements ;-)
Dirk
--
http://dirk.eddelbuettel.com | @eddelbuettel | edd(a)debian.org
9:44 p.m.
On Tue, 21 Nov 2017, Dirk Eddelbuettel wrote:
On 21 November 2017 at 12:06, Allin Cottrell wrote:
| On Tue, 21 Nov 2017, Riccardo (Jack) Lucchetti wrote:
|
| > On Tue, 21 Nov 2017, Dirk Eddelbuettel wrote:
| >
| >> I _could_ change the standard build to always use openmpi. May open yet
| >> another can of worms, but at least it is less involved.
| >
| > Would gretl_x11 work on a system without openmpi installed, or would it
| > complain about the missing library?
|
| It will work OK, as will gretlcli. It's just gretlmpi that won't work
| without the MPI library.
|
| We've taken some pains to arrange this: neither gretl_x11 nor gretlcli
| are linked against the MPI library; we get the required symbols at
| runtime via dlopen, if they're available.
That sounds good.
So ... one never needs one of the MPI frontend, ie orte_run? Or, conversely,
one gets this only when gretl is launched this way? Ie user opts in, no side
effects for others?
Yes, it's an opt-in thing. We need a back-end that supports "mpirun"
or "mpiexec" -- e.g. orte_run -- but only if/when gretlmpi gets
called, not for the regular programs.
Can't stand in the way of such Pareto-improvements ;-)
Heh, that's right.
Allin
11:21 p.m.
No luck so far. I added the (virtual) packages mpi-default-dev and
mpi-default-bin (mostly OpenMPI, but some of Debian arches may use MPICH).
They are installed but your configure then report
checking for mpicc... yes
checking for MPI library... Couldn't find MPI library
Hence:
Configuration:
Installation path: /usr
Use readline library: yes
Use gnuplot for graphs: yes
Use pdflatex for typesetting: yes
Use libgsf for zip/unzip: no
sse2 support for RNG: no
OpenMP support: yes
MPI support: no
AVX support for arithmetic: no
Build with GTK version: 2.0
Build gretl documentation: no
Use Lucida fonts: no
Build message catalogs: yes
Build gretl addons: no
X-12-ARIMA support: yes
TRAMO/SEATS support: yes
libR support: yes
libsvm support: no
ODBC support: yes
JSON parsing support: yes
Experimental audio support: no
Use xdg-utils in installation: no
LAPACK libraries:
-llapack -lblas -lgfortran
Now type 'make' to build gretl.
Could you configure possibly rely on this:
edd@brad:~$ pkg-config --cflags mpi
-pthread -I/usr/lib/x86_64-linux-gnu/openmpi/include
-I/usr/lib/x86_64-linux-gnu/openmpi/include/openmpi
edd@brad:~$ pkg-config --libs mpi
-L/usr/lib/x86_64-linux-gnu/openmpi/lib -L/usr//lib -lmpi
edd@brad:~$
Happy to test a patch.
Dirk
--
http://dirk.eddelbuettel.com | @eddelbuettel | edd(a)debian.org
11:40 p.m.
On Sat, 25 Nov 2017, Dirk Eddelbuettel wrote:
No luck so far. I added the (virtual) packages mpi-default-dev and
mpi-default-bin (mostly OpenMPI, but some of Debian arches may use MPICH).
They are installed but your configure then report
checking for mpicc... yes
checking for MPI library... Couldn't find MPI library
Hence:
Configuration:
Installation path: /usr
Use readline library: yes
Use gnuplot for graphs: yes
Use pdflatex for typesetting: yes
Use libgsf for zip/unzip: no
sse2 support for RNG: no
OpenMP support: yes
MPI support: no
[...]
Could you[r] configure possibly rely on this:
edd@brad:~$ pkg-config --cflags mpi
-pthread -I/usr/lib/x86_64-linux-gnu/openmpi/include
-I/usr/lib/x86_64-linux-gnu/openmpi/include/openmpi
edd@brad:~$ pkg-config --libs mpi
-L/usr/lib/x86_64-linux-gnu/openmpi/lib -L/usr//lib -lmpi
edd@brad:~$
Hmm, maybe, but I think Debian is unusual in supplying an mpi.pc
file (not present on Fedora or Arch anyway).
Could you try using the config variable(s) --with-mpi-lib= (and
perhaps also --with-mpi-include=)?
Thanks,
Allin
11:50 p.m.
On 26 November 2017 at 18:40, Allin Cottrell wrote:
| On Sat, 25 Nov 2017, Dirk Eddelbuettel wrote:
|
| > No luck so far. I added the (virtual) packages mpi-default-dev and
| > mpi-default-bin (mostly OpenMPI, but some of Debian arches may use MPICH).
| > They are installed but your configure then report
| >
| > checking for mpicc... yes
| > checking for MPI library... Couldn't find MPI library
| >
| > Hence:
| >
| > Configuration:
| >
| > Installation path: /usr
| > Use readline library: yes
| > Use gnuplot for graphs: yes
| > Use pdflatex for typesetting: yes
| > Use libgsf for zip/unzip: no
| > sse2 support for RNG: no
| > OpenMP support: yes
| > MPI support: no
| [...]
| >
| > Could you[r] configure possibly rely on this:
| >
| > edd@brad:~$ pkg-config --cflags mpi
| > -pthread -I/usr/lib/x86_64-linux-gnu/openmpi/include
-I/usr/lib/x86_64-linux-gnu/openmpi/include/openmpi
| > edd@brad:~$ pkg-config --libs mpi
| > -L/usr/lib/x86_64-linux-gnu/openmpi/lib -L/usr//lib -lmpi
| > edd@brad:~$
|
| Hmm, maybe, but I think Debian is unusual in supplying an mpi.pc
| file (not present on Fedora or Arch anyway).
|
| Could you try using the config variable(s) --with-mpi-lib= (and
| perhaps also --with-mpi-include=)?
Not easily. See how those values have x86_64 in them? We have over 10
different architectures (even if x86_64 may cover 98% of users) and my build
recipe is supposed to cover all equally.
Dirk
--
http://dirk.eddelbuettel.com | @eddelbuettel | edd(a)debian.org
7:15 a.m.
On Sun, 26 Nov 2017, Allin Cottrell wrote:
> No luck so far. I added the (virtual) packages mpi-default-dev
and
> mpi-default-bin (mostly OpenMPI, but some of Debian arches may use MPICH).
> They are installed but your configure then report
I have no problems here with libopenmpi-dev.
-------------------------------------------------------
Riccardo (Jack) Lucchetti
Dipartimento di Scienze Economiche e Sociali (DiSES)
Università Politecnica delle Marche
(formerly known as Università di Ancona)
r.lucchetti(a)univpm.it
http://www2.econ.univpm.it/servizi/hpp/lucchetti
-------------------------------------------------------
4:25 p.m.
On Mon, 27 Nov 2017, Riccardo (Jack) Lucchetti wrote:
On Sun, 26 Nov 2017, Allin Cottrell wrote:
>> No luck so far. I added the (virtual) packages mpi-default-dev and
>> mpi-default-bin (mostly OpenMPI, but some of Debian arches may use MPICH).
>> They are installed but your configure then report
I have no problems here with libopenmpi-dev.
Good to hear, but I think in Dirk's case he's (cross-) building gretl
for several architectures on a given server, which adds a layer of
complication.
I'm working on a revision of our mpi.m4 which hopefully should sort
this out.
Allin
10:55 p.m.
On Sun, 26 Nov 2017, Dirk Eddelbuettel wrote:
On 26 November 2017 at 18:40, Allin Cottrell wrote:
| On Sat, 25 Nov 2017, Dirk Eddelbuettel wrote:
|
| > No luck so far. I added the (virtual) packages mpi-default-dev and
| > mpi-default-bin (mostly OpenMPI, but some of Debian arches may use MPICH).
| > They are installed but your configure then report
| >
| > checking for mpicc... yes
| > checking for MPI library... Couldn't find MPI library
| >
| > Hence:
| >
| > Configuration:
| >
| > Installation path: /usr
| > Use readline library: yes
| > Use gnuplot for graphs: yes
| > Use pdflatex for typesetting: yes
| > Use libgsf for zip/unzip: no
| > sse2 support for RNG: no
| > OpenMP support: yes
| > MPI support: no
| [...]
| >
| > Could you[r] configure possibly rely on this:
| >
| > edd@brad:~$ pkg-config --cflags mpi
| > -pthread -I/usr/lib/x86_64-linux-gnu/openmpi/include
-I/usr/lib/x86_64-linux-gnu/openmpi/include/openmpi
| > edd@brad:~$ pkg-config --libs mpi
| > -L/usr/lib/x86_64-linux-gnu/openmpi/lib -L/usr//lib -lmpi
| > edd@brad:~$
|
| Hmm, maybe, but I think Debian is unusual in supplying an mpi.pc
| file (not present on Fedora or Arch anyway).
|
| Could you try using the config variable(s) --with-mpi-lib= (and
| perhaps also --with-mpi-include=)?
Not easily. See how those values have x86_64 in them? We have
over 10 different architectures (even if x86_64 may cover 98% of
users) and my build recipe is supposed to cover all equally.
OK, understood. Please try the attached patch for the configure
script in gretl-2017d.
I'm keeping Jack in cc in case he has time to experiment, in an
attempt to avoid the case where we support a multi-arch build
environment at the cost of breaking a single-arch one.
Allin
11:37 p.m.
On 27 November 2017 at 17:55, Allin Cottrell wrote:
| On Sun, 26 Nov 2017, Dirk Eddelbuettel wrote:
|
| > On 26 November 2017 at 18:40, Allin Cottrell wrote:
| > | On Sat, 25 Nov 2017, Dirk Eddelbuettel wrote:
| > |
| > | > No luck so far. I added the (virtual) packages mpi-default-dev and
| > | > mpi-default-bin (mostly OpenMPI, but some of Debian arches may use MPICH).
| > | > They are installed but your configure then report
| > | >
| > | > checking for mpicc... yes
| > | > checking for MPI library... Couldn't find MPI library
| > | >
| > | > Hence:
| > | >
| > | > Configuration:
| > | >
| > | > Installation path: /usr
| > | > Use readline library: yes
| > | > Use gnuplot for graphs: yes
| > | > Use pdflatex for typesetting: yes
| > | > Use libgsf for zip/unzip: no
| > | > sse2 support for RNG: no
| > | > OpenMP support: yes
| > | > MPI support: no
| > | [...]
| > | >
| > | > Could you[r] configure possibly rely on this:
| > | >
| > | > edd@brad:~$ pkg-config --cflags mpi
| > | > -pthread -I/usr/lib/x86_64-linux-gnu/openmpi/include
-I/usr/lib/x86_64-linux-gnu/openmpi/include/openmpi
| > | > edd@brad:~$ pkg-config --libs mpi
| > | > -L/usr/lib/x86_64-linux-gnu/openmpi/lib -L/usr//lib -lmpi
| > | > edd@brad:~$
| > |
| > | Hmm, maybe, but I think Debian is unusual in supplying an mpi.pc
| > | file (not present on Fedora or Arch anyway).
| > |
| > | Could you try using the config variable(s) --with-mpi-lib= (and
| > | perhaps also --with-mpi-include=)?
| >
| > Not easily. See how those values have x86_64 in them? We have
| > over 10 different architectures (even if x86_64 may cover 98% of
| > users) and my build recipe is supposed to cover all equally.
|
| OK, understood. Please try the attached patch for the configure
| script in gretl-2017d.
Nice work, and I really appreciate it. Seems to work too:
OpenMP support: yes
MPI support: yes
Build in progress so I better not jinx it ...
| I'm keeping Jack in cc in case he has time to experiment, in an
| attempt to avoid the case where we support a multi-arch build
| environment at the cost of breaking a single-arch one.
Sure. Earlier you wrote
On 27 November 2017 at 11:25, Allin Cottrell wrote:
| On Mon, 27 Nov 2017, Riccardo (Jack) Lucchetti wrote:
|
| > On Sun, 26 Nov 2017, Allin Cottrell wrote:
| >
| >>> No luck so far. I added the (virtual) packages mpi-default-dev and
| >>> mpi-default-bin (mostly OpenMPI, but some of Debian arches may use MPICH).
| >>> They are installed but your configure then report
| >
| > I have no problems here with libopenmpi-dev.
|
| Good to hear, but I think in Dirk's case he's (cross-) building gretl
| for several architectures on a given server, which adds a layer of
| complication.
Not really. I build on my platform, and then ship my sources. Autobuilders
pick those up and build on theirs. So I am lower-tech -- no magic
multiplatform builders here.
Dirk
--
http://dirk.eddelbuettel.com | @eddelbuettel | edd(a)debian.org
2340
days inactive
2347
days old
17 comments
6 participants
participants (6)
-
Allin Cottrell -
Dirk Eddelbuettel -
Hélio Guilherme -
Johannes Lips -
Riccardo (Jack) Lucchetti -
Sven Schreiber