[ADMB Users] windows version of admb
dave fournier
I have made a more windows friendly packjage for the mingw32 gcc432
compiler that you can test if you want.It uses the inno installer.
give it a try and see what you think.
http://otter-rsch.com/admb-org-rc/setupADMB_gcc432_mingw32.exe
--
David A. Fournier
P.O. Box 2040,
Sidney, B.C. V8l 3S3
Canada
Phone/FAX 250-655-3364
http://otter-rsch.com
_______________________________________________
Users mailing list
Us...@admb-project.org
http://lists.admb-project.org/mailman/listinfo/users
Schnute, Jon
finding a Windows binary version of gcc 4.3.2
(http://gcc.gnu.org/gcc-4.3/). The seemingly official distribution site
for Windows versions of gcc
http://sourceforge.net/project/showfiles.php?group_id=2435
suggests that gcc 4.x is still at a testing phase and that the current
release is gcc 3.4.5. The older release 3.4.2 from 2004 is still
available.
Do you perhaps mean gcc 3.4.2, rather than gcc 4.3.2 as the name of your
exe file suggests? If you really are using 4.3.2, where did you get the
binary distribution for Windows?
Thanks for your help,
Jon
******************************************
Jon Schnute, Scientist Emeritus
Pacific Biological Station
3190 Hammond Bay Road
Nanaimo, BC V9T 6N7
CANADA
NEW email: Jon.S...@dfo-mpo.gc.ca
(formerly schn...@pac.dfo-mpo.gc.ca)
******************************************
Gabor Grothendieck
a 4.2.1 binary of g++ together with an various other assorted tools.
See:
http://www.murdoch-sutherland.com/Rtools/
C:\tmp2>g++ --version
g++ (GCC) 4.2.1-sjlj (mingw32-2)
Copyright (C) 2007 Free Software Foundation, Inc.
This is free software; see the source for copying conditions. There is NO
warranty; not even for MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.
Jim Ianelli
with the link that Dave provided below. Thanks for that.
My initial impression is that the compile-time is considerably slower than
the Borland compiler (old version 5.2) . Runtime seems is improved
considerable (all within windows vista 32bit). The Borland compiled code
for a basic assessment model took 39 seconds while the gcc compiled version
took only 25 seconds(no screen IO, I used the redirect ">NUL"). The same
model compiled in "safe" admb mode using the gcc compiler took 33 seconds.
Regarding the distribution of CPU load over the 2 processors was the same in
both cases (appearing to load one by about 1/3 and the other by about 2/3
capacity).
So it seems that using admb with the Borland compiler for development has
advantages (fast compile times, workable debugger) but for execution, the
gcc has clear advantages.
I would like to open the discussion of debuggers. I notice that the rtools
distribution doesn't seem to include the gdb tools or maybe I'm missing
what's commonly used. Have folks that use MSVS C++ had luck configuring the
debugger to work well?
Cheers,
Jim
James Ianelli
REFM Division, Alaska Fisheries Science Center, NOAA
7600 Sand Point Way NE
Seattle WA 98115
Ph. 206 526 6510
Schnute, Jon
http://otter-rsch.com/admb-org-rc/setupADMB_gcc432_mingw32.exe
works for me too. Great! This is the first open source version I've seen
that works on a Windows platform. I really like the idea of using Rtools
as the appropriate version of mingw because most serious ADMB users
probably also use R. If they make R packages, they'll need Rtools
anyway.
I've noticed three minor problems with this version that could easily be
fixed:
1. The main directory (\ADMB) of this distribution has a file
"version.txt" that says "ADMB for MinGW gcc 3.4.2". That is incorrect
(as discussed in earlier email), and it should say something like "ADMB
for gcc (GCC) 4.2.1-sjlj (mingw32-2), as distributed in Rtools28.exe,
available from http://www.murdoch-sutherland.com/Rtools/." I strongly
encourage the developers always to include a file that tells users how
to get the relevant compiler, at least for Windows versions.
2. The name of the setup file should be changed to reflect this version
of gcc (421, not 432). I'm not sure how important the sjlj (short
jump/long jump) version would be for ADMB, although I understand that it
matters for R.
3. The "uninstall" feature does not work for this installation. It
reports that "The uninstall log file ...\unins000.dat is in a format not
recognized by this version of the uninstaller. Cannot uninstall."
Probably, different versions of the Inno installer have become involved
somehow.
I also have five questions:
Q1. Given Jim's observation that the Borland compiler works well for
development and the GNU compiler for production (on Windows), could we
get a similar version of ADMB for BCC55 available at
http://cc.codegear.com/Free.aspx?id=24778? If I had both versions, I'd
seriously consider writing an R interface (with PBSmodelling) that would
allow a user to switch easily between the two versions. We already have
an R package that gives such an interface to an old version of ADMB that
works with gcc 3.4.2 on Windows. (It still needs conversion to the new
version of ADMB.)
Q2. Will this installation file (hopefully corrected along the lines
mentioned above) now appear on the ADMB Project site
http://admb-project.org/?
Q3. What is the connection between the sites http://admb-project.org/
and http://code.google.com/p/admb-project/? Which should I consider the
primary source, or are they equivalent?
Q4. The project web site says that "Ultimately, the full source code
will be available, enabling researchers to contribute enhancements and
add-ons." When do you plan to make the source code available? It strikes
me that the user community might surprise everyone with their ability to
add value to the project, even at this early stage. I know some computer
scientists who could probably make the package work on a variety of
platforms if they had the source. Given the numerous flavours of Linux,
versions of gcc, and other operating systems (such as Windows and OS X),
it would help to get the user community more directly involved. In the R
project, various users have undertaken to maintain versions of R for
specific operating systems. (I understand, of course, that you need a
core development team.)
Q5. When users have contributions (code and/or documentation) that they
consider worthwhile, how do they make them broadly available? For
example, my group may soon have an R package that would effectively give
an easy Windows installation of ADMB, along with a handy user interface.
Best wishes to everyone involved,
Jon
PS to Gabor - Thanks for reminding me of g++ --version.