Hi Everyone,
I'am trying to build (compile) a simulation using some private thorns and the Einstein Toolkit (ET_2014_05). The build goes through successfully on my local machine (a laptop running on Gnu/Linux mint 17.1, using GCC compilers).
However, when compiling the same code on a remote cluster (running on Gnu/Linux RedHat plus Intel compilers) the compiling persistently fails with the error: cactus/arrangements/EinsteinAnalysis/AHFinderDirect/src/patch/patch_system.cc(87):
catastrophic error: cannot open source file "hdf5.h" # include <hdf5.h>
Searching on Google, there seems to have been aticket (https://trac.einsteintoolkit.org/ticket/911, or http://lists.einsteintoolkit.org/pipermail/users/2012-July/002333.html) on this issue and it was resolved. I tried the work-around suggested by Roland on the mailing list about modifying the configuration.ccl file in the concerned thorn. But this didn't work.
The ticket connected with this error points to modifications in cactus/lib/sbin/ConfigurationParser.pl file. On the web-page of this ticket/patches, I can see some highlighted sections -- green and red -- and some that are not. But I'm not sure what the difference between them is, especially in terms of what to remove and/or add onto my copy of the file in order to remove the error. (I have attached a copy of my ConfigurationParser.pl file)
Naively, I also suspect the issue could be that the HDF5 library on this cluster was compiled without support for C++. (..because the source file that raises the error is a C++ file which conditionally requests some HDF5 functionality.) But I tested this particular point by pointing the simulation to my own copy of HDF5 which I compiled with C++ support within the cluster. Yet the error still comes up and stops the compilation.
I'd appreciate any ideas to help me past this error.
Regards, Dumsani
On Wed, Aug 19, 2015 at 05:54:51PM +0200, dumsani wrote:
I'am trying to build (compile) a simulation using some private thorns and the Einstein Toolkit (ET_2014_05). The build goes through successfully on my local machine (a laptop running on Gnu/Linux mint 17.1, using GCC compilers).
However, when compiling the same code on a remote cluster (running on Gnu/Linux RedHat plus Intel compilers) the compiling persistently fails with the error: cactus/arrangements/EinsteinAnalysis/AHFinderDirect/src/patch/patch_system.cc(87):
catastrophic error: cannot open source file "hdf5.h" # include <hdf5.h>
This failure, as you probably guessed, means that the compiler cannot find the hdf5 library - not just the C++ part, the library itself.
Searching on Google, there seems to have been aticket (https://trac.einsteintoolkit.org/ticket/911, or http://lists.einsteintoolkit.org/pipermail/users/2012-July/002333.html) on this issue and it was resolved.
Since you are using 2014_05 this should be already resolved in your checkout.
Compile again, using 'make sim VERBOSE=yes'. This should give you the complete compiler command line used for cactus/arrangements/EinsteinAnalysis/AHFinderDirect/src/patch/patch_system.cc. See if this line includes the hdf5 library (it does not have to necessarily). Besides that line, let us know how the HDF5 thorn configured HDF5: do you use a system-provided version, a version you compiled yourself, or a version Cactus built for you?
We would need at least that information to see what is going on. Also - did you try using a newer ET release?
Frank
Hi Frank, Everyone
The problem was solved. Thanks to your suggestion to use the VERBOSE=yes option. This option revealed that the HDF5 library path was not correctly specified in my .cfg file. There was an extra /lib bit on the path, instead of giving the path only up to the top-level library directory. That apparently confused the compiler, and hence the include directory could not be found.
Thank you for your assistance.
Kind regards, Dumsani
On 19/08/2015 18:21, Frank Loeffler wrote:
On Wed, Aug 19, 2015 at 05:54:51PM +0200, dumsani wrote:
I'am trying to build (compile) a simulation using some private thorns and the Einstein Toolkit (ET_2014_05). The build goes through successfully on my local machine (a laptop running on Gnu/Linux mint 17.1, using GCC compilers).
However, when compiling the same code on a remote cluster (running on Gnu/Linux RedHat plus Intel compilers) the compiling persistently fails with the error: cactus/arrangements/EinsteinAnalysis/AHFinderDirect/src/patch/patch_system.cc(87):
catastrophic error: cannot open source file "hdf5.h" # include <hdf5.h>
This failure, as you probably guessed, means that the compiler cannot find the hdf5 library - not just the C++ part, the library itself.
Searching on Google, there seems to have been aticket (https://trac.einsteintoolkit.org/ticket/911, or http://lists.einsteintoolkit.org/pipermail/users/2012-July/002333.html) on this issue and it was resolved.
Since you are using 2014_05 this should be already resolved in your checkout.
Compile again, using 'make sim VERBOSE=yes'. This should give you the complete compiler command line used for cactus/arrangements/EinsteinAnalysis/AHFinderDirect/src/patch/patch_system.cc. See if this line includes the hdf5 library (it does not have to necessarily). Besides that line, let us know how the HDF5 thorn configured HDF5: do you use a system-provided version, a version you compiled yourself, or a version Cactus built for you?
We would need at least that information to see what is going on. Also - did you try using a newer ET release?
Frank
On 20 Aug 2015, at 10:56, dumsani g14n8326@campus.ru.ac.za wrote:
Hi Frank, Everyone
The problem was solved. Thanks to your suggestion to use the VERBOSE=yes option. This option revealed that the HDF5 library path was not correctly specified in my .cfg file. There was an extra /lib bit on the path, instead of giving the path only up to the top-level library directory. That apparently confused the compiler, and hence the include directory could not be found.
Thank you for your assistance.
Perhaps the HDF5 thorn (and all other external library thorns) should check that the paths given actually exist, and abort, or at least give a warning, if they don't?
Kind regards, Dumsani
On 19/08/2015 18:21, Frank Loeffler wrote:
On Wed, Aug 19, 2015 at 05:54:51PM +0200, dumsani wrote:
I'am trying to build (compile) a simulation using some private thorns and the Einstein Toolkit (ET_2014_05). The build goes through successfully on my local machine (a laptop running on Gnu/Linux mint 17.1, using GCC compilers).
However, when compiling the same code on a remote cluster (running on Gnu/Linux RedHat plus Intel compilers) the compiling persistently fails with the error: cactus/arrangements/EinsteinAnalysis/AHFinderDirect/src/patch/patch_system.cc(87):
catastrophic error: cannot open source file "hdf5.h" # include <hdf5.h>
This failure, as you probably guessed, means that the compiler cannot find the hdf5 library - not just the C++ part, the library itself.
Searching on Google, there seems to have been aticket (https://trac.einsteintoolkit.org/ticket/911, or http://lists.einsteintoolkit.org/pipermail/users/2012-July/002333.html) on this issue and it was resolved.
Since you are using 2014_05 this should be already resolved in your checkout.
Compile again, using 'make sim VERBOSE=yes'. This should give you the complete compiler command line used for cactus/arrangements/EinsteinAnalysis/AHFinderDirect/src/patch/patch_system.cc. See if this line includes the hdf5 library (it does not have to necessarily). Besides that line, let us know how the HDF5 thorn configured HDF5: do you use a system-provided version, a version you compiled yourself, or a version Cactus built for you?
We would need at least that information to see what is going on. Also - did you try using a newer ET release?
Frank
Users mailing list Users@einsteintoolkit.org http://lists.einsteintoolkit.org/mailman/listinfo/users
users@lists.einsteintoolkit.org
-
dumsani -
Frank Loeffler -
Ian Hinder