Installation challenges

takluyver commented 4 years ago

What's the recommended way to install SimEx? I've been looking at various things without much success:

The conda environment file seems to be missing several dependencies even to start building, and I can't get far even by adding things. Eventually, with some pointers from @ejcjason I was able to get it to work by disabling almost every component in install.sh and commenting out various imports in the Python code, but this doesn't feel like a good answer.
The install docs refer to a .deb package, but it's from a few years ago, and my target isn't Debian-based anyway.
They also refer to a docker image yakser/simex, which was last updated in 2017.
I tried building a docker image from Docker/simex within the repository, but it fell down pretty quickly. I haven't investigated this much, but I see the docker folder hasn't been touched for 2 years, so I doubt you're using that frequently.

CFGrote commented 4 years ago

Thanks, this is very important information.

The recommended way to install simex is using cmake. install.sh serves as a guideline. More hints may be found in .travis.yml and in scripts under Travis/ .

Installation has successfully been tested on various linux distributions including debian7+ ubuntu 14.04+, centos7.

What is your target? c compiler? hdf version? cmake version? mpi ? mkl?

where does it fail if you enable more components? which ones?

takluyver commented 4 years ago

Thanks Carsten!

In the first instance I want to install it on Maxwell (I know there's already a module, but I want to test a branch @ejcjason is working on). So CentOS 7. I'll focus on the conda installation for now. Once I've got a direct installation working, I might take an interest in using Spack to install it, and/or containerising it.

So, I've recreated a conda environment using conda-requirements.yml. I've modified install.sh to disable everything except SingFEL & WPG. Then I get CMake 3.12 or higher is required. OK, conda install cmake to get a newer version.

This time it worked. Previously I tried for longer with other modules enabled, and something I installed for one of the other modules seemed to break WPG. I'll experiment further and get back to you.

takluyver commented 4 years ago

This is a bit long, but I figure it's worth recording the details of what errors I'm getting and what I'm trying. I should also have mentioned that I turned DEVELOPER_MODE off on @ejcjason's suggestion.

Disable WPG (for now), enable CrystFEL.

/home/kluyvert/.conda/envs/simex/lib/libpng16.so.16: undefined reference to `inflateValidate@ZLIB_1.2.9'

I didn't see this one before. conda install zlib ? No, that doesn't fix it. conda install libpng ? Replaces a conda-forge package with one from the default channel. No, still the same error.

I recall there was a change in the way conda works at some point, and now the recommendation is to install a compiler from conda, so you're not using something outside the environment. Maybe I'm running into some symptom of that. conda install gcc_linux-64 .

src/utils.h:41:10: fatal error: gsl/gsl_matrix.h: No such file or directory

Now we're using a compiler in conda, it no longer uses libraries from the system, if I've understood correctly. This is the first of a series of problems building CrystFEL which I ran into before, so let's install several packages I already found it needs: conda install gsl cairo pango gdk-pixbuf .

Building CrystFEL now works. Re-enable WPG.

x86_64-conda_cos6-linux-gnu-cc: error trying to exec 'cc1plus': execvp: No such file or directory

Missing a C++ compiler? Trying conda install gxx_linux-64 .

/home/kluyvert/.conda/envs/simex/bin/../lib/gcc/x86_64-conda_cos6-linux-gnu/7.3.0/../../../../x86_64-conda_cos6-linux-gnu/bin/ld: cannot find -lgomp

This is where I got stuck last time around. conda install libgomp . A package gets installed, but I still get the same error. This appears to come while building srwlpy for XFELPhotonPropagator.

Summary: Relying on compilers & libraries on the system, I can build WPG, but building CrystFEL fails. By installing compilers in the conda environment, I can get CrystFEL to build, but then WPG fails, and I've not yet worked out how to solve that.

CFGrote commented 4 years ago

since you try this on maxwell, there is a functional installation on maxwell under /data/netapp/s2e/simex

$> module load intel/2018
$> module load mpi/openmpi-x86_64-intel

annoyingly, you have to clean up your PYTHONPATH after this (remove the py2.7 part of it).

then:

$> source /data/netapp/s2e/simex/bin/simex_vars.sh

test:

$> python

>>> from SimEx import *

Hope this helps.

-- Dr. Carsten Fortmann-Grote PaNOSC WP5 leader

On 5/6/20 10:49 AM, Thomas Kluyver wrote:

This is a bit long, but I figure it's worth recording the details of what errors I'm getting and what I'm trying. I should also have mentioned that I turned |DEVELOPER_MODE| off on @ejcjason https://github.com/ejcjason's suggestion.

Disable WPG (for now), enable CrystFEL.

|/home/kluyvert/.conda/envs/simex/lib/libpng16.so.16: undefined reference to `inflateValidate@ZLIB_1.2.9' |

I didn't see this one before. |conda install zlib| ? No, that doesn't fix it. |conda install libpng| ? Replaces a |conda-forge| package with one from the default channel. No, still the same error.

I recall there was a change in the way conda works https://www.anaconda.com/blog/utilizing-the-new-compilers-in-anaconda-distribution-5 at some point, and now the recommendation is to install a compiler from conda, so you're not using something outside the environment. Maybe I'm running into some symptom of that. |conda install gcc_linux-64| .

|src/utils.h:41:10: fatal error: gsl/gsl_matrix.h: No such file or directory |

Now we're using a compiler in conda, it no longer uses libraries from the system, if I've understood correctly. This is the first of a series of problems building CrystFEL which I ran into before, so let's install several packages I already found it needs: |conda install gsl cairo pango gdk-pixbuf| .

Building CrystFEL now works. Re-enable WPG.

|x86_64-conda_cos6-linux-gnu-cc: error trying to exec 'cc1plus': execvp: No such file or directory |

Missing a C++ compiler? Trying |conda install gxx_linux-64| .

|/home/kluyvert/.conda/envs/simex/bin/../lib/gcc/x86_64-conda_cos6-linux-gnu/7.3.0/../../../../x86_64-conda_cos6-linux-gnu/bin/ld: cannot find -lgomp |

This is where I got stuck last time around. |conda install libgomp| . A package gets installed, but I still get the same error. This appears to come while building |srwlpy| for |XFELPhotonPropagator|.

Summary: Relying on compilers & libraries on the system, I can build WPG, but building CrystFEL fails. By installing compilers in the conda environment, I can get CrystFEL to build, but then WPG fails, and I've not yet worked out how to solve that.

— You are receiving this because you commented. Reply to this email directly, view it on GitHub https://github.com/PaNOSC-ViNYL/SimEx/issues/196#issuecomment-624521486, or unsubscribe https://github.com/notifications/unsubscribe-auth/ADWCQRZLGGL6D3UAEFYUYWDRQEQAXANCNFSM4MZWXLWQ.

CFGrote commented 4 years ago

hi thomas, you need intel omp: module load intel

-- Dr. Carsten Fortmann-Grote PaNOSC WP5 leader

On 5/6/20 10:49 AM, Thomas Kluyver wrote:

This is a bit long, but I figure it's worth recording the details of what errors I'm getting and what I'm trying. I should also have mentioned that I turned |DEVELOPER_MODE| off on @ejcjason https://github.com/ejcjason's suggestion.

Disable WPG (for now), enable CrystFEL.

|/home/kluyvert/.conda/envs/simex/lib/libpng16.so.16: undefined reference to `inflateValidate@ZLIB_1.2.9' |

I didn't see this one before. |conda install zlib| ? No, that doesn't fix it. |conda install libpng| ? Replaces a |conda-forge| package with one from the default channel. No, still the same error.

I recall there was a change in the way conda works https://www.anaconda.com/blog/utilizing-the-new-compilers-in-anaconda-distribution-5 at some point, and now the recommendation is to install a compiler from conda, so you're not using something outside the environment. Maybe I'm running into some symptom of that. |conda install gcc_linux-64| .

|src/utils.h:41:10: fatal error: gsl/gsl_matrix.h: No such file or directory |

Now we're using a compiler in conda, it no longer uses libraries from the system, if I've understood correctly. This is the first of a series of problems building CrystFEL which I ran into before, so let's install several packages I already found it needs: |conda install gsl cairo pango gdk-pixbuf| .

Building CrystFEL now works. Re-enable WPG.

|x86_64-conda_cos6-linux-gnu-cc: error trying to exec 'cc1plus': execvp: No such file or directory |

Missing a C++ compiler? Trying |conda install gxx_linux-64| .

|/home/kluyvert/.conda/envs/simex/bin/../lib/gcc/x86_64-conda_cos6-linux-gnu/7.3.0/../../../../x86_64-conda_cos6-linux-gnu/bin/ld: cannot find -lgomp |

This is where I got stuck last time around. |conda install libgomp| . A package gets installed, but I still get the same error. This appears to come while building |srwlpy| for |XFELPhotonPropagator|.

Summary: Relying on compilers & libraries on the system, I can build WPG, but building CrystFEL fails. By installing compilers in the conda environment, I can get CrystFEL to build, but then WPG fails, and I've not yet worked out how to solve that.

— You are receiving this because you commented. Reply to this email directly, view it on GitHub https://github.com/PaNOSC-ViNYL/SimEx/issues/196#issuecomment-624521486, or unsubscribe https://github.com/notifications/unsubscribe-auth/ADWCQRZLGGL6D3UAEFYUYWDRQEQAXANCNFSM4MZWXLWQ.

belfhi commented 4 years ago

@CFGrote I can talk to Frank, we don't support Python 2.7 any more, so maybe it makes sense to modify the module to not set the python path...

takluyver commented 4 years ago

Thanks Carsten. I found the module, but I was trying to test a branch that @ejcjason was working on, so I want to do the installation myself.

I'll try building it with the intel module loaded.

takluyver commented 4 years ago

I tried with module load intel (it seems to have picked intel/2020 as the default version), but I still see the same error about -lgomp.

CFGrote commented 4 years ago

have you run

$> source `which compilervars.sh` intel64

assuming you are on a 64bit machine. that loads the intel lib and include paths.

carsten

-- Dr. Carsten Fortmann-Grote PaNOSC WP5 leader

On 5/15/20 5:06 PM, Thomas Kluyver wrote:

I tried with |module load intel| (it seems to have picked |intel/2020| as the default version), but I still see the same error about |-lgomp|.

— You are receiving this because you were mentioned. Reply to this email directly, view it on GitHub https://github.com/PaNOSC-ViNYL/SimEx/issues/196#issuecomment-629287283, or unsubscribe https://github.com/notifications/unsubscribe-auth/ADWCQR6EBRND2EA6ISKZ66TRRVLA7ANCNFSM4MZWXLWQ.

PaNOSC-ViNYL / SimEx

Installation challenges #196