PLUGIN from alternative location

This was indeed requested/put in place for ATLAS years ago.
Is this not working anymore?

For reference you should put your plugin in a directory named
XXX/MG5aMC_PLUGIN
and put XXX in your PYTHONPATH

I guess that you plan to have the plugin on the same read-only filesystem (or at least control by the same people)
Remember that the plugin have full control on the code and can overwrite any function/...
So if the point of the read-only filesystem is to be sure that the code stay the same, then you need the same level of protection for the plugin directory.

Cheers,

Olivier

> On 18 May 2022, at 18:20, Hannes <email address hidden> wrote:
>
> New question #701855 on MadGraph5_aMC@NLO:
> https://answers.launchpad.net/mg5amcnlo/+question/701855
>
> Hi,
>
> in ATLAS we have a central MG5_aMC installation on a read-only file system that cannot easily be modified.
>
> For that reason it would be very convenient for us to load a plugin from an alternative location outside the MG5_aMC directory (that could for example be specified by an environment variable). Is something like this possible?
>
> Cheers,
> Hannes
>
> --
> You received this question notification because you are an answer
> contact for MadGraph5_aMC@NLO.

Hi Olivier,

I have tried that, I have copied the code from

https://cp3.irmp.ucl.ac.be/projects/madgraph/wiki/FAQ-General-15

to

/tmp/hmildner/test/MG5aMC_PLUGIN/user_filter.py

and added

:/tmp/hmildner/test

to PYTHONPATH

but when I run

python $MADPATH/bin/mg5_aMC

and

generate p p > t t~ --diagram_filter

I still get

Error detected in "generate p p > t t~ --diagram_filter"
write debug file MG5_debug
If you need help with this issue please contact us on https://answers.launchpad.net/mg5amcnlo
str : user filter required to be defined in PLUGIN/user_filter.py with the function remove_diag(ONEDIAG) which returns True if the diagram has to be removed

Cheers,
Hannes

in v2.9.9

Aahhh user_filter is not a real plugin this is a (extremelly dangerous) hack.

So that indeed does not have a way to to be callable via environment variable.
I'm not in favor to have a way to do that actually. But adding such support can be done ... via a PLUGIN (which can be set by environment variable).

Hi Olivier,

thanks. It's an extremely dangerous hack because of the physics implications of removing some diagrams?
How would we go about implementing such a plugin in cases where it might still be required?

Cheers,
Hannes

> thanks. It's an extremely dangerous hack because of the physics implications of removing some diagrams?

yes breaking of gauge invariance in particular.

> How would we go about implementing such a plugin in cases where it might still be required?

I would have say by overwritting the function
 def apply_user_filter(self, diag_list):
of diagram_generation.py

but looks like this is already using the plugin magic function:

 if True:
            remove_diag = misc.plugin_import('user_filter',
                                             'user filter required to be defined in PLUGIN/user_filter.py with the function remove_diag(ONEDIAG) which returns True if the diagram has to be removed',
                                             fcts=['remove_diag'])

and misc.plugin_import has the method implemented for ATLAS and the plugin...

    try:
        _temp = __import__('PLUGIN.%s' % module, globals(), locals(), fcts, level)
    except ImportError:
        try:
            _temp = __import__('MG5aMC_PLUGIN.%s' % module, globals(), locals(), fcts, level)
        except ImportError:
            raise MadGraph5Error(error_msg)

So maybe this is a python3 issue...

Cheers,

Olivier

> On 18 May 2022, at 23:11, Hannes <email address hidden> wrote:
>
> Question #701855 on MadGraph5_aMC@NLO changed:
> https://answers.launchpad.net/mg5amcnlo/+question/701855
>
> Hannes posted a new comment:
> Hi Olivier,
>
> thanks. It's an extremely dangerous hack because of the physics implications of removing some diagrams?
> How would we go about implementing such a plugin in cases where it might still be required?
>
> Cheers,
> Hannes
>
> --
> You received this question notification because you are an answer
> contact for MadGraph5_aMC@NLO.