generate_events tries to remove GpuAbstraction.h

[valassi]

Trying to create a gridpack and use generate_events I get

INFO: Creating gridpack 
Command "generate_events -f" interrupted with error:
OSError : Cannot call rmtree on a symbolic link
Please report this bug on https://bugs.launchpad.net/mg5amcnlo
More information is found in '/data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_tt.mad/run_01_tag_1_debug.log'.
Please attach this file to your report.
removing %s /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/gg_tt.mad/SubProcesses/P1_gg_ttx/GpuAbstraction.h

(I had to add a manual print to show the path, because of mg5amcnlo/mg5amcnlo#123)

I guess it just tries to delete G* and it does not expect GpuAbstraction.h, which it should not remove...

[valassi]

This is a possible fix 8e29029

@oliviermattelaer I guess this is best only in cudacpp, but maybe you have better ideas

(Yes I know this is using patches, but this is what we have now, and this file is already included there)

[oliviermattelaer]

Hi Andrea,

Sorry but not, python file CAN NOT be modified via patch.
For python file, this is not a question of style of coding/support/...

It is just WRONG to do that, the reason is that the code has priority to use MG5DIR/madgraph/interface/madevent_interface.py (same for other file) over bin/internal/madevent_interface.py
But since you do not patch (and you should not) MG5DIR/madgraph/interface/madevent_interface.py
the effect of the patch will be applied for some user and not for others.

I did implement the full framework needed to avoid any patch to any python file, so for sure we do not need any patch for any python file. We can do a peer programing together if you want to see how such change need to be properly handle if you want (or I can do it and you take a look, as you want)

[oliviermattelaer]

Actually here, the best is likely to update upstream madgraph to add a check that this is a directory and not a file (or to change the name of the file to avoid the conflict).

[valassi]

I did implement the full framework needed to avoid any patch to any python file,

Thanks (and apologies), my mistake. Indeed you had implemented this. I should have removed this madevent_interface.py from my instructions.

Now I created a patch upstream instead mg5amcnlo/mg5amcnlo#125

Actually here, the best is likely to update upstream madgraph to add a check that this is a directory and not a file

Thanks, this is exactly what I did in mg5amcnlo/mg5amcnlo#125

[valassi]

(Yes I know this is using patches, but this is what we have now, and this file is already included there)

Specifically, my mistake: "this file is already included there" was wrong, you had already removed it

[valassi]

I did implement the full framework needed to avoid any patch to any python file, so for sure we do not need any patch for any python file

Thanks again. I confirm that now my patches are created without any check on .py files. (I added a comment in #948 using mg5amcnlo/mg5amcnlo#125)

[valassi]

This is fixed in mg5amcnlo/mg5amcnlo#125

Submodule being updated in PR #949

Closing