Detect whether VR was compiled with NOMASS option and change the VR interface accordingly

See discussion on the VELOCIraptor github, issue 66.

Fixes #654 (closed).

If you confirm that this works for you, I'll merge it. Thanks!

@pelahi you might be interested in this as well.

mentioned in issue #654 (closed)

changed the description

This looks like it should fix the problems I found and it gets much further now, but I just had a segfault on the 75Mpc box at the "Sort particles and compute properties" stage.

Could you run with Verbose=1? That might help if you want to try running in the background and don't have time to debug right away.

It crashes on the 12Mpc Eagle DMO low redshift ICs too, but only if I use >1 MPI rank. Here's the last bit of the output with verbose=1:

Getting Hierarchy 4069
Done
1 Sort particles and compute properties of 4069 objects 
1 Calculate properties using minimum potential particle as reference 
1 Sort particles by binding energy
0 Sort particles and compute properties of 2036 objects 
0 Calculate properties using minimum potential particle as reference 
0 Sort particles by binding energy
1 getting CM
1 Done getting CM in 0.00311804
Done FOF masses 0.000142813
1 getting energy
0 getting CM
0 Done getting CM in 0.0100329
Done FOF masses 0.000133038
0 getting energy
1 Have calculated potentials 12.8857
1Done getting energy in 12.9864
1 getting bulk properties

===================================================================================
=   BAD TERMINATION OF ONE OF YOUR APPLICATION PROCESSES
=   PID 207647 RUNNING AT m7013
=   EXIT CODE: 11
=   CLEANING UP REMAINING PROCESSES
=   YOU CAN IGNORE THE BELOW CLEANUP MESSAGES
===================================================================================

I'll try it in ddt.

Odd. The segfault is at this line in Particle.h:

        Double_t Radius() const {return sqrt(position[0]*position[0]+position[1]*position[1]+position[2]*position[2]);}

Somehow position is an invalid pointer. This happens in the loop at line 490 of substructureproperties.cxx where it loop over particles within Rv.

Ok, it's a typo in velociraptor_interface.c, line 524:

#ifdef HAVE_VELOCIRAPTOR_WIH_NOMASS
  /* Assume all particles have the same mass */
  double DM_mass = 0.;
  for (size_t i = 0; i < e->s->nr_gparts; ++i) {
...

Note the missing T in WITH, which explains why my particle mass is 6.2e-323!

Also, massuniformbox in struct siminfo should be mass_uniform_box.

And velociraptor is reading out mass_uniform_box in initvelociraptor(), which is before swift sets it in velociraptorinvoke(). If I move the assignment later velociraptor then it runs to completion on the 12Mpc box. The branch VR_detection_typo has the changes I made in Swift. I'll make a PR for Velociraptor now.

ah thanks! Silly mistake...

added 2 commits

• d541ca06 - Fix typos in velociraptor interface that cause problems with VR_NO_MASS
• f0263198 - Merge branch 'VR_detection_typo' into 'VR_detection'

Compare with previous version

I have merged your fixes in this branch.

If the crash above is fixed then this can be merged I think.

Pascal merged my velociraptor change so I'll check that the 75Mpc box runs with this swift branch and the lastest velociraptor master.

Velociraptor with NO_MASS on the 75Mpc box worked, so I think this can be merged.

Thanks both! Good this is all fixed.

merged

mentioned in commit 1d7fce84