Memory leaks limit number of HepMC events that can be processed

[Sourav Mandal]

Due to serious memory leaks in HepMCConverter, one can only process a limited number of events in an HepMC file. For example, on my machine w/ 4GB of RAM I can only read in 140K LHC @ TeV top-pair events (generated by Herwig++) before Delphes exits.

I am attaching a patch which corrects these memory leaks. Now, I can process an arbitrary number of events without ever consuming more than 200MB of RAM.

The patch contains the following fixes:

1) Clear index_to_particle vector and particle_to_index map when reading in a new event.

2) Add virtual keywords to PdgParticle and PdgTable destructors. This is necessary for the clear method to work.

[Sourav Mandal]

Sorry, the second sentence should read: For example, on my machine w/ 4GB of RAM I can only read in 140K top-pair events (7 TeV LHC, generated by Herwig++) before Delphes quits.

[favereau]

Thanks for this too, merging...

Out of curiosity, why do we need virtual destructors ?

[Sourav Mandal]

Indeed you're right, for simply clearing index_to_particle and particle_to_index the virtual keyword should not required, as they hold instances of PdgParticle itself, not instances of subclasses. Same with the container holding PdgTable.

If nowhere in Delphes are PdgParticle and PdgTable subclassed, it's safe to leave out the virtual keyword.

However, generically, ​virtual destructors are necessary to avoid undefined behavior.