Fix update of the magnetic field

[romainschotter]

When using the standard CCDB loader, the magnetic field is initialised correctly.

However, when processing a second file from a different run, the TGeoGlobalMagField instance is deleted (

AliceO2/Detectors/Base/src/Propagator.cxx

Line 131 in 10eb61e

delete TGeoGlobalMagField::Instance();

), while the corresponding pointer mField is not reset (

AliceO2/Detectors/Base/src/Propagator.cxx

Line 51 in 10eb61e

mField = static_cast<o2::field::MagneticField*>(TGeoGlobalMagField::Instance()->GetField());

).

As a result, it becomes a dangling pointer and can return completely nonsensical magnetic field values (e.g. around 2e+22). The pointer therefore needs to be refreshed by calling updateField. Furthermore, updateField only updates the magnetic field if the pointer mField is a nullptr. If mField exists but points to invalid memory, updateField does not update the magnetic field, so this logic also had to be fixed.

Thank you @fchinu for reporting this bug!

@ddobrigk @dsekihat for your information!

[shahor02]

@romainschotter thanks, the fix is correct.
Could you hint at how you re-initialise it in the analysis for different runs?
In principle, if it is done via GRPGeomHelper, there no need to delete the field: the field GRP is a function if timestamp rather than the run number, one simply updates the current values.