Michele Simionato

Today @VSilva noted that we need to store only mean and std after the liquefaction sampling, so that should solve most of my concerns.

This is blocked because on Windows the event_based_risk tests fail terribly (cannot decrement, permission denied, slist already enabled?). Removing ebrisk helps but not enough.

We discovered recently that h5py 3.4 started using hdf5 1.12 which uses locks on Windows, so in order to go back to the previous behavior one has to set HDF5_USE_FILE_LOCKING:...

After 9 months, is there any news on this issue?

Still nothing after another year.

You are effectively disabling the vectorization, i.e. there is an interpolation for each magnitude, even if the magnitudes are all equal: this is no good performance-wise. I need a test...

@cossatot, the way to extract the rupture IDs is explained in this comment: https://github.com/gem/oq-engine/pull/7961#issuecomment-1184162163

No, the beauty of this approach is that it works for all kind of sources.

> will the rupids returned in this way match that from using source.iter_ruptures() Absolutely yes, this is the point of it. > the first field of the dstore['source_info'] array looks...

There is way to run a preclassical calculation without splitting the sources so that there is correspondence between source_info and get_sources. Here is how to do it: ```python from openquake.calculators.base...