Thread safety.

It seems that the evaluation of shape functions is not thread safe. Is this something that can be fixed easily?

added bug discussion labels

Could you elaborate? Which shape functions are not thread safe?

Maybe related? !220 (merged)

@tkoch !220 (merged) is about thread safety of the cache not the shape functions themselves. I guess to understand the problem one needs to know more details on what is actually meant here.

Personally I never encountered problems with thread safety for parallel assembly in dune-fufem, because (in dune-fufem) all LFEs, LFECaches and shape function caches are thread-local.

... this is because dune-functions global bases are designed to make all mutable data thread-local.

On the meeting we agreed that a single LFE does not need to be thread-safe. However, we would like to guarantee that there's no race conditions between separate copies of the same LFE. @santiago.ospina and @carsten.graeser will take care for documenting this.

Grepping through dune-localfunctions it seems that the (former) generic LFE implementations use shared_ptss which may violate the above given guarantee. @robert.kloefkorn want's to have a look at this.

assigned to @santiago.ospina