Dennis Heimbigner

The zarr code should not interact with DAP code except that both use libcurl. So there should be no problem.

No need to hold up most merges. It is only really big ones that may cause trouble.

Thanks. What is the general state of thread support in pnetcdf?

There is (was?) an hdf5 build flag that serialized access to the hdf5 API. So irrespective of netcdf-c, you need to build hdf5 with that flag set. The most important...

There are two ways to do this, I think. The simplest is to do what hdf5 did, which was to serialize access to the netcdf library by wrapping all calls...

But you need to make sure that you can isolate all metadata accesses thru some closed internal API. You need to do some significant code examination to find all the...

Personally my guess is that locking the global data is better than locking the code. We can do some name changing on the known global data to catch all references.

Maybe, but I bet if you look, you find any number of places where structs are accessed directly by field. These would all have to be changed to use get/set...

Remember, it is not enough to lock modifications. Reads must also be locked.

The short answer is yes. The basic problem is that if some other thread is making changes to the meta-data, modifying or adding an attribute, say. Then without detailed code...