Edward Hartnett

Thanks for these documentation clarifications! Very helpful!!

I think that def_var_deflate() should check for legal types and return a NC_EINVAL.

This is already accomplished by simply not using the NC_CLASSIC_MODEL flag. The reason for the classic model is to keep the large body of existing test codes working (i.e. nc_test)....

NC_CLASSIC_MODEL promises code that will work in netcdf classic applications, just by changing the mode flag. It's a way of guaranteeing that users can always test code with netCDF classic,...

I think the idea has merit but the changes in current behavior break code compatibility. Even a minor change in the operation of nc_redef() is causing my PIO tests to...

I believe this issue can be closed...

I strongly disagree with the goal. Also there is more to the classic model than just enddef/redef. The point of the classic model is to ensure that a large code...

I would suggest that those other formats should behave like netCDF-4 without classic. That is, enddef/redef should be called automatically when needed. You may also feel the need to implement...

Did you build netcdf-c with --enable-parallel-tests? And did those tests pass?

I am taking a look at this issue today. I have added the test code above to tst_parallel6.c and indeed have reproduced the problem. I also added a sequential version...