In discussion #6, I noted a difference in how checksums were computed.
The essence of this problem is as follows:
The difference then is that the netcdf checksum includes the counts for variable length data and Hyrax excludes the counts.
Including the counts allows checksum computation at a lower level of the stack independent (more or less) of the dmr. I have implemented both and including the counts is very slightly faster, but not by much.
So, as a stake in the ground, I am proposing that we include the count when computing the checksum.
In discussion #6, I noted a difference in how checksums were computed.
The essence of this problem is as follows:
Including the counts allows checksum computation at a lower level of the stack independent (more or less) of the dmr. I have implemented both and including the counts is very slightly faster, but not by much.
So, as a stake in the ground, I am proposing that we include the count when computing the checksum.