[MITgcm-devel] missing-value in meta files

[Jean-Michel Campin]

Hi,

I am going to add the missing-value in the meta file,
but skipping this additional information if the argument for the missing-value
is the unlikely missing-value of one (the missing missing-value case).

But there is a problem when trying to use this missing-value from the
meta file (ascii) since it could be truncated differently from the value
in the binary file, and so making things more difficult to use.
For this reason, I would like to change, in pkg/diagnostics, the default
missing value to correspond to an integer value; this would reduce
(hopefully, even eliminate) the risk of different truncations.

I also think it would make easier/safer the use of missing value even with
NetCDF output from pkg/diagnostics. The truncation between the
array value and the stored missing-value is generally not a problem with
NetCDF files, since both are stored in the same file, using the same
precision. However, in the (very) special case where I put myself some
missing value (currently, UNSET_RL) in an _RS array, and fill-up
a diagnostics with this array to do a snap-shot output, I run into a
problem of different truncation when _RS is real*4.
I tested this and rdmnc.m was not able to identify the missing value
I added in the array.

The remaining question, assuming we go for changing the default missing value
in pkg/diagnostics, is which integer value do we pick ?
I prefer something different from the 123456789 (current UNSET_I),
to avoid confusion with UNSET_RL/RS.
I kind of like -999 or -999999 or any of the -9[9 ... 9] family, but it's
just my preference.

And just a short question to Martin: do we really need (or will need
one day) this misvalInt array ? Since all the storage arrays for diagnostics
are all _RL, not clear to me what it could be used for.

Cheers,
Jean-Michel