re: https://github.com/Unidata/netcdf-c/issues/1330
The ncgen utility is documented to accept octal integer constants
if the leading digit is zero. This was not implemented. Fix ncgen.l
to properly handle such constants. Also add a test to c0.cdl.
re: https://github.com/Unidata/thredds/issues/1224
[note that this is an issue in thredds, but the fix is in netcdf-c]
A thredds server can encode a netcdf-4 file into DAP2
by flattening names to include the containing group path,
where the group names are separated by '/'.
But the '/' is prohibited in netcdf names even if escaped
(a decision before my time).
So, if the netcdf-c/libdap2 code encounters a DAP2 name with '/'
characters, the '/' characters are converted to the string
%2f. Unfortunately, there is a glitch, namely that converting
the leading '/' produces a name that is still illegal. This PR
modifies the code to just drop the leading '/' character.
re: issue https://github.com/Unidata/netcdf-c/issues/1316
The DAP2 data model does not have a signed byte type,
but netcdf-3 does have (only) a signed byte type.
So, when converting a netcdf-3 signed byte typed variable to
a DAP2 unsigned byte, the following actions are taken by thredds:
1. The variable type is marked as DAP2 (unsigned) byte.
2. A special attribute, "_Unsigned=false" is set for the variable
3. The corresponding "_FillValue" attribute, if any, is up-converted
to the DAP2 Int16 type in order to hold, with sign, any signed byte
fill value.
On the netcdf-c side, this looks like a fillvalue type mismatch and causes
an error. Instead, the netcdf-c dap2 conversion code needs to recognize
this hack and undo it locally.
So this change looks for the above case, and if found, then it properly
converts the _FillValue type to netcdf-3 signed byte.
Since DAP2 supports both signed and unsigned integers of sizes 16 and 32 bits,
this should be the only hack needed (famous last words).
It may later be desirable for the thredds DAP2 converter to modify its
behavior as well.
