(primarily from libsrc4)
into its own dispatch library
called libsrc5.
2. Fixed part of Jira NCF-253
by removing the need for the
pnetcdf_ndims field.
For some reason, the original
code tried to cache the variable
ranks rather than computing them
as needed. Fixed by doing
an ...inq_varndims call as needed.
3. found some places where NC_MAX_DIMS
was being stack allocated and changed
to heap allocation.
Still some cases in nc_test4.
(primarily from libsrc4)
into its own dispatch library
called libsrc5.
2. Fixed part of Jira NCF-253
by removing the need for the
pnetcdf_ndims field.
For some reason, the original
code tried to cache the variable
ranks rather than computing them
as needed. Fixed by doing
an ...inq_varndims call as needed.
3. found some places where NC_MAX_DIMS
was being stack allocated and changed
to heap allocation.
Still some cases in nc_test4.
So, it turns out that just freeing
the nc4_info is not enough;
The root group must also be reclaimed.
So, it appears the best approach
is to invoke an abort on the
failed file.
Problem was that the NC_create
code was not checking for the NC_CLASSIC_MODEL
mode flag in deciding what dispatch table to use.
This meant that it was then defaulting to use
the default format, and if that was changed
to e.g. NC_FORMAT_NETCDF4, then it would try
to create a netcdf-4 format file, even is
NC_CLASSIC_MODEL mode flag was set.
to do prefetch on either a lazy
or eager basis. Lazy means that
the prefetch does not occur
until and unless the client actually
makes a get_var request.
Also repaired a problem where
doing prefetch wrt a url that
has a constraint will prefetch
a whole variable if its constrained
size is small enough, even if the
underlying variable is too large
to warrant prefetch.
