GMT crashes when dealing with classic netcdf files after 4.7.4 update

[PaulWessel]

After my macports distro updated to netcdf 4.7.4 overnight, our GMT test suite reports these types of messages for any GMT command that tries to read one of our netcdf grids:

gmt (gmt_api.c:4658(api_import_grid)): NetCDF: Attempting netcdf-4 operation on netcdf-3 file [tut_bathy.nc]

There has been no change to GMT in regards to grid reading for weeks and all was working well yesterday. Something backwards incompatible must have been introduced - ideas?

[PaulWessel]

Traced in the debugger to this line:

nc_inq_var_deflate (ncid, z_id, &shuffle, &deflate, &deflate_level);

So previous behavior was (probably) to just return if the netcdf file is not netcdf 4 but now it instead throws this message. Was the change intentional? The file in question is netcdf3 classic.

[PaulWessel]

We are able to work around this in the GMT repo by adding a check if the grid is netcdf4 before calling this function, but users of older GMT releases may get a surprise when their repo updates to netcdf 4.7.4.

[DennisHeimbigner]

There were some cases where the older netcdf c code was ignoring attempts to use netcdf-4 functions (like filters and chunking) on netcdf-3 files. In the latest release we had modified the code to return an error instead of ignoring it. Sorry you got caught in that change. =Dennis Heimbigner   Unidata

…

On 4/4/2020 3:08 PM, Paul Wessel wrote: Traced in the debugger to this line: |nc_inq_var_deflate (ncid, z_id, &shuffle, &deflate, &deflate_level);| So previous behavior was (probably) to just return if the netcdf file is not netcdf 4 but now it instead throws this message. Was the change intentional? The file in question is netcdf3 classic. — You are receiving this because you are subscribed to this thread. Reply to this email directly, view it on GitHub <#1689 (comment)>, or unsubscribe <https://github.com/notifications/unsubscribe-auth/AAG47W3POHFLVGECEO7EXNDRK6OWJANCNFSM4L7EHKKA>.

[PaulWessel]

OK, we have already fixed this issue in our GitHub repo. I can understand you may consider the change a netCDF bug fix in that regard, but it also breaks 10 years of backward compatibility for others I would think. Probably deserving of an entry in the release notes, though. Best, Paul

[dopplershift]

I think this was fixed by #1691.

[edwardhartnett]

This may not have been fixed by #1691. That PR fixed this problem with nc_inq_var_szip/nc_inq_var_deflate. The problem still exists with nc_inq_var_chunking and nc_inq_var_filter, if I understand correctly. I believe these will be fixed by @DennisHeimbigner .

Let me know if you need help getting these fixed.

[WardF]

Fwiw, once these final fixes are in, we'll be doing a quick 4.7.x release and then moving on to 4.8.x.

[DennisHeimbigner]

What happened to the idea of using dispatch entries
NC_SHOWOFF_xxx?

[edhartnett]

Yes, did you want me to do that? Hold on, I will put up a PR shortly...

…

On Mon, Apr 13, 2020 at 3:21 PM Dennis Heimbigner ***@***.***> wrote: What happened to the idea of using dispatch entries NC_SHOWOFF_xxx? — You are receiving this because you are subscribed to this thread. Reply to this email directly, view it on GitHub <#1689 (comment)>, or unsubscribe <https://github.com/notifications/unsubscribe-auth/ABCSXXBUBITG64NBMJPJV63RMN66FANCNFSM4L7EHKKA> .