ncap2 ignores "--cmp" when new dimensions are defined about nco HOT 7 CLOSED

The intent of your ncap2 command is unclear to me. However, I was able to reproduce an issue with ncap2 not compressing new variables. While doing so, I notice that using -L 1 instead of --cmp='dfl' might help. Please LMK if that workaround solves the immediate problem.

Also, var.set_miss(var2d) might be causing problems. set_miss() expects a scalar argument for the missing value, not a 2D array.

from nco.

Hi @czender ,

Thanks for the quick response. The intent is to transform a 2d array into time record 3d using the same data and applying the mask/missing value. I will revert back to using the L flag for ncap2 now and try to fix the fill_missing argument. Spotted the problem when using new filters. I can't trust my memory but I think the fill_missing call was tthe shortest path to have the array properly broadcasted when I coded ages ago.

from nco.

The snippet I typed above was wrong (see edited version now), so I was set/get missing correctly.

Indeed, using old and tried -LN flag will do the compression. However, some burden/overhead by the lacking of that is quite predictable:

...
opts="--cmp 'myfavcompression'"
...
ncks $opts ... a.nc b.nc 
#force -L1 or handroll a flag argument conversion (we don't want that...)
ncap2 -L1 -s ... b.nc c.nc #if we end here, compression is not what I asked for.
ncks $opts ... c.nc d.nc #if we got more on the plate, we have a small read-overhead from reading a larger c.nc

Not the end of the world, but not optimal either. I assume most users somehow inject/force default arguments to all calls, particularly compression.

from nco.

@hmb1 As this thread shows, ncap2 does not support the --cmp option when simultaneously creating new variables (or something like that). I thought I had hooked up the --cmp support in all appropriate ncap2 subroutines but apparently I missed one. Any idea which one? i.e., where does ncap2 call nc_def_var() without an immediately following call to nco_flt_def_*?

from nco.

Okay, nevermind. I think I found the locations that I had missed before. There are two in prs_cls.cc. I'll add this to the TODO list.

from nco.

@ocehugo the latest NCO snapshot includes a fix for the problem you reported. If you can, please test the snapshot and LMK if it works for you. Otherwise it will be in NCO 5.1.6 to be released in a week or two.

from nco.

Please reopen if this does not work for you with NCO 5.1.9

from nco.