Remove the use of average_* variables and use time_bnds variables instead

[uramirez8707]

Fixes #273
Fixes #274

I tried this using output from the old diag manager (which includes the average_* variables) and this updated reproduces the master branch.

I tried this using output from the new diag manager (which does not have the average_* variables). This update reproduces the output from the old diag manager.

[uramirez8707]

FYI @ceblanton

[ilaflott]

tested on GFDL work-station, using a NetCDF file from fre-python-tool's time-averager.

exact function call to create averaged file: ./postprocessing/timavg/timavg.csh -r8 -mb -o FOO atmos.197901-198312.LWP.nc

compared against output of old timavg.csh output genereated for fre-python-tools tests.

stored the header info from ncdump -h into text files, then diff, yields

1c1
< netcdf FOO {
---
> netcdf fre_nctools_timavg_CLI_test_r8_mb_atmos_LWP_1979_5y {
16a17
> 		LWP:time_avg_info = "average_T1,average_T2,average_DT" ;
46a48,56
> 	double average_T1(time) ;
> 		average_T1:long_name = "Start time for average period" ;
> 		average_T1:units = "days since 1979-01-01 00:00:00" ;
> 	double average_T2(time) ;
> 		average_T2:long_name = "End time for average period" ;
> 		average_T2:units = "days since 1979-01-01 00:00:00" ;
> 	double average_DT(time) ;
> 		average_DT:long_name = "Length of average period" ;
> 		average_DT:units = "days" ;

this is the difference in metadata desired and described in issue 273 and issue 274. so, YAY!

numerical results bitwise identical as well, can be seen by ncdiff calls between the two files.

[ilaflott]

all looks good to me

[ceblanton]

Let's confirm that timavg.csh chokes as we think it does and should if the time_bounds are missing.

if timavg.csh hard fails when run on old/uncompliant history files (that don't have time_bounds) no one will end up with incorrect output, at least.

[uramirez8707]

Let's confirm that timavg.csh chokes as we think it does and should if the time_bounds are missing.

if timavg.csh hard fails when run on old/uncompliant history files (that don't have time_bounds) no one will end up with incorrect output, at least.

This is what I get when I use a file that does not have the time_bounds:

 NETCDF ERROR in program time_average
 error getting varid for time bounds
 NetCDF: Variable not found

Program aborted. Backtrace:
#0  0x402c39 in ???
#1  0x403a63 in ???
#2  0x40b554 in ???
#3  0x40240c in ???
#4  0x2b70608e5554 in ???
#5  0x40243c in ???
#6  0xffffffffffffffff in ???
Abort

[ceblanton]

Thanks for checking that. I'll merge and deploy into fre-nctools/test at GFDL now.