Global/local attributes in TS files

[aclsce]

Global/local attributes defined for a file are not reported in Time Series files.
Example :

<file name="file1" output_freq="1d" timeseries="both" >

<variable id="project_id" type="string" > CMIP6/CMIP6 </variable>
<field field_ref="soce" name="soce">

...

The file "file1.nc" contains
:project_id = "CMIP6/CMIP6" ;
whereas in TS file "file1_soce.nc" there is no :
:project_id = "CMIP6/CMIP6" ;

[rlacroix]

If I understand correctly the variables defined for a file should be copied to all the automatically generated files when timeseries are enabled. Is that correct?

[aclsce]

Yes, that's correct.

[rlacroix]

Fixed in r928.

[aclsce]

Thanks to the commit 928, that's ok for global attribute (variable of a file). But, variables defined for a field (which is included in a file) should be copied to all the automatically generated files containing this variable (including Time series file).
Example :

<file name="file1" timeseries="both" >
 <field field_ref="toce" name="thetao" >
     <variable name="realm" type="string" > ocean </variable>
...

The file "file1.nc" contains

float thetao(time_counter, axis_oce, y, x) ;
		thetao:realm = "ocean" ;

whereas in file "file1_thetao.nc", the variable "thetao" does not contain :

thetao:realm = "ocean" ;

as attribute of thetao variable.

[rlacroix]

Ok, I will fix that shortly.

[rlacroix]

If I use a test case similar to what you showed above, I get a strange crash (unrelated to my change in r928) with the latest revision. Can you confirm that you are not seeing this?

[aclsce]

I tested with last revision 953 and I had no crash with my small test...only problem I mentionned above about local attributes and TS files.

[rlacroix]

Can you try the attached xml file with test_client.exe?

As far as I can tell there is an infinite recursion somewhere but I don't understand why you are not affected.

[aclsce]

OK, I am affected too. My small test was not the test_client (it was another one), that's why I haven't encountered a crash. I confirm that there is a crash by using the iodef.xml you attached and the test_client. It seems there is no crash if I put
<field field_ref="field_A" name="field_A" ts_enabled="false" />
but I don't know why...

[rlacroix]

I have some ideas but I couldn't really look into it yet.

[rlacroix]

I was able to confirm that my intuition was correct. Currently there is a conceptual bug that might result in infinite recursion. I will commit a fix later today.

[rlacroix]

Replying to rlacroix:

I was able to confirm that my intuition was correct. Currently there is a conceptual bug that might result in infinite recursion. I will commit a fix later today.

Fixed in r957.

But, variables defined for a field (which is included in a file) should be copied to all the automatically generated files containing this variable (including Time series file).

Fixed in r958, let's hope everything is fine this time.

[aclsce]

I confirm that seems ok now, thanks Remi !