cn_dir corrupts AGRIF input filenames

[fschwarzkopf]

Context

Input files (forcing etc.) cannot be found if cn_dir differs from ./ in nested configurations.

Analysis

In AGRIF configurations, when cn_dir differs from ./ inputfiles for the child grid(s) cannot be found. The prefix for the child(s)-filenames is inserted after the first / in cn_dir and not in front of the filename itself.

Example:

filename = 'myfile'
cn_dir = './'

--> /1_myfile

filename = 'myfile'
cn_dir = '/path/to/my/file/'

--> /1_path/to/my/file/myfile
instead of /path/to/my/file/1_myfile

[clevy]

Yes, this is the way AGRIF works...

[mscheinert]

On one hand, the effective path to some input-file is meant to be composed by cn_dir//clname, isn't it?
On the other hand any nest-specific file (not the path) needs a prefix consisting of the number of the nest and an underscore, e.g. '1_'.

And this works perfectly without any nests or with nests but with cn_dir only containing a simple './'. But with AGRIF the logic fails if the path in cn_dir has more than one sub-directory in it, because the Fortran function "INDEX()" (see code below) only returns the first position of a pattern, which is "/" in the case on hand. However, the actual position of interest in this pattern-search is the LAST occurrence of "/" in the string clname - in order to extract the filename from the full path without knowing about the original cn_dir.

see iom.F90:

      IF ( .NOT. Agrif_Root() .AND. .NOT. lliof ) THEN

         iln    = INDEX(clname,'/')

         cltmpn = clname(1:iln)
         clname = clname(iln+1:LEN_TRIM(clname))
         clname=TRIM(cltmpn)//TRIM(Agrif_CFixed())//'_'//TRIM(clname)
      ENDIF

Erroneous result for the given example

Given cn_dir='/path/to/my/file/' and an already composed clname='/path/to/my/file/my_file', the INDEX() call results in iln=1 instead of iln=17 and thus the extracted filename becomes 'path/to/my/file/my_file'. And consequently for the first nest we get: '1_path/to/my/file/my_file' instead of the intended '/path/to/my/file/1_my_file'.

Proposed SOLUTION:

Find the last occurrence of "/" by searching from the right to the left. Enable back-search in the INDEX() call when diagnosing iln in the code excerpt from above:

         iln    = INDEX(clname,'/',BACK=.true.)

[fschwarzkopf]

The above provided fix also works in NEMO4 (iom.F90 line 699; subroutine iom_open).

A similar erroneous behaviour occurs for writing output (using xios) into a directory different from the current working directory defined in file_definition;

e.g. <file_definition type="one_file" name="/path/to/output/@expname@_@freq@_@startdate@_@enddate@" sync_freq="1mo" min_digits="4">

This can be analogously fixed in subroutine iom_update_file_name, substituting line 2323:

  IF( jn == 1 .AND. TRIM(Agrif_CFixed()) /= '0' )   clname = TRIM(Agrif_CFixed())//"_"//TRIM(clname)

by

         IF( jn == 1 .AND. TRIM(Agrif_CFixed()) /= '0' ) THEN
          iln    = INDEX(clname,'/',BACK=.true.)
          cltmpn = clname(1:iln)
          clname = clname(iln+1:LEN_TRIM(clname))
          clname = TRIM(cltmpn)//TRIM(Agrif_CFixed())//'_'//TRIM(clname)
         ENDIF  

iln and cltmpn need to be declared:

    CHARACTER(LEN=256) :: cltmpn
    INTEGER            :: iln

Btw: only one file_definition is used. I.e. if file_definitions in file_def_nemo-oce.xml, file_def_nemo-ice.xml and others differ, only the last one given in context_nemo.xml is used.

[fschwarzkopf]

Corrected iom.F90 (NEMO4.0)

[jchanut]

In 11554:

Error: Failed to load processor CommitTicketReference

No macro or processor named 'CommitTicketReference' found