Turn on the xarray convention for NCZarr format

re: https://github.com/pydata/xarray/issues/6374

As a result of a discussion about Xarray (see above issue),
I decided to turn on the xarray convention for NCZarr datasets
where possible so that xarray can read a larger set of nczarr
generated datasets.

This causes the following changes:
* If the user wants to generate a pure zarr file, then the mode "zarr" must be explicitly used; it is no longer the case that "mode=xarray" or mode="noxarray"
implies "mode=zarr".
* It is still the case that "mode=noxarray" will turn off the XArray convention.

The following conditions will cause ''_ARRAY_DIMENSIONS'' to not be written.
* The variable is not in the root group,
* Any dimension referenced by the variable is not in the root group.

.github/workflows/run_tests_ubuntu.yml

@@ -4,7 +4,7 @@
 
 name: Run Ubuntu/Linux netCDF Tests
 
-on: [ pull_request ]
+on: [pull_request]
 
 jobs:
 

RELEASE_NOTES.md

@@ -7,6 +7,7 @@ This file contains a high-level description of this package's evolution. Release
 
 ## 4.8.2 - TBD
 
+* [Enhancement] Turn on the XArray convention for NCZarr files by default. See [Github #????](https://github.com/Unidata/netcdf-c/pull/????).
 * [Enhancement] Improve filter support. More specifically (1) add nc_inq_filter_avail to check if a filter is available, (2) add the notion of standard filters, (3) cleanup szip support to fix interaction with NCZarr. See [Github #2245](https://github.com/Unidata/netcdf-c/pull/2245).
 * [Bug Fix] Require that the type of the variable in nc_def_var_filter is not variable length. See [Github #/2231](https://github.com/Unidata/netcdf-c/pull/2231).
 * [File Change] Apply HDF5 v1.8 format compatibility when writing to previous files, as well as when creating new files.  The superblock version remains at 2 for newly created files.  Full backward read/write compatibility for netCDF-4 is maintained in all cases.  See [Github #2176](https://github.com/Unidata/netcdf-c/issues/2176).

docs/nczarr.md

@@ -136,16 +136,18 @@ Note that It should be the case that zipping a _file_
 format directory tree will produce a file readable by the
 _zip_ storage format, and vice-versa.
 
-By default, _mode=zarr_ also supports the XArray _\_ARRAY\_DIMENSIONS_ convention. The _noxarray_ mode tells the library to disable the XArray support.
+By default, the XArray convention is supported and used for
+both NCZarr files and pure Zarr files. This
+means that every variable in the root group whose named dimensions
+are also in the root group will have an attribute called
+*\_ARRAY\_DIMENSIONS* that stores those dimension names.
+The _noxarray_ mode tells the library to disable the XArray support.
 
 The netcdf-c library is capable of inferring additional mode flags based on the flags it finds. Currently we have the following inferences.
-
-- _xarray_ => _zarr_
-- _noxarray_ => _zarr_
 - _zarr_ => _nczarr_
 
-So for example: ````...#mode=noxarray,zip```` is equivalent to this.
-````...#mode=nczarr,zarr,noxarray,zip
+So for example: ````...#mode=zarr,zip```` is equivalent to this.
+````...#mode=nczarr,zarr,zip
 ````
 <!--
 - log=&lt;output-stream&gt;: this control turns on logging output,
@@ -434,10 +436,13 @@ The value of this attribute is a list of dimension names (strings).
 An example might be ````["time", "lon", "lat"]````.
 It is essentially equivalent to the ````_NCZARR_ARRAY "dimrefs" list````, except that the latter uses fully qualified names so the referenced dimensions can be anywhere in the dataset.
 
-As of _netcdf-c_ version 4.8.1, The Xarray ''_ARRAY_DIMENSIONS'' attribute is supported.
-This attribute will be read/written by default, but can be suppressed if the mode value "noxarray" is specified.
+As of _netcdf-c_ version 4.8.2, The Xarray ''_ARRAY_DIMENSIONS'' attribute is supported for both NCZarr and pure Zarr.
+If possible, this attribute will be read/written by default,
+but can be suppressed if the mode value "noxarray" is specified.
 If detected, then these dimension names are used to define shared dimensions.
-Note that "noxarray" or "xarray" implies pure zarr format.
+The following conditions will cause ''_ARRAY_DIMENSIONS'' to not be written.
+* The variable is not in the root group,
+* Any dimension referenced by the variable is not in the root group.
 
 # Examples {#nczarr_examples}
 

libdispatch/dinfermodel.c

@@ -137,9 +137,9 @@ static const struct MACRODEF {
 {"dap4","mode",{"dap4",NULL}},
 {"s3","mode",{"s3","nczarr",NULL}},
 {"bytes","mode",{"bytes",NULL}},
-{"xarray","mode",{"nczarr","zarr","xarray",NULL}},
-{"noxarray","mode",{"nczarr","zarr","noxarray",NULL}},
-{"zarr","mode",{"nczarr","zarr","xarray",NULL}},
+{"xarray","mode",{"nczarr", NULL}},
+{"noxarray","mode",{"nczarr", "noxarray", NULL}},
+{"zarr","mode",{"nczarr","zarr", NULL}},
 {NULL,NULL,{NULL}}
 };
 
@@ -149,9 +149,8 @@ static const struct MODEINFER {
     char* inference;
 } modeinferences[] = {
 {"zarr","nczarr"},
-{"zarr","xarray"},
-{"xarray","zarr"},
-{"noxarray","zarr"},
+{"xarray","nczarr"},
+{"noxarray","nczarr"},
 {NULL,NULL}
 };
 

libnczarr/zarr.c

@@ -384,13 +384,12 @@ applycontrols(NCZ_FILE_INFO_T* zinfo)
     }
     /* Process the modelist first */
     zinfo->controls.mapimpl = NCZM_DEFAULT;
+    zinfo->controls.flags |= FLAG_XARRAYDIMS; /* Always support XArray convention where possible */
     for(i=0;i<nclistlength(modelist);i++) {
         const char* p = nclistget(modelist,i);
-	if(strcasecmp(p,PUREZARRCONTROL)==0) zinfo->controls.flags |= (FLAG_PUREZARR|FLAG_XARRAYDIMS);
-	else if(strcasecmp(p,XARRAYCONTROL)==0) zinfo->controls.flags |= (FLAG_XARRAYDIMS|FLAG_PUREZARR); /*xarray=>zarr*/
+	if(strcasecmp(p,PUREZARRCONTROL)==0) zinfo->controls.flags |= (FLAG_PUREZARR);
 	else if(strcasecmp(p,NOXARRAYCONTROL)==0) {
 	    noflags |= FLAG_XARRAYDIMS;
-	    zinfo->controls.flags |= FLAG_PUREZARR; /*noxarray=>zarr*/
 	}
 	else if(strcasecmp(p,"zip")==0) zinfo->controls.mapimpl = NCZM_ZIP;
 	else if(strcasecmp(p,"file")==0) zinfo->controls.mapimpl = NCZM_FILE;

libnczarr/zsync.c

@@ -695,20 +695,33 @@ ncz_sync_atts(NC_FILE_INFO_T* file, NC_OBJ* container, NCindex* attlist, int isc
 
     if(container->sort == NCVAR) { 
         if(isrootgroup && isxarray) {
+	    int dimsinroot = 1;
 	    /* Insert the XARRAY _ARRAY_ATTRIBUTE attribute */
 	    if((stat = NCJnew(NCJ_ARRAY,&jdimrefs)))
 	        goto done;
-	    /* Walk the dimensions and capture the names */
+	    /* Walk the dimensions to check in root group */
 	    for(i=0;i<var->ndims;i++) {
 	        NC_DIM_INFO_T* dim = var->dim[i];
-		char* dimname = strdup(dim->hdr.name);
-		if(dimname == NULL) {stat = NC_ENOMEM; goto done;}
-	        NCJaddstring(jdimrefs,NCJ_STRING,dimname);
-   	        nullfree(dimname); dimname = NULL;
+		/* Verify that the dimension is in the root group */
+		if(dim->container && dim->container->parent != NULL) {
+		    dimsinroot = 0; /* dimension is not in root */
+		    break;
+		}
+	    }
+	    if(dimsinroot) {
+		/* Walk the dimensions and capture the names */
+		for(i=0;i<var->ndims;i++) {
+		    char* dimname;
+	            NC_DIM_INFO_T* dim = var->dim[i];
+		    dimname = strdup(dim->hdr.name);
+		    if(dimname == NULL) {stat = NC_ENOMEM; goto done;}
+	            NCJaddstring(jdimrefs,NCJ_STRING,dimname);
+   	            nullfree(dimname); dimname = NULL;
+		}
+	        /* Add the _ARRAY_DIMENSIONS attribute */
+	        if((stat = NCJinsert(jatts,NC_XARRAY_DIMS,jdimrefs))) goto done;
+	        jdimrefs = NULL;
 	    }
-	    /* Add the _ARRAY_DIMENSIONS attribute */
-	    if((stat = NCJinsert(jatts,NC_XARRAY_DIMS,jdimrefs))) goto done;
-	    jdimrefs = NULL;
         }
     }
     /* Add Quantize Attribute */
@@ -736,9 +749,11 @@ ncz_sync_atts(NC_FILE_INFO_T* file, NC_OBJ* container, NCindex* attlist, int isc
 	    /* Insert the _NCZARR_ATTR attribute */
             if((stat = NCJnew(NCJ_DICT,&jdict)))
                 goto done;
-            if((stat = NCJinsert(jdict,"types",jtypes))) goto done;
+	    if(jtypes != NULL)
+                {if((stat = NCJinsert(jdict,"types",jtypes))) goto done;}
             jtypes = NULL;
-            if((stat = NCJinsert(jatts,NCZ_V2_ATTR,jdict))) goto done;      
+	    if(jdict != NULL)
+                {if((stat = NCJinsert(jatts,NCZ_V2_ATTR,jdict))) goto done;}
             jdict = NULL;
 	}
         /* write .zattrs path */
@@ -1449,7 +1464,7 @@ define_vars(NC_FILE_INFO_T* file, NC_GRP_INFO_T* grp, NClist* varnames)
 
     if(zinfo->controls.flags & FLAG_PUREZARR) purezarr = 1;
     if(zinfo->controls.flags & FLAG_NCZARR_V1) formatv1 = 1;
-    if(zinfo->controls.flags & FLAG_XARRAYDIMS) {purezarr = 1; xarray = 1;}
+    if(zinfo->controls.flags & FLAG_XARRAYDIMS) {xarray = 1;}
 
     /* Load each var in turn */
     for(i = 0; i < nclistlength(varnames); i++) {
@@ -2246,10 +2261,8 @@ computedimrefs(NC_FILE_INFO_T* file, NC_VAR_INFO_T* var, int purezarr, int xarra
 
     assert(zfile && zvar);
 
-    /* xarray => purezarr */
-    assert(!xarray || purezarr);
-
-    if(xarray) {/* Read in the attributes to get xarray dimdef attribute; Note that it might not exist */
+    if(purezarr && xarray) {/* Read in the attributes to get xarray dimdef attribute; Note that it might not exist */
+	/* Note that if xarray && !purezarr, then xarray will be superceded by the nczarr dimensions key */
         char zdimname[4096];
 	if(zvar->xarray == NULL) {
 	    assert(nclistlength(dimnames) == 0);

nc_test4/tst_broken_files.c

@@ -27,8 +27,9 @@ main() {
         if(fwrite(TRUNCATED_FILE_CONTENT, sizeof(char), sizeof(TRUNCATED_FILE_CONTENT), fp) != sizeof(TRUNCATED_FILE_CONTENT)) ERR;
         fclose(fp);
 
-        int  ncid;
-        if (nc_open(FILE_NAME, 0, &ncid) != NC_EHDFERR) ERR;
+        int  ncid, stat;
+        stat = nc_open(FILE_NAME, 0, &ncid);
+        if (stat != NC_EHDFERR && stat != NC_ENOFILTER && stat != NC_ENOTNC) ERR;
     }
 
     {

nczarr_test/ref_newformatpure.cdl

@@ -1,9 +1,10 @@
 netcdf ref_oldformat {
 dimensions:
+	lat = 8 ;
 	_zdim_8 = 8 ;
 	_zdim_10 = 10 ;
 variables:
-	int lat(_zdim_8) ;
+	int lat(lat) ;
 		lat:_FillValue = -1 ;
 		lat:lat_attr = "latitude" ;
 data:

nczarr_test/run_purezarr.sh

@@ -15,7 +15,7 @@ testcase() {
 zext=$1
 
 echo "*** Test: pure zarr write then read; format=$zext"
-fileargs tmp_purezarr "mode=noxarray,$zext"
+fileargs tmp_purezarr "mode=zarr,noxarray,$zext"
 deletemap $zext $file
 ${NCGEN} -4 -b -o "$fileurl" $srcdir/ref_purezarr_base.cdl
 ${NCDUMP} $fileurl > tmp_purezarr_${zext}.cdl
@@ -27,9 +27,9 @@ fileargs tmp_xarray "mode=zarr,$zext"
 ${NCGEN} -4 -b -o "$fileurl" $srcdir/ref_purezarr_base.cdl
 ${NCDUMP} $fileurl > tmp_xarray_${zext}.cdl
 diff -b ${srcdir}/ref_xarray.cdl tmp_xarray_${zext}.cdl
-echo "*** Test: pure zarr reading nczarr; format=$zext"
-fileargs tmp_nczarr "mode=nczarr,$zext"
 
+echo "*** Test: pure zarr reading nczarr; format=$zext"
+fileargs tmp_nczarr "mode=nczarr,noxarray,$zext"
 deletemap $zext $file
 ${NCGEN} -4 -b -o "$fileurl" $srcdir/ref_whole.cdl
 fileargs tmp_nczarr "mode=zarr,$zext"