Add H5Dchunk_iter method for iterating over chunks (#6)

* Add H5Dchunk_iter method for iterating over chunks

This method iterates over all chunks in dataset, calling a user-supplied
callback with the chunk information and optional user supplied data.

The iterator is stopped when ITER_STOP is returned by the user-supplied
callback or the iterator is exhausted.

Existing methods to get chunk_info performs an iteration each time, so
to get many or all chunks causes SUM(i) for i = 0 -> N operations for N
chunks, as opposed to N operations when using this iterator for this use case.

* H5Dchunk_iter: test iterating all chunks, some chunks and failing iteration.

* H5D: move H5Dchunk_iter private methods to specific

* trace: add H5D_chunk_iter_op_t and trace H5D.c

* chunks-iter: document chunk_iter

* chunk-iter: chunk add FUNC_ENTER/FUNC_LEAVE macros

* Committing clang-format changes

Co-authored-by: github-actions <41898282+github-actions[bot]@users.noreply.github.com>

bin/trace

@@ -44,6 +44,7 @@ $Source = "";
                "H5D_vds_view_t"             => "Dv",
                "H5FD_mpio_xfer_t"           => "Dt",
                "H5FD_splitter_vfd_config_t" => "Dr",
+               "H5D_chunk_iter_op_t"        => "x",
                "herr_t"                     => "e",
                "H5E_direction_t"            => "Ed",
                "H5E_error_t"                => "Ee",

fortran/src/Makefile.in

@@ -839,9 +839,9 @@ LT_VERS_REVISION = 0
 LT_CXX_VERS_INTERFACE = 201
 LT_CXX_VERS_REVISION = 0
 LT_CXX_VERS_AGE = 1
-LT_F_VERS_INTERFACE = 200
-LT_F_VERS_REVISION = 1
-LT_F_VERS_AGE = 0
+LT_F_VERS_INTERFACE = 201
+LT_F_VERS_REVISION = 0
+LT_F_VERS_AGE = 1
 LT_HL_VERS_INTERFACE = 200
 LT_HL_VERS_REVISION = 1
 LT_HL_VERS_AGE = 0
@@ -851,12 +851,12 @@ LT_HL_CXX_VERS_AGE = 0
 LT_HL_F_VERS_INTERFACE = 200
 LT_HL_F_VERS_REVISION = 1
 LT_HL_F_VERS_AGE = 0
-LT_JAVA_VERS_INTERFACE = 200
-LT_JAVA_VERS_REVISION = 1
-LT_JAVA_VERS_AGE = 0
-LT_TOOLS_VERS_INTERFACE = 200
-LT_TOOLS_VERS_REVISION = 1
-LT_TOOLS_VERS_AGE = 0
+LT_JAVA_VERS_INTERFACE = 201
+LT_JAVA_VERS_REVISION = 0
+LT_JAVA_VERS_AGE = 1
+LT_TOOLS_VERS_INTERFACE = 201
+LT_TOOLS_VERS_REVISION = 0
+LT_TOOLS_VERS_AGE = 1
 AM_FCLIBS = $(LIBHDF5)
 
 # This is our main target, the fortran library

release_docs/RELEASE.txt

@@ -309,6 +309,21 @@ New Features
 
     Library:
     --------
+    - H5Dchunk_iter() added for iterating over chunks
+
+      This method iterates over all chunks in a dataset, calling a user-supplied
+      callback with the chunk information and optional user supplied data.
+
+      The iterator is stopped when ITER_STOP is returned by the user-supplied
+      callback or the iterator is exhausted.
+
+      Existing methods to get chunk_info perform an iteration each time, so
+      whereas getting many or all chunks with existing methods caused SUM(i) 
+      for i = 0 -> N operations for N chunks, using this iterator for this 
+      use case will get N chunks with N operations.
+
+      (2021/05/19)
+
     - H5Gcreate1() now rejects size_hint parameters larger than UINT32_MAX
 
       The size_hint value is ultimately stored in a uint32_t struct field,

src/H5D.c

@@ -1126,3 +1126,68 @@ H5Dget_chunk_info_by_coord(hid_t dset_id, const hsize_t *offset, unsigned *filte
 done:
     FUNC_LEAVE_API(ret_value)
 } /* end H5Dget_chunk_info_by_coord() */
+
+/*-------------------------------------------------------------------------
+ * Function:    H5Dchunk_iter
+ *
+ * Purpose:     Iterates over all chunks in dataset with given callback and user data.
+ *
+ * Parameters:
+ *              hid_t dset_id;          IN: Chunked dataset ID
+ *              H5D_chunk_iter_op_t cb  IN: User callback function, called for every chunk.
+ *              void *op_data           IN/OUT: Optional user data passed on to user callback.
+ *
+ * Callback information:
+ *      H5D_chunk_iter_op_t is defined as:
+ *
+ *        typedef int (*H5D_chunk_iter_op_t)(
+ *            const hsize_t *offset,
+ *            uint32_t filter_mask,
+ *            haddr_t addr,
+ *            uint32_t nbytes,
+ *            void *op_data);
+ *
+ *      H5D_chunk_iter_op_t parameters:
+ *          hsize_t *offset;        IN/OUT: Array of starting logical coordinates of chunk.
+ *          uint32_t filter_mask;   IN: Filter mask of chunk.
+ *          haddr_t addr;           IN: Offset in file of chunk data.
+ *          uint32_t nbytes;        IN: Size in number of bytes of chunk data in file.
+ *          void *op_data;          IN/OUT: Pointer to any user-defined data
+ *                                  associated with the operation.
+ *
+ *      The return values from an operator are:
+ *          Zero (H5_ITER_CONT) causes the iterator to continue, returning zero when all
+ *              elements have been processed.
+ *          Positive (H5_ITER_STOP) causes the iterator to immediately return that positive
+ *              value, indicating short-circuit success.
+ *          Negative (H5_ITER_ERROR) causes the iterator to immediately return that value,
+ *              indicating failure.
+ *
+ * Return:      Non-negative on success, negative on failure
+ *
+ * Programmer:  Gaute Hope
+ *              August 2020
+ *
+ *-------------------------------------------------------------------------
+ */
+herr_t
+H5Dchunk_iter(hid_t dset_id, H5D_chunk_iter_op_t cb, void *op_data)
+{
+    H5VL_object_t *vol_obj   = NULL; /* Dataset for this operation */
+    herr_t         ret_value = SUCCEED;
+
+    FUNC_ENTER_API(FAIL)
+    H5TRACE3("e", "ix*x", dset_id, cb, op_data);
+
+    /* Check arguments */
+    if (NULL == (vol_obj = (H5VL_object_t *)H5I_object_verify(dset_id, H5I_DATASET)))
+        HGOTO_ERROR(H5E_ARGS, H5E_BADTYPE, FAIL, "invalid dataset identifier")
+
+    /* Call private function to get the chunk info given the chunk's index */
+    if (H5VL_dataset_specific(vol_obj, H5VL_DATASET_CHUNK_ITER, H5P_DATASET_XFER_DEFAULT, H5_REQUEST_NULL, cb,
+                              op_data) < 0)
+        HGOTO_ERROR(H5E_DATASET, H5E_CANTGET, FAIL, "Can't iterate over chunks")
+
+done:
+    FUNC_LEAVE_API(ret_value)
+} /* end H5Dchunk_iter() */

src/H5Dchunk.c

@@ -245,6 +245,11 @@ typedef struct H5D_chunk_coll_info_t {
 } H5D_chunk_coll_info_t;
 #endif /* H5_HAVE_PARALLEL */
 
+typedef struct H5D_chunk_iter_cb_data_t {
+    H5D_chunk_iter_op_t cb;      /* User defined callback */
+    void *              op_data; /* User data for user defined callback */
+} H5D_chunk_iter_cb_data_t;
+
 /********************/
 /* Local Prototypes */
 /********************/
@@ -269,6 +274,7 @@ static herr_t H5D__chunk_dest(H5D_t *dset);
 static int H5D__get_num_chunks_cb(const H5D_chunk_rec_t *chunk_rec, void *_udata);
 static int H5D__get_chunk_info_cb(const H5D_chunk_rec_t *chunk_rec, void *_udata);
 static int H5D__get_chunk_info_by_coord_cb(const H5D_chunk_rec_t *chunk_rec, void *_udata);
+static int H5D__chunk_iter_cb(const H5D_chunk_rec_t *chunk_rec, void *udata);
 
 /* "Nonexistent" layout operation callback */
 static ssize_t H5D__nonexistent_readvv(const H5D_io_info_t *io_info, size_t chunk_max_nseq,
@@ -7451,3 +7457,95 @@ H5D__get_chunk_info_by_coord(const H5D_t *dset, const hsize_t *offset, unsigned
 done:
     FUNC_LEAVE_NOAPI_TAG(ret_value)
 } /* end H5D__get_chunk_info_by_coord() */
+
+/*-------------------------------------------------------------------------
+ * Function:    H5D__chunk_iter
+ *
+ * Purpose:     Iterate over all the chunks in the dataset with given callbak.
+ *
+ * Return:      Success:        Non-negative
+ *              Failure:        Negative
+ *
+ * Programmer:  Gaute Hope
+ *              August 2020
+ *
+ *-------------------------------------------------------------------------
+ */
+herr_t
+H5D__chunk_iter(const H5D_t *dset, H5D_chunk_iter_op_t cb, void *op_data)
+{
+    const H5O_layout_t *layout = NULL;       /* Dataset layout */
+    const H5D_rdcc_t *  rdcc   = NULL;       /* Raw data chunk cache */
+    H5D_rdcc_ent_t *    ent;                 /* Cache entry index */
+    H5D_chk_idx_info_t  idx_info;            /* Chunked index info */
+    herr_t              ret_value = SUCCEED; /* Return value */
+
+    FUNC_ENTER_PACKAGE_TAG(dset->oloc.addr)
+
+    /* Check args */
+    HDassert(dset);
+    HDassert(dset->shared);
+
+    /* Get dataset layout and raw data chunk cache */
+    layout = &(dset->shared->layout);
+    rdcc   = &(dset->shared->cache.chunk);
+    HDassert(layout);
+    HDassert(rdcc);
+    HDassert(H5D_CHUNKED == layout->type);
+
+    /* Search for cached chunks that haven't been written out */
+    for (ent = rdcc->head; ent; ent = ent->next)
+        /* Flush the chunk out to disk, to make certain the size is correct later */
+        if (H5D__chunk_flush_entry(dset, ent, FALSE) < 0)
+            HGOTO_ERROR(H5E_IO, H5E_WRITEERROR, FAIL, "cannot flush indexed storage buffer")
+
+    /* Compose chunked index info struct */
+    idx_info.f       = dset->oloc.file;
+    idx_info.pline   = &dset->shared->dcpl_cache.pline;
+    idx_info.layout  = &dset->shared->layout.u.chunk;
+    idx_info.storage = &dset->shared->layout.storage.u.chunk;
+
+    /* If the dataset is not written, return without errors */
+    if (H5F_addr_defined(idx_info.storage->idx_addr)) {
+        H5D_chunk_iter_cb_data_t data;
+        data.cb      = cb;
+        data.op_data = op_data;
+
+        /* Iterate over the allocated chunks calling the iterator callback */
+        if ((dset->shared->layout.storage.u.chunk.ops->iterate)(&idx_info, H5D__chunk_iter_cb, &data) < 0)
+            HGOTO_ERROR(H5E_DATASET, H5E_CANTGET, FAIL, "unable to iterate over chunks.")
+    } /* end if H5F_addr_defined */
+
+done:
+    FUNC_LEAVE_NOAPI_TAG(ret_value)
+} /* end H5D__chunk_iter() */
+
+/*-------------------------------------------------------------------------
+ * Function:    H5D__chunk_iter_cb
+ *
+ * Purpose:     Call the user-defined function with the chunk data. The iterator continues if
+ *              the user-defined function returns H5_ITER_CONT, and stops if H5_ITER_STOP is
+ *              returned.
+ *
+ * Return:      Success:    H5_ITER_CONT or H5_ITER_STOP
+ *              Failure:    Negative (H5_ITER_ERROR)
+ *
+ * Programmer:  Gaute Hope
+ *              August 2020
+ *
+ *-------------------------------------------------------------------------
+ */
+static int
+H5D__chunk_iter_cb(const H5D_chunk_rec_t *chunk_rec, void *udata)
+{
+    int ret_value = 0;
+
+    FUNC_ENTER_STATIC_NOERR
+
+    const H5D_chunk_iter_cb_data_t *data = (H5D_chunk_iter_cb_data_t *)udata;
+
+    ret_value = (data->cb)(chunk_rec->scaled, chunk_rec->filter_mask, chunk_rec->chunk_addr,
+                           chunk_rec->nbytes, data->op_data);
+
+    FUNC_LEAVE_NOAPI(ret_value)
+} /* end H5D__chunk_iter_cb */

src/H5Dpkg.h

@@ -565,6 +565,7 @@ H5_DLL herr_t  H5D__get_chunk_info(const H5D_t *dset, const H5S_t *space, hsize_
                                    unsigned *filter_mask, haddr_t *offset, hsize_t *size);
 H5_DLL herr_t  H5D__get_chunk_info_by_coord(const H5D_t *dset, const hsize_t *coord, unsigned *filter_mask,
                                             haddr_t *addr, hsize_t *size);
+H5_DLL herr_t  H5D__chunk_iter(const H5D_t *dset, H5D_chunk_iter_op_t cb, void *op_data);
 H5_DLL haddr_t H5D__get_offset(const H5D_t *dset);
 H5_DLL herr_t  H5D__vlen_get_buf_size(H5D_t *dset, hid_t type_id, hid_t space_id, hsize_t *size);
 H5_DLL herr_t  H5D__vlen_get_buf_size_gen(H5VL_object_t *vol_obj, hid_t type_id, hid_t space_id,

src/H5Dpublic.h

@@ -160,6 +160,14 @@ typedef herr_t (*H5D_scatter_func_t)(const void **src_buf /*out*/, size_t *src_b
 typedef herr_t (*H5D_gather_func_t)(const void *dst_buf, size_t dst_buf_bytes_used, void *op_data);
 //! <!-- [H5D_gather_func_t_snip] -->
 
+//! <!-- [H5D_chunk_iter_op_t_snip] -->
+/**
+ * Define the operator function pointer for H5Dchunk_iter()
+ */
+//! <!-- [H5D_chunk_iter_op_t_snip] -->
+typedef int (*H5D_chunk_iter_op_t)(const hsize_t *offset, uint32_t filter_mask, haddr_t addr, uint32_t nbytes,
+                                   void *op_data);
+
 /********************/
 /* Public Variables */
 /********************/
@@ -601,6 +609,51 @@ H5_DLL herr_t H5Dget_num_chunks(hid_t dset_id, hid_t fspace_id, hsize_t *nchunks
 H5_DLL herr_t H5Dget_chunk_info_by_coord(hid_t dset_id, const hsize_t *offset, unsigned *filter_mask,
                                          haddr_t *addr, hsize_t *size);
 
+/**
+ * --------------------------------------------------------------------------
+ * \ingroup H5D
+ *
+ * \brief Iterate over all chunks
+ *
+ * \dset_id
+ * \param[in]  cb       User callback function, called for every chunk.
+ * \param[in]  op_data  User-defined pointer to data required by op
+ *
+ * \return \herr_t
+ *
+ * \details H5Dget_chunk_iter iterates over all chunks in the dataset, calling the
+ *          user supplied callback with the details of the chunk and the supplied
+ *          \p op_data.
+ *
+ *          Callback information:
+ *            H5D_chunk_iter_op_t is defined as:
+ *
+ *              typedef int (*H5D_chunk_iter_op_t)(
+ *                const hsize_t *offset,
+ *                uint32_t filter_mask,
+ *                haddr_t addr,
+ *                uint32_t nbytes,
+ *                void *op_data);
+ *
+ *          H5D_chunk_iter_op_t parameters:
+ *              hsize_t *offset;        IN/OUT: Array of starting logical coordinates of chunk.
+ *              uint32_t filter_mask;   IN: Filter mask of chunk.
+ *              haddr_t addr;           IN: Offset in file of chunk data.
+ *              uint32_t nbytes;        IN: Size in number of bytes of chunk data in file.
+ *              void *op_data;          IN/OUT: Pointer to any user-defined data
+ *                                      associated with the operation.
+ *
+ *          The return values from an operator are:
+ *              Zero (H5_ITER_CONT) causes the iterator to continue, returning zero when all
+ *                  elements have been processed.
+ *              Positive (H5_ITER_STOP) causes the iterator to immediately return that positive
+ *                  value, indicating short-circuit success.
+ *              Negative (H5_ITER_ERROR) causes the iterator to immediately return that value,
+ *                  indicating failure.
+ *
+ */
+H5_DLL herr_t H5Dchunk_iter(hid_t dset_id, H5D_chunk_iter_op_t cb, void *op_data);
+
 /**
  * --------------------------------------------------------------------------
  * \ingroup H5D

src/H5VLconnector.h

@@ -81,7 +81,9 @@ typedef enum H5VL_dataset_get_t {
 typedef enum H5VL_dataset_specific_t {
     H5VL_DATASET_SET_EXTENT, /* H5Dset_extent                       */
     H5VL_DATASET_FLUSH,      /* H5Dflush                            */
-    H5VL_DATASET_REFRESH     /* H5Drefresh                          */
+    H5VL_DATASET_REFRESH,    /* H5Drefresh                          */
+    H5VL_DATASET_WAIT,       /* H5Dwait                             */
+    H5VL_DATASET_CHUNK_ITER  /* H5Dchunk_iter                       */
 } H5VL_dataset_specific_t;
 
 /* Typedef for VOL connector dataset optional VOL operations */

src/H5VLnative_dataset.c

@@ -346,6 +346,31 @@ H5VL__native_dataset_specific(void *obj, H5VL_dataset_specific_t specific_type,
             break;
         }
 
+        case H5VL_DATASET_WAIT: { /* H5Dwait */
+            /* The native VOL connector doesn't support asynchronous
+             *      operations, so this is a no-op.
+             */
+            break;
+        }
+
+        case H5VL_DATASET_CHUNK_ITER: { /* H5Dchunk_iter */
+            H5D_chunk_iter_op_t cb      = HDva_arg(arguments, H5D_chunk_iter_op_t);
+            void *              op_data = HDva_arg(arguments, void *);
+
+            HDassert(dset->shared);
+
+            /* Make sure the dataset is chunked */
+            if (H5D_CHUNKED != dset->shared->layout.type) {
+                HGOTO_ERROR(H5E_ARGS, H5E_BADTYPE, FAIL, "not a chunked dataset")
+            }
+
+            /* Call private function */
+            if (H5D__chunk_iter(dset, cb, op_data) < 0)
+                HGOTO_ERROR(H5E_DATASET, H5E_CANTGET, FAIL, "can't iterate over chunks")
+
+            break;
+        }
+
         default:
             HGOTO_ERROR(H5E_VOL, H5E_UNSUPPORTED, FAIL, "invalid specific operation")
     } /* end switch */