Chapter 9
HDF5 Error Handling

9.1. Introduction

The HDF5 Library provides an error reporting mechanism for both the library itself and for user application programs. It can trace errors through function stack and error information like file name, function name, line number, and error description.

Section 2 of this chapter discusses the HDF5 error handling programming model.

Section 3 presents summaries of HDF5’s error handling functions.

Section 4 discusses the basic error concepts such as error stack, error record, and error message and describes the related API functions. These concepts and functions are sufficient for application programs to trace errors inside the HDF5 Library.

Section 5 talks about the advanced concepts of error class and error stack handle and talks about the related functions. With these concepts and functions, an application library or program using the HDF5 Library can have its own error report blended with HDF5’s error report.

Starting with Release 1.8, we have a new set of Error Handling API functions. For the purpose of backward compatibility with version 1.6 and before, we still keep the old API functions, H5Epush, H5Eprint, H5Ewalk, H5Eclear, H5Eget_auto, H5Eset_auto. These functions do not have the error stack as parameter. The library allows them to operate on the default error stack. Users do not have to change their code to catch up with the new Error API but are encouraged to do so.

The old API is similar to functionality discussed in Section 4. The functionality discussed in Section 5, the ability of allowing applications to add their own error records, is the library new design for the Error API.

9.2. Programming Model

This section is under construction.

9.3. Error Handling (H5E) Function Summaries

Functions that can be used to handle errors (H5E functions) are listed below.


Function Listing 1. Error handling functions (H5E)
C Function
Fortran Function
Purpose
H5Eauto_is_v2
(none)
Determines the type of error stack.
H5Eclear
h5eclear_f
Clears the error stack for the current thread. The C function is a macro: see “API Compatibility Macros in HDF5.”
H5Eclear_stack
(none)
Clears the error stack for the current thread.
H5Eclose_msg
(none)
Closes an error message identifier.
H5Eclose_stack
(none)
Closes object handle for error stack.
H5Ecreate_msg
(none)
Add major error message to an error class.
H5Eget_auto
h5eget_auto_f
Returns the current settings for the automatic error stack traversal function and its data. The C function is a macro: see “API Compatibility Macros in HDF5.”
H5Eget_class_name
(none)
Retrieves error class name.
H5Eget_current_stack
(none)
Registers the current error stack.
H5Eget_msg
(none)
Retrieves an error message.
H5Eget_num
(none)
Retrieves the number of error messages in an error stack.
H5Epop
(none)
Deletes specified number of error messages from the error stack.
H5Eprint
h5eprint_f
Prints the error stack in a default manner. The C function is a macro: see “API Compatibility Macros in HDF5.”
H5Epush
(none)
Pushes new error record onto error stack. The C function is a macro: see “API Compatibility Macros in HDF5.”
H5Eregister_class
(none)
Registers a client library or application program to the HDF5 error API.
H5Eset_auto
h5eset_auto_f
Turns automatic error printing on or off. The C function is a macro: see “API Compatibility Macros in HDF5.”
H5Eset_current_stack
(none)
Replaces the current error stack.
H5Eunregister_class
(none)
Removes an error class.
H5Ewalk
(none)
Walks the error stack for the current thread, calling a specified function. The C function is a macro: see “API Compatibility Macros in HDF5.”

9.4. Basic Error Handling Operations

9.4.1. Introduction

Let us first try to understand the error stack. An error stack is a collection of error records. Error records can be pushed onto or popped off the error stack. By default, when an error occurs deep within the HDF5 Library, an error record is pushed onto an error stack and that function returns a failure indication. Its caller detects the failure, pushes another record onto the stack, and returns a failure indication. This continues until the API function called by the application returns a failure indication. The next API function being called will reset the error stack. All HDF5 Library error records belong to the same error class (explained in Section 5).

9.4.2. Error Stack and Error Message

In normal circumstances, an error causes the stack to be printed on the standard error stream automatically. This automatic error stack is the library’s default stack. For all the functions in this section, whenever an error stack ID is needed as a parameter, H5E_DEFAULT can be used to indicate the library’s default stack. The first error record of the error stack, number #000, is produced by the API function itself and is usually sufficient to indicate to the application what went wrong.

Example: An Error Report

If an application calls H5Tclose on a predefined datatype, then the message in the example below is printed on the standard error stream. This is a simple error that has only one component, the API function; other errors may have many components.


HDF5-DIAG: Error detected in HDF5 (1.6.4) thread 0.
  #000: H5T.c line 462 in H5Tclose(): predefined datatype
    major: Function argument
    minor: Bad value
Example 1. An error report

In the example above, we can see that an error record has a major message and a minor message. A major message generally indicates where the error happens. The location can be a dataset or a dataspace, for example. A minor message explains further details of the error. An example is “unable to open file”. Another specific detail about the error can be found at the end of the first line of each error record. This error description is usually added by the library designer to tell what exactly goes wrong. In the example above, the “predefined datatype” is an error description.

9.4.3. Print and Clear an Error Stack

Besides the automatic error report, the error stack can also be printed and cleared by the functions H5Eprint() and H5Eclear_stack(). If an application wishes to make explicit calls to H5Eprint() to print the error stack, the automatic printing should be turned off to prevent error messages from being displayed twice (see H5Eset_auto() below).

To print an error stack

herr_t H5Eprint(hid_t error_stack, FILE * stream)

This function prints the error stack specified by error_stack on the specified stream, stream. If the error stack is empty, a one-line message will be printed. The following is an example of such a message. This message would be generated if the error was in the HDF5 Library.

HDF5-DIAG: Error detected in HDF5 Library version: 1.5.62 thread 0.

To clear an error stack

herr_t H5Eclear_stack(hid_t error_stack)

The H5Eclear_stack function shown above clears the error stack specified by error_stack. H5E_DEFAULT can be passed in to clear the current error stack. The current stack is also cleared whenever an API function is called; there are certain exceptions to this rule such as H5Eprint().

9.4.4. Mute Error Stack

Sometimes an application calls a function for the sake of its return value, fully expecting the function to fail; sometimes the application wants to call H5Eprint() explicitly. In these situations, it would be misleading if an error message were still automatically printed. Using the H5Eset_auto() function can control the automatic printing of error messages.

To enable or disable automatic printing of errors

herr_t H5Eset_auto(hid_t error_stack, H5E_auto_t func, void *client_data)

The H5Eset_auto function can be used to turns on or off the automatic printing of errors for the error stack specified by error_stack. When turned on (non-null func pointer), any API function which returns an error indication will first call func, passing it client_data as an argument. When the library is first initialized the auto printing function is set to H5Eprint() (cast appropriately) and client_data is the standard error stream pointer, stderr.

To see the current settings

herr_t H5Eget_auto(hid_t error_stack, H5E_auto_t * func, void **client_data )

The function above returns the current settings for the automatic error stack traversal function, func, and its data, client_data. If either or both of the arguments are null, then the value is not returned.

Example: Error Control

An application can temporarily turn off error messages while “probing” a function. See the example below.


/* Save old error handler */
herr_t (*old_func)(void*);
void *old_client_data;

H5Eget_auto(error_stack, &old_func, &old_client_data);

/* Turn off error handling */
H5Eset_auto(error_stack, NULL, NULL);

/* Probe. Likely to fail, but that’s okay */
status = H5Fopen (......);

/* Restore previous error handler */
H5Eset_auto(error_stack, old_func, old_client_data);
Example 2. Turn off error messages while probing a function

Or automatic printing can be disabled altogether and error messages can be explicitly printed.


/* Turn off error handling permanently */
H5Eset_auto(error_stack, NULL, NULL);
/* If failure, print error message */
if (H5Fopen (....)<0) {
    H5Eprint(H5E_DEFAULT, stderr);
    exit (1);
}
Example 3. Disable automatic printing and explicitly print error messages

9.4.5. Customized Printing of an Error Stack

Applications are allowed to define an automatic error traversal function other than the default H5Eprint(). For instance, one can define a function that prints a simple, one-line error message to the standard error stream and then exits. The first example below defines a such a function. The second example below installs the function as the error handler.


herr_t
my_hdf5_error_handler(void *unused)
{
   fprintf (stderr, “An HDF5 error was detected. Bye.\n”);
   exit (1);
}
Example 4. Defining a function to print a simple error message



H5Eset_auto(H5E_DEFAULT, my_hdf5_error_handler, NULL);
Example 5. The user-defined error handler

9.4.6. Walk through the Error Stack

The H5Eprint() function is actually just a wrapper around the more complex H5Ewalk() function which traverses an error stack and calls a user-defined function for each member of the stack. The example below shows how H5Ewalk is used.

herr_t H5Ewalk(hid_t err_stack, H5E_direction_t direction, H5E_walk_t func, void *client_data)

The error stack err_stack is traversed and func is called for each member of the stack. Its arguments are an integer sequence number beginning at zero (regardless of direction) and the client_data pointer. If direction is H5E_WALK_UPWARD, then traversal begins at the inner-most function that detected the error and concludes with the API function. Use H5E_WALK_DOWNWARD for the opposite order.

9.4.7. Traverse an Error Stack with a Callback Function

An error stack traversal callback function takes three arguments: n is a sequence number beginning at zero for each traversal, eptr is a pointer to an error stack member, and client_data is the same pointer used in the example above passed to H5Ewalk(). See the example below.

typedef herr_t (*H5E_walk_t)(unsigned n, H5E_error2_t *eptr, void *client_data)

The H5E_error2_t structure is shown below.

typedef struct {
    hid_t	cls_id;
    hid_t       maj_num;
    hid_t       min_num;
    unsigned    line;
    const char  *func_name;
    const char  *file_name;
    const char  *desc;
} H5E_error2_t;

The maj_num and min_num are major and minor error IDs, func_name is the name of the function where the error was detected, file_name and line locate the error within the HDF5 Library source code, and desc points to a description of the error.

Example: Callback Function

The following example shows a user-defined callback function.


#define MSG_SIZE       64

herr_t
custom_print_cb(unsigned n, const H5E_error2_t *err_desc, void* client_data)
{
    FILE		*stream  = (FILE *)client_data;
    char                maj[MSG_SIZE];
    char                min[MSG_SIZE];
    char                cls[MSG_SIZE];
    const int		indent = 4;

    /* Get descriptions for the major and minor error numbers */
    if(H5Eget_class_name(err_desc->cls_id, cls, MSG_SIZE)<0)
        TEST_ERROR;

    if(H5Eget_msg(err_desc->maj_num, NULL, maj, MSG_SIZE)<0)
        TEST_ERROR;

    if(H5Eget_msg(err_desc->min_num, NULL, min, MSG_SIZE)<0)
        TEST_ERROR;

    fprintf (stream, “%*serror #%03d: %s in %s(): line %u\n”,
	     indent, “”, n, err_desc->file_name,
	     err_desc->func_name, err_desc->line);
    fprintf (stream, “%*sclass: %s\n”, indent*2, “”, cls);
    fprintf (stream, “%*smajor: %s\n”, indent*2, “”, maj);
    fprintf (stream, “%*sminor: %s\n”, indent*2, “”, min);

    return 0;

  error:
    return -1;
}
Example 6. A user-defined callback function

9.5. Advanced Error Handling Operations

9.5.1. Introduction

Section 4 discusses the basic error handling operations of the library. In that section, all the error records on the error stack are from the library itself. In this section, we are going to introduce the operations that allow an application program to push its own error records onto the error stack once it declares an error class of its own through the HDF5 Error API.

Example: An Error Report

An error report shows both the library’s error record and the application’s error records. See the example below.


Error Test-DIAG: Error detected in Error Program (1.0) thread 8192:
  #000: ../../hdf5/test/error_test.c line 468 in main(): Error test failed
    major: Error in test
    minor: Error in subroutine
  #001: ../../hdf5/test/error_test.c line 150 in test_error(): H5Dwrite failed 
      as supposed to
    major: Error in IO
    minor: Error in H5Dwrite
HDF5-DIAG: Error detected in HDF5 (1.7.5) thread 8192:
  #002: ../../hdf5/src/H5Dio.c line 420 in H5Dwrite(): not a dataset
    major: Invalid arguments to routine
    minor: Inappropriate type 
    
Example 7. An error report

In the line above error record #002 in the example above, the starting phrase is HDF5. This is the error class name of the HDF5 Library. All of the library’s error messages (major and minor) are in this default error class. The Error Test in the beginning of the line above error record #000 is the name of the application’s error class. The first two error records, #000 and #001, are from application’s error class.

By definition, an error class is a group of major and minor error messages for a library (the HDF5 Library or an application library built on top of the HDF5 Library) or an application program. The error class can be registered for a library or program through the HDF5 Error API. Major and minor messages can be defined in an error class. An application will have object handles for the error class and for major and minor messages for further operation. See the example below.


#define MSG_SIZE       64

herr_t
custom_print_cb(unsigned n, const H5E_error2_t *err_desc, void* client_data)
{
    FILE		*stream  = (FILE *)client_data;
    char                maj[MSG_SIZE];
    char                min[MSG_SIZE];
    char                cls[MSG_SIZE];
    const int		indent = 4;

    /* Get descriptions for the major and minor error numbers */
    if(H5Eget_class_name(err_desc->cls_id, cls, MSG_SIZE)<0)
        TEST_ERROR;

    if(H5Eget_msg(err_desc->maj_num, NULL, maj, MSG_SIZE)<0)
        TEST_ERROR;

    if(H5Eget_msg(err_desc->min_num, NULL, min, MSG_SIZE)<0)
        TEST_ERROR;

    fprintf (stream, “%*serror #%03d: %s in %s(): line %u\n”,
	     indent, “”, n, err_desc->file_name,
	     err_desc->func_name, err_desc->line);
    fprintf (stream, “%*sclass: %s\n”, indent*2, “”, cls);
    fprintf (stream, “%*smajor: %s\n”, indent*2, “”, maj);
    fprintf (stream, “%*sminor: %s\n”, indent*2, “”, min);

    return 0;

  error:
    return -1;
}
Example 8. Defining an error class

9.5.2. More Error API Functions

The Error API has functions that can be used to register or unregister an error class, to create or close error messages, and to query an error class or error message. These functions are illustrated below.

To register an error class

hid_t H5Eregister_class(const char* cls_name, const char* lib_name, const char* version)

This function registers an error class with the HDF5 Library so that the application library or program can report errors together with the HDF5 Library.

To add an error message to an error class

hid_t H5Ecreate_msg(hid_t class, H5E_type_t msg_type, const char* mesg)

This function adds an error message to an error class defined by an application library or program. The error message can be either major or minor which is indicated by parameter msg_type.

To get the name of an error class

ssize_t H5Eget_class_name(hid_t class_id, char* name, size_t size)

This function retrieves the name of the error class specified by the class ID.

To retrieve an error message

ssize_t H5Eget_msg(hid_t mesg_id, H5E_type_t* mesg_type, char* mesg, size_t size)

This function retrieves the error message including its length and type.

To close an error message

herr_t H5Eclose_msg(hid_t mesg_id)

This function closes an error message.

To remove an error class

herr_t H5Eunregister_class(hid_t class_id)

This function removes an error class from the Error API.


Example: Error Class and its Message

The example below shows how an application creates an error class and error messages.


/* Create an error class */
class_id = H5Eregister_class(ERR_CLS_NAME, PROG_NAME, PROG_VERS);

/* Retrieve class name */
H5Eget_class_name(class_id, cls_name, cls_size); 

/* Create a major error message in the class */
maj_id = H5Ecreate_msg(class_id, H5E_MAJOR, “... ...”);

/* Create a minor error message in the class */
min_id = H5Ecreate_msg(class_id, H5E_MINOR, “... ...”);
Example 9. Create an error class and error messages

The example below shows how an application closes error messages and unregisters the error class.


H5Eclose_msg(maj_id);
H5Eclose_msg(min_id);
H5Eunregister_class(class_id);
Example 10. Closing error messages and unregistering the error class

9.5.3. Pushing an Application Error Message onto Error Stack

An application can push error records onto or pop error records off of the error stack just as the library does internally. An error stack can be registered, and an object handle can be returned to the application so that the application can manipulate a registered error stack.

To register the current stack

hid_t H5Eget_current_stack(void)

This function registers the current error stack, returns an object handle, and clears the current error stack. An empty error stack will also be assigned an ID.

To replace the current error stack with another

herr_t H5Eset_current_stack(hid_t error_stack)

This function replaces the current error stack with another error stack specified by error_stack and clears the current error stack. The object handle error_stack is closed after this function call.

To push a new error record to the error stack

herr_t H5Epush(hid_t error_stack, const char* file, const char* func, unsigned line, hid_t cls_id, hid_t major_id, hid_t minor_id, const char* desc, ... )

This function pushes a new error record onto the error stack for the current thread.

To delete some error messages

herr_t H5Epop(hid_t error_stack, size_t count)

This function deletes some error messages from the error stack.

To retrieve the number of error records

int H5Eget_num(hid_t error_stack)

This function retrieves the number of error records from an error stack.

To clear the error stack

herr_t H5Eclear_stack(hid_t error_stack)

This function clears the error stack.

To close the object handle for an error stack

herr_t H5Eclose_stack(hid_t error_stack)

This function closes the object handle for an error stack and releases its resources.


Example: Working with an Error Stack

The example below shows how an application pushes an error record onto the default error stack.


/* Make call to HDF5 I/O routine */
if((dset_id=H5Dopen(file_id, dset_name, access_plist))<0)
{
	/* Push client error onto error stack */
	H5Epush(H5E_DEFAULT,__FILE__,FUNC,__LINE__,cls_id,CLIENT_ERR_MAJ_IO,
		CLIENT_ERR_MINOR_OPEN,“H5Dopen failed”);

	/* Indicate error occurred in function */
	return(0);
}
Example 11. Pushing an error message to an error stack

The example below shows how an application registers the current error stack and creates an object handle to avoid another HDF5 function from clearing the error stack.


if(H5Dwrite(dset_id, mem_type_id, mem_space_id, file_space_id, dset_xfer_plist_id, buf)<0)
{
	/* Push client error onto error stack */
	H5Epush(H5E_DEFAULT,__FILE__,FUNC,__LINE__,cls_id,CLIENT_ERR_MAJ_IO,
		    CLIENT_ERR_MINOR_HDF5,“H5Dwrite failed”);

	/* Preserve the error stack by assigning an object handle to it */
    error_stack = H5Eget_current_stack();
                
	/* Close dataset */
	H5Dclose(dset_id);
               
    /* Replace the current error stack with the preserved one */
    H5Eset_current_stack(error_stack);

	Return(0);
} 
Example 12. Registering the error stack