Error Reporting: GLib Reference Manual

Error Reporting

Error Reporting — a system for reporting errors

Functions

GError *	g_error_new ()
GError *	g_error_new_literal ()
GError *	g_error_new_valist ()
void	g_error_free ()
GError *	g_error_copy ()
gboolean	g_error_matches ()
void	g_set_error ()
void	g_set_error_literal ()
void	g_propagate_error ()
void	g_clear_error ()
void	g_prefix_error ()
void	g_propagate_prefixed_error ()

Types and Values

struct

GError

Includes

#include <glib.h>

Description

GLib provides a standard method of reporting errors from a called function to the calling code. (This is the same problem solved by exceptions in other languages.) It's important to understand that this method is both a data type (the GError struct) and a set of rules. If you use GError incorrectly, then your code will not properly interoperate with other code that uses GError, and users of your API will probably get confused. In most cases, using GError is preferred over numeric error codes, but there are situations where numeric error codes are useful for performance.

First and foremost: GError should only be used to report recoverable runtime errors, never to report programming errors. If the programmer has screwed up, then you should use g_warning(), g_return_if_fail(), g_assert(), g_error(), or some similar facility. (Incidentally, remember that the g_error() function should only be used for programming errors, it should not be used to print any error reportable via GError.)

Examples of recoverable runtime errors are "file not found" or "failed to parse input." Examples of programming errors are "NULL passed to strcmp()" or "attempted to free the same pointer twice." These two kinds of errors are fundamentally different: runtime errors should be handled or reported to the user, programming errors should be eliminated by fixing the bug in the program. This is why most functions in GLib and GTK+ do not use the GError facility.

Functions that can fail take a return location for a GError as their last argument. On error, a new GError instance will be allocated and returned to the caller via this argument. For example:

gboolean g_file_get_contents (const gchar  *filename,
                              gchar       **contents,
                              gsize        *length,
                              GError      **error);

If you pass a non-NULL value for the error argument, it should point to a location where an error can be placed. For example:

gchar *contents;
GError *err = NULL;

g_file_get_contents ("foo.txt", &contents, NULL, &err);
g_assert ((contents == NULL && err != NULL) || (contents != NULL && err == NULL));
if (err != NULL)
  {
    // Report error to user, and free error
    g_assert (contents == NULL);
    fprintf (stderr, "Unable to read file: %s\n", err->message);
    g_error_free (err);
  }
else
  {
    // Use file contents
    g_assert (contents != NULL);
  }

Note that err != NULL in this example is a reliable indicator of whether g_file_get_contents() failed. Additionally, g_file_get_contents() returns a boolean which indicates whether it was successful.

Because g_file_get_contents() returns FALSE on failure, if you are only interested in whether it failed and don't need to display an error message, you can pass NULL for the error argument:

if (g_file_get_contents ("foo.txt", &contents, NULL, NULL)) // ignore errors
  // no error occurred 
  ;
else
  // error
  ;

The GError object contains three fields: domain indicates the module the error-reporting function is located in, code indicates the specific error that occurred, and message is a user-readable error message with as many details as possible. Several functions are provided to deal with an error received from a called function: g_error_matches() returns TRUE if the error matches a given domain and code, g_propagate_error() copies an error into an error location (so the calling function will receive it), and g_clear_error() clears an error location by freeing the error and resetting the location to NULL. To display an error to the user, simply display the message , perhaps along with additional context known only to the calling function (the file being opened, or whatever - though in the g_file_get_contents() case, the message already contains a filename).

When implementing a function that can report errors, the basic tool is g_set_error(). Typically, if a fatal error occurs you want to g_set_error(), then return immediately. g_set_error() does nothing if the error location passed to it is NULL. Here's an example:

gint
foo_open_file (GError **error)
{
  gint fd;
  int saved_errno;

  fd = open ("file.txt", O_RDONLY);
  saved_errno = errno;

  if (fd < 0)
    {
      g_set_error (error,
                   FOO_ERROR,                 // error domain
                   FOO_ERROR_BLAH,            // error code
                   "Failed to open file: %s", // error message format string
                   g_strerror (saved_errno));
      return -1;
    }
  else
    return fd;
}

Things are somewhat more complicated if you yourself call another function that can report a GError. If the sub-function indicates fatal errors in some way other than reporting a GError, such as by returning TRUE on success, you can simply do the following:

gboolean
my_function_that_can_fail (GError **err)
{
  g_return_val_if_fail (err == NULL || *err == NULL, FALSE);

  if (!sub_function_that_can_fail (err))
    {
      // assert that error was set by the sub-function
      g_assert (err == NULL || *err != NULL);
      return FALSE;
    }

  // otherwise continue, no error occurred
  g_assert (err == NULL || *err == NULL);
}

If the sub-function does not indicate errors other than by reporting a GError (or if its return value does not reliably indicate errors) you need to create a temporary GError since the passed-in one may be NULL. g_propagate_error() is intended for use in this case.

gboolean
my_function_that_can_fail (GError **err)
{
  GError *tmp_error;

  g_return_val_if_fail (err == NULL || *err == NULL, FALSE);

  tmp_error = NULL;
  sub_function_that_can_fail (&tmp_error);

  if (tmp_error != NULL)
    {
      // store tmp_error in err, if err != NULL,
      // otherwise call g_error_free() on tmp_error
      g_propagate_error (err, tmp_error);
      return FALSE;
    }

  // otherwise continue, no error occurred
}

Error pileups are always a bug. For example, this code is incorrect:

gboolean
my_function_that_can_fail (GError **err)
{
  GError *tmp_error;

  g_return_val_if_fail (err == NULL || *err == NULL, FALSE);

  tmp_error = NULL;
  sub_function_that_can_fail (&tmp_error);
  other_function_that_can_fail (&tmp_error);

  if (tmp_error != NULL)
    {
      g_propagate_error (err, tmp_error);
      return FALSE;
    }
}

tmp_error should be checked immediately after sub_function_that_can_fail(), and either cleared or propagated upward. The rule is: after each error, you must either handle the error, or return it to the calling function.

Note that passing NULL for the error location is the equivalent of handling an error by always doing nothing about it. So the following code is fine, assuming errors in sub_function_that_can_fail() are not fatal to my_function_that_can_fail():

gboolean
my_function_that_can_fail (GError **err)
{
  GError *tmp_error;

  g_return_val_if_fail (err == NULL || *err == NULL, FALSE);

  sub_function_that_can_fail (NULL); // ignore errors

  tmp_error = NULL;
  other_function_that_can_fail (&tmp_error);

  if (tmp_error != NULL)
    {
      g_propagate_error (err, tmp_error);
      return FALSE;
    }
}

Note that passing NULL for the error location ignores errors; it's equivalent to try { sub_function_that_can_fail(); } catch (...) {} in C++. It does not mean to leave errors unhandled; it means to handle them by doing nothing.

Error domains and codes are conventionally named as follows:

The error domain is called <NAMESPACE>_<MODULE>_ERROR, for example G_SPAWN_ERROR or G_THREAD_ERROR:

#define G_SPAWN_ERROR g_spawn_error_quark ()

GQuark
g_spawn_error_quark (void)
{
  return g_quark_from_static_string ("g-spawn-error-quark");
}

The quark function for the error domain is called <namespace>_<module>_error_quark, for example g_spawn_error_quark() or g_thread_error_quark().
The error codes are in an enumeration called <Namespace><Module>Error; for example, GThreadError or GSpawnError.
Members of the error code enumeration are called <NAMESPACE>_<MODULE>_ERROR_<CODE>, for example G_SPAWN_ERROR_FORK or G_THREAD_ERROR_AGAIN.
If there's a "generic" or "unknown" error code for unrecoverable errors it doesn't make sense to distinguish with specific codes, it should be called <NAMESPACE>_<MODULE>_ERROR_FAILED, for example G_SPAWN_ERROR_FAILED. In the case of error code enumerations that may be extended in future releases, you should generally not handle this error code explicitly, but should instead treat any unrecognized error code as equivalent to FAILED.

Comparison of GError and traditional error handling

GError has several advantages over traditional numeric error codes: importantly, tools like gobject-introspection understand GErrors and convert them to exceptions in bindings; the message includes more information than just a code; and use of a domain helps prevent misinterpretation of error codes.

GError has disadvantages though: it requires a memory allocation, and formatting the error message string has a performance overhead. This makes it unsuitable for use in retry loops where errors are a common case, rather than being unusual. For example, using G_IO_ERROR_WOULD_BLOCK means hitting these overheads in the normal control flow. String formatting overhead can be eliminated by using g_set_error_literal() in some cases.

These performance issues can be compounded if a function wraps the GErrors returned by the functions it calls: this multiplies the number of allocations and string formatting operations. This can be partially mitigated by using g_prefix_error().

Rules for use of GError

Summary of rules for use of GError:

Do not report programming errors via GError.
The last argument of a function that returns an error should be a location where a GError can be placed (i.e. "GError** error"). If GError is used with varargs, the GError** should be the last argument before the "...".
The caller may pass NULL for the GError** if they are not interested in details of the exact error that occurred.
If NULL is passed for the GError** argument, then errors should not be returned to the caller, but your function should still abort and return if an error occurs. That is, control flow should not be affected by whether the caller wants to get a GError.
If a GError is reported, then your function by definition had a fatal failure and did not complete whatever it was supposed to do. If the failure was not fatal, then you handled it and you should not report it. If it was fatal, then you must report it and discontinue whatever you were doing immediately.
If a GError is reported, out parameters are not guaranteed to be set to any defined value.
A GError* must be initialized to NULL before passing its address to a function that can report errors.
"Piling up" errors is always a bug. That is, if you assign a new GError to a GError* that is non-NULL, thus overwriting the previous error, it indicates that you should have aborted the operation instead of continuing. If you were able to continue, you should have cleared the previous error with g_clear_error(). g_set_error() will complain if you pile up errors.
By convention, if you return a boolean value indicating success then TRUE means success and FALSE means failure. Avoid creating functions which have a boolean return value and a GError parameter, but where the boolean does something other than signal whether the GError is set. Among other problems, it requires C callers to allocate a temporary error. Instead, provide a "gboolean *" out parameter. There are functions in GLib itself such as g_key_file_has_key() that are deprecated because of this. If FALSE is returned, the error must be set to a non-NULL value. One exception to this is that in situations that are already considered to be undefined behaviour (such as when a g_return_val_if_fail() check fails), the error need not be set. Instead of checking separately whether the error is set, callers should ensure that they do not provoke undefined behaviour, then assume that the error will be set on failure.
A NULL return value is also frequently used to mean that an error occurred. You should make clear in your documentation whether NULL is a valid return value in non-error cases; if NULL is a valid value, then users must check whether an error was returned to see if the function succeeded.
When implementing a function that can report errors, you may want to add a check at the top of your function that the error return location is either NULL or contains a NULL error (e.g. g_return_if_fail (error == NULL || *error == NULL);).

Functions

g_error_new ()

GError *
g_error_new (GQuark domain,
             gint code,
             const gchar *format,
             ...);

Creates a new GError with the given domain and code , and a message formatted with format .

Parameters

domain	error domain
code	error code
format	printf()-style format for error message
...	parameters for message format

Returns

a new GError

g_error_new_literal ()

GError *
g_error_new_literal (GQuark domain,
                     gint code,
                     const gchar *message);

Creates a new GError; unlike g_error_new(), message is not a printf()-style format string. Use this function if message contains text you don't have control over, that could include printf() escape sequences.

Parameters

domain	error domain
code	error code
message	error message

Returns

a new GError

g_error_new_valist ()

GError *
g_error_new_valist (GQuark domain,
                    gint code,
                    const gchar *format,
                    va_list args);

Creates a new GError with the given domain and code , and a message formatted with format .

Parameters

domain	error domain
code	error code
format	printf()-style format for error message
args	va_list of parameters for the message format

Returns

a new GError

Since: 2.22

g_error_free ()

void
g_error_free (GError *error);

Frees a GError and associated resources.

Parameters

error

a GError

g_error_copy ()

GError *
g_error_copy (const GError *error);

Makes a copy of error .

Parameters

error

a GError

Returns

a new GError

g_error_matches ()

gboolean
g_error_matches (const GError *error,
                 GQuark domain,
                 gint code);

Returns TRUE if error matches domain and code , FALSE otherwise. In particular, when error is NULL, FALSE will be returned.

If domain contains a FAILED (or otherwise generic) error code, you should generally not check for it explicitly, but should instead treat any not-explicitly-recognized error code as being equivalent to the FAILED code. This way, if the domain is extended in the future to provide a more specific error code for a certain case, your code will still work.

Parameters

error	a GError.	[nullable]
domain	an error domain
code	an error code

Returns

whether error has domain and code

g_set_error ()

void
g_set_error (GError **err,
             GQuark domain,
             gint code,
             const gchar *format,
             ...);

Does nothing if err is NULL; if err is non-NULL, then *err must be NULL. A new GError is created and assigned to *err .

Parameters

err	a return location for a GError.	[out callee-allocates][optional]
domain	error domain
code	error code
format	printf()-style format
...	args for `format`

g_set_error_literal ()

void
g_set_error_literal (GError **err,
                     GQuark domain,
                     gint code,
                     const gchar *message);

Does nothing if err is NULL; if err is non-NULL, then *err must be NULL. A new GError is created and assigned to *err . Unlike g_set_error(), message is not a printf()-style format string. Use this function if message contains text you don't have control over, that could include printf() escape sequences.

Parameters

err	a return location for a GError.	[out callee-allocates][optional]
domain	error domain
code	error code
message	error message

Since: 2.18

g_propagate_error ()

void
g_propagate_error (GError **dest,
                   GError *src);

If dest is NULL, free src ; otherwise, moves src into *dest . The error variable dest points to must be NULL.

src must be non-NULL.

Note that src is no longer valid after this call. If you want to keep using the same GError*, you need to set it to NULL after calling this function on it.

Parameters

dest	error return location.	[out callee-allocates][optional][nullable]
src	error to move into the return location.	[transfer full]

g_clear_error ()

void
g_clear_error (GError **err);

If err or *err is NULL, does nothing. Otherwise, calls g_error_free() on *err and sets *err to NULL.

Parameters

err

a GError return location

g_prefix_error ()

void
g_prefix_error (GError **err,
                const gchar *format,
                ...);

Formats a string according to format and prefix it to an existing error message. If err is NULL (ie: no error variable) then do nothing.

If *err is NULL (ie: an error variable is present but there is no error condition) then also do nothing. Whether or not it makes sense to take advantage of this feature is up to you.

Parameters

err	a return location for a GError.	[inout][optional][nullable]
format	printf()-style format string
...	arguments to `format`

Since: 2.16

g_propagate_prefixed_error ()

void
g_propagate_prefixed_error (GError **dest,
                            GError *src,
                            const gchar *format,
                            ...);

If dest is NULL, free src ; otherwise, moves src into *dest . *dest must be NULL. After the move, add a prefix as with g_prefix_error().

Parameters

dest	error return location
src	error to move into the return location
format	printf()-style format string
...	arguments to `format`

Since: 2.16

Types and Values

struct GError

struct GError {
  GQuark       domain;
  gint         code;
  gchar       *message;
};

The GError structure contains information about an error that has occurred.

Members

GQuark `domain`;	error domain, e.g. G_FILE_ERROR
gint `code`;	error code, e.g. `G_FILE_ERROR_NOENT`
gchar *`message`;	human-readable informative error message