MPI-4: Initial error handler

ompi/communicator/comm_init.c

@@ -3,7 +3,7 @@
  * Copyright (c) 2004-2005 The Trustees of Indiana University and Indiana
  *                         University Research and Technology
  *                         Corporation.  All rights reserved.
- * Copyright (c) 2004-2017 The University of Tennessee and The University
+ * Copyright (c) 2004-2020 The University of Tennessee and The University
  *                         of Tennessee Research Foundation.  All rights
  *                         reserved.
  * Copyright (c) 2004-2005 High Performance Computing Center Stuttgart,
@@ -135,8 +135,8 @@ int ompi_comm_init(void)
     ompi_mpi_comm_world.comm.c_remote_group = group;
     OBJ_RETAIN(ompi_mpi_comm_world.comm.c_remote_group);
     ompi_mpi_comm_world.comm.c_cube_dim     = opal_cube_dim((int)size);
-    ompi_mpi_comm_world.comm.error_handler  = &ompi_mpi_errors_are_fatal.eh;
-    OBJ_RETAIN( &ompi_mpi_errors_are_fatal.eh );
+    ompi_mpi_comm_world.comm.error_handler  = ompi_initial_error_handler_eh;
+    OBJ_RETAIN( ompi_mpi_comm_world.comm.error_handler );
     OMPI_COMM_SET_PML_ADDED(&ompi_mpi_comm_world.comm);
     opal_pointer_array_set_item (&ompi_mpi_communicators, 0, &ompi_mpi_comm_world);
 
@@ -188,8 +188,8 @@ int ompi_comm_init(void)
     ompi_mpi_comm_self.comm.c_local_group  = group;
     ompi_mpi_comm_self.comm.c_remote_group = group;
     OBJ_RETAIN(ompi_mpi_comm_self.comm.c_remote_group);
-    ompi_mpi_comm_self.comm.error_handler  = &ompi_mpi_errors_are_fatal.eh;
-    OBJ_RETAIN( &ompi_mpi_errors_are_fatal.eh );
+    ompi_mpi_comm_self.comm.error_handler  = ompi_initial_error_handler_eh;
+    OBJ_RETAIN( ompi_mpi_comm_self.comm.error_handler );
     OMPI_COMM_SET_PML_ADDED(&ompi_mpi_comm_self.comm);
     opal_pointer_array_set_item (&ompi_mpi_communicators, 1, &ompi_mpi_comm_self);
 
@@ -214,8 +214,10 @@ int ompi_comm_init(void)
     ompi_mpi_comm_null.comm.c_contextid    = 2;
     ompi_mpi_comm_null.comm.c_my_rank      = MPI_PROC_NULL;
 
+    /* unlike world, self, and parent, comm_null does not inherit the initial error
+     * handler */
     ompi_mpi_comm_null.comm.error_handler  = &ompi_mpi_errors_are_fatal.eh;
-    OBJ_RETAIN( &ompi_mpi_errors_are_fatal.eh );
+    OBJ_RETAIN( ompi_mpi_comm_null.comm.error_handler );
     opal_pointer_array_set_item (&ompi_mpi_communicators, 2, &ompi_mpi_comm_null);
 
     opal_string_copy(ompi_mpi_comm_null.comm.c_name, "MPI_COMM_NULL",
@@ -228,6 +230,8 @@ int ompi_comm_init(void)
     OBJ_RETAIN(&ompi_mpi_comm_null);
     OBJ_RETAIN(&ompi_mpi_group_null.group);
     OBJ_RETAIN(&ompi_mpi_errors_are_fatal.eh);
+    /* During dyn_init, the comm_parent error handler will be set to the same
+     * as comm_world (thus, the initial error handler). */
 
     /* initialize communicator requests (for ompi_comm_idup) */
     ompi_comm_request_init ();

ompi/dpm/dpm.c

@@ -782,6 +782,7 @@ int ompi_dpm_spawn(int count, const char *array_of_commands[],
     int flag=0;
     char cwd[OPAL_PATH_MAX];
     char host[OPAL_MAX_INFO_VAL];  /*** should define OMPI_HOST_MAX ***/
+    char init_errh[OPAL_MAX_INFO_VAL];
     char prefix[OPAL_MAX_INFO_VAL];
     char stdin_target[OPAL_MAX_INFO_VAL];
     char params[OPAL_MAX_INFO_VAL];
@@ -814,6 +815,7 @@ int ompi_dpm_spawn(int count, const char *array_of_commands[],
        - "file": filename, where additional information is provided.
        - "soft": see page 92 of MPI-2.
        - "host": desired host where to spawn the processes
+       - "mpi_initial_errhandler": the error handler attached to predefined communicators.
       Non-standard keys:
        - "hostfile": hostfile containing hosts where procs are
        to be spawned
@@ -968,6 +970,15 @@ int ompi_dpm_spawn(int count, const char *array_of_commands[],
             }
 #endif
 
+            /* check for 'mpi_initial_errhandler' */
+            ompi_info_get (array_of_info[i], "mpi_initial_errhandler", sizeof(init_errh) - 1, init_errh, &flag);
+            if ( flag ) {
+                /* this is set as an environment because it must be available
+                 * before pmix_init */
+                opal_setenv("OMPI_MCA_mpi_initial_errhandler", init_errh, true, &app->env);
+                continue;
+            }
+
             /* 'path', 'arch', 'file', 'soft'  -- to be implemented */
 
             /* non-standard keys

ompi/errhandler/errcode.c

@@ -131,8 +131,17 @@ do {                                                                      \
     opal_pointer_array_set_item(&ompi_mpi_errcodes, (ERRCODE), &(VAR));   \
 } while (0)
 
+static opal_mutex_t errcode_init_lock = OPAL_MUTEX_STATIC_INIT;
+
 int ompi_mpi_errcode_init (void)
 {
+    opal_mutex_lock(&errcode_init_lock);
+    if ( 0 != ompi_mpi_errcode_lastpredefined ) {
+        /* Already initialized (presumably by an API call before MPI_init */
+        opal_mutex_unlock(&errcode_init_lock);
+        return OMPI_SUCCESS;
+    }
+
     /* Initialize the pointer array, which will hold the references to
        the error objects */
     OBJ_CONSTRUCT(&ompi_mpi_errcodes, opal_pointer_array_t);
@@ -223,6 +232,7 @@ int ompi_mpi_errcode_init (void)
        MPI_ERR_LASTCODE.  So just start it as == MPI_ERR_LASTCODE. */
     ompi_mpi_errcode_lastused = MPI_ERR_LASTCODE;
     ompi_mpi_errcode_lastpredefined = MPI_ERR_LASTCODE;
+    opal_mutex_unlock(&errcode_init_lock);
     return OMPI_SUCCESS;
 }
 
@@ -231,6 +241,7 @@ int ompi_mpi_errcode_finalize(void)
     int i;
     ompi_mpi_errcode_t *errc;
 
+    opal_mutex_lock(&errcode_init_lock);
     for (i=ompi_mpi_errcode_lastpredefined+1; i<=ompi_mpi_errcode_lastused; i++) {
         /*
          * there are some user defined error-codes, which
@@ -317,6 +328,8 @@ int ompi_mpi_errcode_finalize(void)
     OBJ_DESTRUCT(&ompi_t_err_invalid_name);
 
     OBJ_DESTRUCT(&ompi_mpi_errcodes);
+    ompi_mpi_errcode_lastpredefined = 0;
+    opal_mutex_unlock(&errcode_init_lock);
     return OMPI_SUCCESS;
 }
 

ompi/errhandler/errcode.h

@@ -3,7 +3,7 @@
  * Copyright (c) 2004-2005 The Trustees of Indiana University and Indiana
  *                         University Research and Technology
  *                         Corporation.  All rights reserved.
- * Copyright (c) 2004-2007 The University of Tennessee and The University
+ * Copyright (c) 2004-2020 The University of Tennessee and The University
  *                         of Tennessee Research Foundation.  All rights
  *                         reserved.
  * Copyright (c) 2004-2005 High Performance Computing Center Stuttgart,
@@ -55,11 +55,68 @@ OMPI_DECLSPEC extern int ompi_mpi_errcode_lastpredefined;
 
 OMPI_DECLSPEC extern ompi_mpi_errcode_t ompi_err_unknown;
 
+/**
+ * Initialize the error codes
+ *
+ * @returns OMPI_SUCCESS Upon success
+ * @returns OMPI_ERROR Otherwise
+ *
+ * Invoked from ompi_mpi_init(); sets up all static MPI error codes,
+ */
+int ompi_mpi_errcode_init(void);
+
+/**
+ * Finalize the error codes.
+ *
+ * @returns OMPI_SUCCESS Always
+ *
+ * Invokes from ompi_mpi_finalize(); tears down the error code array.
+ */
+int ompi_mpi_errcode_finalize(void);
+
+/**
+ * Add an error code
+ *
+ * @param: error class to which this new error code belongs to
+ *
+ * @returns the new error code on SUCCESS (>0)
+ * @returns OMPI_ERROR otherwise
+ *
+ */
+int ompi_mpi_errcode_add (int errclass);
+
+/**
+ * Add an error class
+ *
+ * @param: none
+ *
+ * @returns the new error class on SUCCESS (>0)
+ * @returns OMPI_ERROR otherwise
+ *
+ */
+int ompi_mpi_errclass_add (void);
+
+/**
+ * Add an error string to an error code
+ *
+ * @param: error code for which the string is defined
+ * @param: error string to add
+ * @param: length of the string
+ *
+ * @returns OMPI_SUCCESS on success
+ * @returns OMPI_ERROR on error
+ */
+int ompi_mpi_errnum_add_string (int errnum, const char* string, int len);
+
 /**
  * Check for a valid error code
  */
 static inline bool ompi_mpi_errcode_is_invalid(int errcode)
 {
+    if (OPAL_UNLIKELY( 0 == ompi_mpi_errcode_lastpredefined )) {
+        ompi_mpi_errcode_init();
+    }
+
     if ( errcode >= 0 && errcode <= ompi_mpi_errcode_lastused )
         return 0;
     else
@@ -73,23 +130,31 @@ static inline int ompi_mpi_errcode_get_class (int errcode)
 {
     ompi_mpi_errcode_t *err = NULL;
 
+    if (OPAL_UNLIKELY( 0 == ompi_mpi_errcode_lastpredefined )) {
+        ompi_mpi_errcode_init();
+    }
+
     if (errcode >= 0) {
         err = (ompi_mpi_errcode_t *)opal_pointer_array_get_item(&ompi_mpi_errcodes, errcode);
         /* If we get a bogus errcode, return MPI_ERR_UNKNOWN */
     }
 
     if (NULL != err) {
-	if ( err->code != MPI_UNDEFINED ) {
-	    return err->cls;
-	}
+        if ( err->code != MPI_UNDEFINED ) {
+            return err->cls;
+        }
     }
     return ompi_err_unknown.cls;
 }
 
 static inline int ompi_mpi_errcode_is_predefined ( int errcode )
 {
+    if (OPAL_UNLIKELY( 0 == ompi_mpi_errcode_lastpredefined )) {
+        ompi_mpi_errcode_init();
+    }
+
     if ( errcode >= 0 && errcode <= ompi_mpi_errcode_lastpredefined )
-	return true;
+        return true;
 
     return false;
 }
@@ -98,23 +163,27 @@ static inline int ompi_mpi_errnum_is_class ( int errnum )
 {
     ompi_mpi_errcode_t *err;
 
+    if (OPAL_UNLIKELY( 0 == ompi_mpi_errcode_lastpredefined )) {
+        ompi_mpi_errcode_init();
+    }
+
     if (errnum < 0) {
         return false;
     }
 
     if ( errnum <= ompi_mpi_errcode_lastpredefined ) {
-	/* Predefined error values represent an error code and
-	   an error class at the same time */
-	return true;
+        /* Predefined error values represent an error code and
+           an error class at the same time */
+        return true;
     }
 
     err = (ompi_mpi_errcode_t *)opal_pointer_array_get_item(&ompi_mpi_errcodes, errnum);
     if (NULL != err) {
-	if ( MPI_UNDEFINED == err->code) {
-	    /* Distinction between error class and error code is that for the
-	       first one the code section is set to MPI_UNDEFINED  */
-	    return true;
-	}
+        if ( MPI_UNDEFINED == err->code) {
+            /* Distinction between error class and error code is that for the
+               first one the code section is set to MPI_UNDEFINED  */
+            return true;
+        }
     }
 
     return false;
@@ -128,6 +197,10 @@ static inline char* ompi_mpi_errnum_get_string (int errnum)
 {
     ompi_mpi_errcode_t *err = NULL;
 
+    if (OPAL_UNLIKELY( 0 == ompi_mpi_errcode_lastpredefined )) {
+        ompi_mpi_errcode_init();
+    }
+
     if (errnum >= 0) {
         err = (ompi_mpi_errcode_t *)opal_pointer_array_get_item(&ompi_mpi_errcodes, errnum);
         /* If we get a bogus errcode, return a string indicating that this
@@ -142,59 +215,6 @@ static inline char* ompi_mpi_errnum_get_string (int errnum)
 }
 
 
-/**
- * Initialize the error codes
- *
- * @returns OMPI_SUCCESS Upon success
- * @returns OMPI_ERROR Otherwise
- *
- * Invoked from ompi_mpi_init(); sets up all static MPI error codes,
- */
-int ompi_mpi_errcode_init(void);
-
-/**
- * Finalize the error codes.
- *
- * @returns OMPI_SUCCESS Always
- *
- * Invokes from ompi_mpi_finalize(); tears down the error code array.
- */
-int ompi_mpi_errcode_finalize(void);
-
-/**
- * Add an error code
- *
- * @param: error class to which this new error code belongs to
- *
- * @returns the new error code on SUCCESS (>0)
- * @returns OMPI_ERROR otherwise
- *
- */
-int ompi_mpi_errcode_add (int errclass);
-
-/**
- * Add an error class
- *
- * @param: none
- *
- * @returns the new error class on SUCCESS (>0)
- * @returns OMPI_ERROR otherwise
- *
- */
-int ompi_mpi_errclass_add (void);
-
-/**
- * Add an error string to an error code
- *
- * @param: error code for which the string is defined
- * @param: error string to add
- * @param: length of the string
- *
- * @returns OMPI_SUCCESS on success
- * @returns OMPI_ERROR on error
- */
-int ompi_mpi_errnum_add_string (int errnum, const char* string, int len);
-
 END_C_DECLS
 
 #endif /* OMPI_MPI_ERRCODE_H */