X-Git-Url: http://info.iut-bm.univ-fcomte.fr/pub/gitweb/simgrid.git/blobdiff_plain/86051ed1b66342bae0e21fd6643e66331ac2c07d..4b70e717ae88822e3985f35e1c8f37bf330976a2:/tools/tesh/run_context.c

diff --git a/tools/tesh/run_context.c b/tools/tesh/run_context.c
index 3f65821eb4..b234a084ac 100644
--- a/tools/tesh/run_context.c
+++ b/tools/tesh/run_context.c
@@ -13,32 +13,81 @@
 #include <sys/types.h>
 #include <sys/wait.h>
 
-
 XBT_LOG_EXTERNAL_DEFAULT_CATEGORY(tesh);
 
 xbt_dynar_t bg_jobs = NULL;
+rctx_t armageddon_initiator = NULL;
+xbt_mutex_t armageddon_mutex = NULL;
 
 /* 
  * Module management
  */
 
-static void join_it(void*t) {  
-  xbt_thread_t th = *(xbt_thread_t*)t;
-  VERB1("Join thread %p which were running a background cmd",th);
-  xbt_thread_join(th,NULL);
+static void kill_it(void*r) {  
+  rctx_t rctx = *(rctx_t*)r;
+
+  VERB1("Join thread %p which were running a background cmd",rctx->runner);
+  xbt_thread_join(rctx->runner,NULL);
+  rctx_free(rctx);
 }
 
 void rctx_init(void) {
-  bg_jobs = xbt_dynar_new(sizeof(xbt_thread_t),join_it);
+  bg_jobs = xbt_dynar_new(sizeof(rctx_t),kill_it);
+  armageddon_mutex = xbt_mutex_init();
+  armageddon_initiator = NULL;
 }
 
 void rctx_exit(void) {
-  xbt_dynar_free(&bg_jobs);
+  if (bg_jobs)
+    xbt_dynar_free(&bg_jobs);
+  xbt_mutex_destroy(armageddon_mutex);
 }
 
 void rctx_wait_bg(void) {
   xbt_dynar_free(&bg_jobs);
-  bg_jobs = xbt_dynar_new(sizeof(xbt_thread_t),join_it);
+  bg_jobs = xbt_dynar_new(sizeof(rctx_t),kill_it);
+}
+
+void rctx_armageddon(rctx_t initiator, int exitcode) {
+  rctx_t rctx;
+  int cpt;
+
+  xbt_mutex_lock(armageddon_mutex);
+  if (armageddon_initiator != NULL) {
+    VERB0("Armageddon already started. Let it go");
+    return;
+  }
+  armageddon_initiator = initiator;
+  xbt_mutex_unlock(armageddon_mutex);
+
+
+  /* Kill any background commands */
+  xbt_dynar_foreach(bg_jobs,cpt,rctx) {
+    if (rctx != initiator) {
+      xbt_mutex_lock(rctx->interruption);
+      rctx->interrupted = 1;
+      INFO2("Kill <%s> because <%s> failed",rctx->filepos,initiator->filepos);
+      if (!rctx->reader_done) {
+	kill(rctx->pid,SIGTERM);
+	usleep(100);
+	kill(rctx->pid,SIGKILL);          
+      }
+      xbt_mutex_unlock(rctx->interruption);
+    }
+  }
+
+  /* Remove myself from the tasks */
+  if (xbt_dynar_member(bg_jobs, &initiator)) {
+    int mypos = xbt_dynar_search(bg_jobs, &initiator);
+    rctx_t myself;
+    xbt_dynar_remove_at(bg_jobs,mypos,&myself);
+    //    rctx_free(myself);
+  } 
+
+  /* Cleanup the place */
+  xbt_dynar_free(&bg_jobs);
+
+  exit(exitcode);
 }
 
 /*
@@ -49,12 +98,16 @@ void rctx_empty(rctx_t rc) {
   if (rc->cmd)
     free(rc->cmd);
   rc->cmd = NULL;
+  if (rc->filepos)
+    free(rc->filepos);
+  rc->filepos = NULL;
   rc->is_empty = 1;
   rc->is_background = 0;
   rc->is_stoppable = 0;
-  rc->check_output = 1;
+  rc->output = e_output_check;
   rc->brokenpipe = 0;
   rc->timeout = 0;
+  rc->interrupted = 0;
   buff_empty(rc->input);
   buff_empty(rc->output_wanted);
   buff_empty(rc->output_got);
@@ -66,6 +119,7 @@ rctx_t rctx_new() {
   res->input=buff_new();
   res->output_wanted=buff_new();
   res->output_got=buff_new();
+  res->interruption = xbt_mutex_init();
   rctx_empty(res);
   return res;
 }
@@ -78,6 +132,9 @@ void rctx_free(rctx_t rctx) {
 
   if (rctx->cmd)
     free(rctx->cmd);
+  if (rctx->filepos)
+    free(rctx->filepos);
+  xbt_mutex_destroy(rctx->interruption);
   buff_free(rctx->input);
   buff_free(rctx->output_got);
   buff_free(rctx->output_wanted);
@@ -109,7 +166,8 @@ void rctx_pushline(const char* filepos, char kind, char *line) {
 	ERROR2("[%s] More than one command in this chunk of lines (previous: %s).\n"
 	       " Dunno which input/output belongs to which command.",
 	       filepos,rctx->cmd);
-	exit(1);
+	ERROR1("Test suite `%s': NOK (syntax error)",testsuite_name);
+	rctx_armageddon(rctx,1);
       }
       rctx_start();
       VERB1("[%s] More than one command in this chunk of lines",filepos);
@@ -120,7 +178,8 @@ void rctx_pushline(const char* filepos, char kind, char *line) {
       rctx->is_background = 0;
       
     rctx->cmd = xbt_strdup(line);
-    INFO3("[%s] %s%s",filepos,line,
+    rctx->filepos = xbt_strdup(filepos);
+    INFO3("[%s] %s%s",filepos,rctx->cmd,
 	  ((rctx->is_background)?" (background command)":""));
 
     break;
@@ -157,13 +216,18 @@ void rctx_pushline(const char* filepos, char kind, char *line) {
       VERB2("[%s] (next command must return code %d)",
 	    filepos, rctx->expected_return);
 
-    } else if (!strncmp(line,"ignore output",strlen("ignore output"))) {
-      rctx->check_output = 0;
+    } else if (!strncmp(line,"output ignore",strlen("output ignore"))) {
+      rctx->output = e_output_ignore;
+      VERB1("[%s] (ignore output of next command)", filepos);
+       
+    } else if (!strncmp(line,"output display",strlen("output display"))) {
+      rctx->output = e_output_display;
       VERB1("[%s] (ignore output of next command)", filepos);
        
     } else {
       ERROR2("%s: Malformed metacommand: %s",filepos,line);
-      exit(1);
+      ERROR1("Test suite `%s': NOK (syntax error)",testsuite_name);
+      rctx_armageddon(rctx,1);
     }
     break;
   }
@@ -190,7 +254,8 @@ static void* thread_writer(void *r) {
 	rctx->brokenpipe = 1;
       } else if (errno!=EINTR && errno!=EAGAIN && errno!=EPIPE) {
 	perror("Error while writing input to child");
-	  exit(4);
+	ERROR1("Test suite `%s': NOK (system error)",testsuite_name);
+	rctx_armageddon(rctx,4);
       }
     }
     DEBUG1("written %d chars so far",posw);
@@ -207,13 +272,14 @@ static void* thread_writer(void *r) {
 static void *thread_reader(void *r) {
   rctx_t rctx = (rctx_t)r;
   char *buffout=malloc(4096);
-  int posr;
+  int posr, got_pid;
 
   do {
     posr=read(rctx->child_from,buffout,4095);
     if (posr<0 && errno!=EINTR && errno!=EAGAIN) {
       perror("Error while reading output of child");
-      exit(4);
+      ERROR1("Test suite `%s': NOK (system error)", testsuite_name);
+      rctx_armageddon(rctx,4);
     }
     if (posr>0) {
       buffout[posr]='\0';
@@ -223,6 +289,15 @@ static void *thread_reader(void *r) {
     }
   } while (!rctx->timeout && posr!=0);
   free(buffout);
+
+  /* let this thread wait for the child so that the main thread can detect the timeout without blocking on the wait */
+  got_pid = waitpid(rctx->pid,&rctx->status,0);
+  if (got_pid != rctx->pid) {
+    perror(bprintf("Cannot wait for the child %s",rctx->cmd));
+    ERROR1("Test suite `%s': NOK (system error)", testsuite_name);
+    rctx_armageddon(rctx,4);
+  }
+   
   rctx->reader_done = 1;
   return NULL;
 } 
@@ -238,13 +313,15 @@ void rctx_start(void) {
   VERB2("Start %s %s",rctx->cmd,(rctx->is_background?"(background job)":""));
   if (pipe(child_in) || pipe(child_out)) {
     perror("Cannot open the pipes");
-    exit(4);
+    ERROR1("Test suite `%s': NOK (system error)", testsuite_name);
+    rctx_armageddon(rctx,4);
   }
 
   rctx->pid=fork();
   if (rctx->pid<0) {
     perror("Cannot fork the command");
-    exit(4);
+    ERROR1("Test suite `%s': NOK (system error)", testsuite_name);
+    rctx_armageddon(rctx,4);
   }
 
   if (rctx->pid) { /* father */
@@ -256,6 +333,7 @@ void rctx_start(void) {
 
     rctx->end_time = time(NULL) + timeout_value;
 
+    rctx->reader_done = 0;
     rctx->reader = xbt_thread_create(thread_reader,(void*)rctx);
     rctx->writer = xbt_thread_create(thread_writer,(void*)rctx);
 
@@ -287,9 +365,10 @@ void rctx_start(void) {
 
     DEBUG2("Launch a thread to wait for %s %d",old->cmd,old->pid);
     runner = xbt_thread_create(rctx_wait,(void*)old);
+    old->runner = runner;
     VERB3("Launched thread %p to wait for %s %d",
 	  runner,old->cmd, old->pid);
-    xbt_dynar_push(bg_jobs,&runner);
+    xbt_dynar_push(bg_jobs,&old);
   }
 }
 
@@ -301,8 +380,7 @@ void rctx_start(void) {
 void *rctx_wait(void* r) {
   rctx_t rctx = (rctx_t)r;
   int errcode = 0;
-  int res;
-  int status;
+  int now = time(NULL);
     
   rctx_dump(rctx,"wait");
 
@@ -310,100 +388,142 @@ void *rctx_wait(void* r) {
     THROW1(unknown_error,0,"Cmd '%s' not started yet. Cannot wait it",
 	   rctx->cmd);
 
-  /* Wait for the child to die or the timeout to happen */
-  while (!rctx->reader_done && rctx->end_time > time(NULL)) {
+  /* Wait for the child to die or the timeout to happen (or an armageddon to happen) */
+  while (!rctx->interrupted && !rctx->reader_done && rctx->end_time >= now) {
     usleep(100);
+    now = time(NULL);
   }
+   
+  xbt_mutex_lock(rctx->interruption);
 
-  if (!rctx->reader_done) {
-    INFO1("Child '%s' timeouted. Kill it",rctx->cmd);
+  if (!rctx->interrupted && rctx->end_time < now) {    
+    INFO1("<%s> timeouted. Kill the process.",rctx->filepos);
     rctx->timeout = 1;
-    kill(rctx->pid,SIGKILL);
+    kill(rctx->pid,SIGTERM);
+    usleep(100);
+    kill(rctx->pid,SIGKILL);    
+    rctx->reader_done = 1;
   }
+   
   /* Make sure helper threads die.
      Cannot block since they wait for the child we just killed
      if not already dead. */
   xbt_thread_join(rctx->writer,NULL);
   xbt_thread_join(rctx->reader,NULL);
 
+  /*  xbt_mutex_unlock(rctx->interruption);
+  if (rctx->interrupted)
+    return NULL;
+    xbt_mutex_lock(rctx->interruption);*/
+ 
+  buff_chomp(rctx->output_got);
+  buff_chomp(rctx->output_wanted);
+  buff_trim(rctx->output_got);
+  buff_trim(rctx->output_wanted);
+
   /* Check for broken pipe */
   if (rctx->brokenpipe)
     VERB0("Warning: Child did not consume all its input (I got broken pipe)");
 
   /* Check for timeouts */
   if (rctx->timeout) {
-    ERROR1("Child timeouted (waited %d sec)",timeout_value);
-    exit(3);
+    if (rctx->output_got->data[0])
+      INFO2("<%s> Output on timeout:\n%s",
+	    rctx->filepos,rctx->output_got->data);
+    else
+      INFO1("<%s> No output before timeout",
+	    rctx->filepos);
+    ERROR3("Test suite `%s': NOK (<%s> timeout after %d sec)", 
+	   testsuite_name,rctx->filepos,timeout_value);
+    if (!rctx->interrupted)
+      rctx_armageddon(rctx, 3);
   }
       
-  DEBUG2("Wait for %s (%d)",rctx->cmd,rctx->pid);
-  res = waitpid(rctx->pid,&status,0);
-  if (res != rctx->pid) {
-    perror(bprintf("Cannot wait for the child %s",rctx->cmd));
-    exit(1);
-  }
   DEBUG2("RCTX=%p (pid=%d)",rctx,rctx->pid);
-  DEBUG3("Status(%s|%d)=%d",rctx->cmd,rctx->pid,status);
-
-  if (WIFSIGNALED(status) && !rctx->expected_signal) {
-    ERROR2("Child \"%s\" got signal %s.", rctx->cmd,
-	    signal_name(WTERMSIG(status),NULL));
-    errcode = WTERMSIG(status)+4;	
-  }
-
-  if (WIFSIGNALED(status) && rctx->expected_signal &&
-      strcmp(signal_name(WTERMSIG(status),rctx->expected_signal),
-	     rctx->expected_signal)) {
-    ERROR3("Child \"%s\" got signal %s instead of signal %s", rctx->cmd,
-	    signal_name(WTERMSIG(status),rctx->expected_signal),
-	    rctx->expected_signal);
-    errcode = WTERMSIG(status)+4;	
-  }
-  
-  if (!WIFSIGNALED(status) && rctx->expected_signal) {
-    ERROR2("Child \"%s\" didn't got expected signal %s",
-	   rctx->cmd, rctx->expected_signal);
-    errcode = 5;
-  }
-
-  if (WIFEXITED(status) && WEXITSTATUS(status) != rctx->expected_return ) {
-    if (rctx->expected_return) 
-      ERROR3("Child \"%s\" returned code %d instead of %d", rctx->cmd,
-	     WEXITSTATUS(status), rctx->expected_return);
-    else
-      ERROR2("Child \"%s\" returned code %d", rctx->cmd, WEXITSTATUS(status));
-    errcode = 40+WEXITSTATUS(status);
-  }
-  rctx->expected_return = 0;
+  DEBUG3("Status(%s|%d)=%d",rctx->cmd,rctx->pid,rctx->status);
+
+  if (!rctx->interrupted) {
+    if (WIFSIGNALED(rctx->status) && !rctx->expected_signal) {
+      ERROR3("Test suite `%s': NOK (<%s> got signal %s)", 
+	     testsuite_name, rctx->filepos,
+	     signal_name(WTERMSIG(rctx->status),NULL));
+      errcode = WTERMSIG(rctx->status)+4;	
+    }
+    
+    if (WIFSIGNALED(rctx->status) && rctx->expected_signal &&
+	strcmp(signal_name(WTERMSIG(rctx->status),rctx->expected_signal),
+	       rctx->expected_signal)) {
+      ERROR4("Test suite `%s': NOK (%s got signal %s instead of %s)", 
+	     testsuite_name, rctx->filepos,
+	     signal_name(WTERMSIG(rctx->status),rctx->expected_signal),
+	     rctx->expected_signal);
+      errcode = WTERMSIG(rctx->status)+4;	
+    }
+    
+    if (!WIFSIGNALED(rctx->status) && rctx->expected_signal) {
+      ERROR3("Test suite `%s': NOK (child %s expected signal %s)", 
+	     testsuite_name, rctx->filepos,
+	     rctx->expected_signal);
+      errcode = 5;
+    }
+    
+    if (WIFEXITED(rctx->status) && WEXITSTATUS(rctx->status) != rctx->expected_return ) {
+      if (rctx->expected_return) 
+	ERROR4("Test suite `%s': NOK (<%s> returned code %d instead of %d)",
+	       testsuite_name, rctx->filepos,
+	       WEXITSTATUS(rctx->status), rctx->expected_return);
+      else
+	ERROR3("Test suite `%s': NOK (<%s> returned code %d)",
+	       testsuite_name, rctx->filepos, WEXITSTATUS(rctx->status));
+      errcode = 40+WEXITSTATUS(rctx->status);
+      
+    }
+    rctx->expected_return = 0;
   
-  if(rctx->expected_signal){
-    free(rctx->expected_signal);
-    rctx->expected_signal = NULL;
+    if(rctx->expected_signal){
+      free(rctx->expected_signal);
+      rctx->expected_signal = NULL;
+    }
   }
 
-  buff_chomp(rctx->output_got);
-  buff_chomp(rctx->output_wanted);
-  buff_trim(rctx->output_got);
-  buff_trim(rctx->output_wanted);
-
-  if (   rctx->check_output 
+  if (   rctx->output == e_output_check
       && (    rctx->output_got->used != rctx->output_wanted->used
 	   || strcmp(rctx->output_got->data, rctx->output_wanted->data))) {
-    char *diff= xbt_str_diff(rctx->output_wanted->data,rctx->output_got->data);
-    ERROR2("Output of child \"%s\" don't match expectations. Here is a diff between expected and got output:\n%s",
-	   rctx->cmd,diff);
-    free(diff);
+    if (XBT_LOG_ISENABLED(tesh,xbt_log_priority_info)) {
+       char *diff= xbt_str_diff(rctx->output_wanted->data,rctx->output_got->data);	  
+       ERROR1("Output mismatch:\n%s",
+	      diff);
+       free(diff);
+    }     
+    ERROR2("Test suite `%s': NOK (<%s> output mismatch)", 
+	   testsuite_name,rctx->filepos);
+     
     errcode=2;
-  } else if (!rctx->check_output) {
+  } else if (rctx->output == e_output_ignore) {
     INFO0("(ignoring the output as requested)");
+  } else if (rctx->output == e_output_display) {
+    xbt_dynar_t a = xbt_str_split(rctx->output_got->data, "\n");
+    char *out = xbt_str_join(a,"\n||");
+    xbt_dynar_free(&a);
+    INFO1("Here is the (ignored) command output: \n||%s",out);
+    free(out);
+  } else if (errcode || rctx->interrupted) {
+    /* checking output, and matching */
+    xbt_dynar_t a = xbt_str_split(rctx->output_got->data, "\n");
+    char *out = xbt_str_join(a,"\n||");
+    xbt_dynar_free(&a);
+    INFO2("Output of <%s> so far: \n||%s",rctx->filepos,out);
+    free(out);    
   }
 
-  if (rctx->is_background)
-    rctx_free(rctx);
-  else
+  if (!rctx->is_background) {
     rctx_empty(rctx);
-  if (errcode)
-    exit (errcode);
+  }
+  if (errcode) {
+    if (!rctx->interrupted)
+      rctx_armageddon(rctx, errcode);
+  }
+  xbt_mutex_unlock(rctx->interruption);
 
   return NULL;
 }