X-Git-Url: http://info.iut-bm.univ-fcomte.fr/pub/gitweb/simgrid.git/blobdiff_plain/40dc499e35463c28c4572e077206864092452e6a..29493aabe8625e8359c877d61b143c3f98bb1072:/tools/tesh/run_context.c

diff --git a/tools/tesh/run_context.c b/tools/tesh/run_context.c
index b234a084ac..130c7f969e 100644
--- a/tools/tesh/run_context.c
+++ b/tools/tesh/run_context.c
@@ -17,7 +17,7 @@ XBT_LOG_EXTERNAL_DEFAULT_CATEGORY(tesh);
 
 xbt_dynar_t bg_jobs = NULL;
 rctx_t armageddon_initiator = NULL;
-xbt_mutex_t armageddon_mutex = NULL;
+xbt_os_mutex_t armageddon_mutex = NULL;
 
 /* 
  * Module management
@@ -26,67 +26,76 @@ xbt_mutex_t armageddon_mutex = 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);
+  VERB2("Join thread %p which were running background cmd <%s>",rctx->runner,rctx->filepos);
+  xbt_os_thread_join(rctx->runner,NULL);
   rctx_free(rctx);
 }
 
 void rctx_init(void) {
-  bg_jobs = xbt_dynar_new(sizeof(rctx_t),kill_it);
-  armageddon_mutex = xbt_mutex_init();
+  bg_jobs = xbt_dynar_new_sync(sizeof(rctx_t),kill_it);
+  armageddon_mutex = xbt_os_mutex_init();
   armageddon_initiator = NULL;
 }
 
 void rctx_exit(void) {
-  if (bg_jobs)
+  if (bg_jobs) {
+    /* Do not use xbt_dynar_free or it will lock the dynar, preventing armageddon from working */
+    while (xbt_dynar_length(bg_jobs)) {
+       rctx_t rctx;
+       xbt_dynar_pop(bg_jobs,&rctx);
+       kill_it(&rctx);
+    }
     xbt_dynar_free(&bg_jobs);
-  xbt_mutex_destroy(armageddon_mutex);
+  }
+  xbt_os_mutex_destroy(armageddon_mutex);
 }
 
 void rctx_wait_bg(void) {
-  xbt_dynar_free(&bg_jobs);
-  bg_jobs = xbt_dynar_new(sizeof(rctx_t),kill_it);
+  if (bg_jobs) {
+    /* Do not use xbt_dynar_free or it will lock the dynar, preventing armageddon from working */
+    while (xbt_dynar_length(bg_jobs)) {
+       rctx_t rctx;
+       xbt_dynar_pop(bg_jobs,&rctx);
+       kill_it(&rctx);
+    }
+    xbt_dynar_free(&bg_jobs);
+  }
+  bg_jobs = xbt_dynar_new_sync(sizeof(rctx_t),kill_it);
 }
 
 void rctx_armageddon(rctx_t initiator, int exitcode) {
   rctx_t rctx;
-  int cpt;
 
-  xbt_mutex_lock(armageddon_mutex);
+  DEBUG2("Armageddon request by <%s> (exit=%d)",initiator->filepos,exitcode);
+  xbt_os_mutex_lock(armageddon_mutex);
   if (armageddon_initiator != NULL) {
     VERB0("Armageddon already started. Let it go");
+    xbt_os_mutex_unlock(initiator->interruption);
+    xbt_os_mutex_unlock(armageddon_mutex);
     return;
   }
+  DEBUG1("Armageddon request by <%s> got the lock. Let's go amok",initiator->filepos);
   armageddon_initiator = initiator;
-  xbt_mutex_unlock(armageddon_mutex);
-
+  xbt_os_mutex_unlock(armageddon_mutex);
 
   /* Kill any background commands */
-  xbt_dynar_foreach(bg_jobs,cpt,rctx) {
+  while (xbt_dynar_length(bg_jobs)) {
+    xbt_dynar_pop(bg_jobs,&rctx);
     if (rctx != initiator) {
-      xbt_mutex_lock(rctx->interruption);
-      rctx->interrupted = 1;
       INFO2("Kill <%s> because <%s> failed",rctx->filepos,initiator->filepos);
+      xbt_os_mutex_lock(rctx->interruption);
+      rctx->interrupted = 1;
+      xbt_os_mutex_unlock(rctx->interruption);
+      INFO2("Do 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);
-
+  VERB0("Shut everything down!");
   exit(exitcode);
 }
 
@@ -119,7 +128,7 @@ rctx_t rctx_new() {
   res->input=buff_new();
   res->output_wanted=buff_new();
   res->output_got=buff_new();
-  res->interruption = xbt_mutex_init();
+  res->interruption = xbt_os_mutex_init();
   rctx_empty(res);
   return res;
 }
@@ -134,7 +143,7 @@ void rctx_free(rctx_t rctx) {
     free(rctx->cmd);
   if (rctx->filepos)
     free(rctx->filepos);
-  xbt_mutex_destroy(rctx->interruption);
+  xbt_os_mutex_destroy(rctx->interruption);
   buff_free(rctx->input);
   buff_free(rctx->output_got);
   buff_free(rctx->output_wanted);
@@ -168,6 +177,7 @@ void rctx_pushline(const char* filepos, char kind, char *line) {
 	       filepos,rctx->cmd);
 	ERROR1("Test suite `%s': NOK (syntax error)",testsuite_name);
 	rctx_armageddon(rctx,1);
+	return;
       }
       rctx_start();
       VERB1("[%s] More than one command in this chunk of lines",filepos);
@@ -200,8 +210,11 @@ void rctx_pushline(const char* filepos, char kind, char *line) {
     if (rctx->cmd)
       rctx_start();
 
-    if (!strncmp(line,"set timeout ",strlen("set timeout "))) {
-      timeout_value=atoi(line+strlen("set timeout"));
+    if (!strncmp(line,"timeout no",strlen("timeout no"))) {
+      VERB1("[%s] (disable timeout)", filepos);
+      timeout_value = -1;
+    } else if (!strncmp(line,"timeout ",strlen("timeout "))) {
+      timeout_value=atoi(line+strlen("timeout"));
       VERB2("[%s] (new timeout value: %d)",
 	     filepos,timeout_value);
 
@@ -228,6 +241,7 @@ void rctx_pushline(const char* filepos, char kind, char *line) {
       ERROR2("%s: Malformed metacommand: %s",filepos,line);
       ERROR1("Test suite `%s': NOK (syntax error)",testsuite_name);
       rctx_armageddon(rctx,1);
+      return;
     }
     break;
   }
@@ -256,6 +270,7 @@ static void* thread_writer(void *r) {
 	perror("Error while writing input to child");
 	ERROR1("Test suite `%s': NOK (system error)",testsuite_name);
 	rctx_armageddon(rctx,4);
+	return NULL;
       }
     }
     DEBUG1("written %d chars so far",posw);
@@ -280,6 +295,7 @@ static void *thread_reader(void *r) {
       perror("Error while reading output of child");
       ERROR1("Test suite `%s': NOK (system error)", testsuite_name);
       rctx_armageddon(rctx,4);
+      return NULL;
     }
     if (posr>0) {
       buffout[posr]='\0';
@@ -296,6 +312,7 @@ static void *thread_reader(void *r) {
     perror(bprintf("Cannot wait for the child %s",rctx->cmd));
     ERROR1("Test suite `%s': NOK (system error)", testsuite_name);
     rctx_armageddon(rctx,4);
+    return NULL;
   }
    
   rctx->reader_done = 1;
@@ -322,6 +339,7 @@ void rctx_start(void) {
     perror("Cannot fork the command");
     ERROR1("Test suite `%s': NOK (system error)", testsuite_name);
     rctx_armageddon(rctx,4);
+    return;
   }
 
   if (rctx->pid) { /* father */
@@ -331,11 +349,14 @@ void rctx_start(void) {
     close(child_out[1]);
     rctx->child_from = child_out[0];
 
-    rctx->end_time = time(NULL) + timeout_value;
+    if (timeout_value > 0)
+       rctx->end_time = time(NULL) + timeout_value;
+    else 
+       rctx->end_time = -1;
 
     rctx->reader_done = 0;
-    rctx->reader = xbt_thread_create(thread_reader,(void*)rctx);
-    rctx->writer = xbt_thread_create(thread_writer,(void*)rctx);
+    rctx->reader = xbt_os_thread_create("reader",thread_reader,(void*)rctx);
+    rctx->writer = xbt_os_thread_create("writer",thread_writer,(void*)rctx);
 
   } else { /* child */
 
@@ -358,13 +379,13 @@ void rctx_start(void) {
   } else {
     /* Damn. Copy the rctx and launch a thread to handle it */
     rctx_t old = rctx;
-    xbt_thread_t runner;
+    xbt_os_thread_t runner;
 
     rctx = rctx_new();
     DEBUG2("RCTX: new bg=%p, new fg=%p",old,rctx);
 
     DEBUG2("Launch a thread to wait for %s %d",old->cmd,old->pid);
-    runner = xbt_thread_create(rctx_wait,(void*)old);
+    runner = xbt_os_thread_create(old->cmd,rctx_wait,(void*)old);
     old->runner = runner;
     VERB3("Launched thread %p to wait for %s %d",
 	  runner,old->cmd, old->pid);
@@ -375,7 +396,7 @@ void rctx_start(void) {
 /* Waits for the child to end (or to timeout), and check its 
    ending conditions. This is launched from rctx_start but either in main
    thread (for foreground jobs) or in a separate one for background jobs. 
-   That explains the prototype, forced by xbt_thread_create. */
+   That explains the prototype, forced by xbt_os_thread_create. */
 
 void *rctx_wait(void* r) {
   rctx_t rctx = (rctx_t)r;
@@ -389,14 +410,13 @@ void *rctx_wait(void* r) {
 	   rctx->cmd);
 
   /* 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) {
+  while (!rctx->interrupted && !rctx->reader_done && (rctx->end_time <0 ||rctx->end_time >= now)) {
     usleep(100);
     now = time(NULL);
   }
    
-  xbt_mutex_lock(rctx->interruption);
-
-  if (!rctx->interrupted && rctx->end_time < now) {    
+  xbt_os_mutex_lock(rctx->interruption);
+  if (!rctx->interrupted && rctx->end_time > 0 && rctx->end_time < now) {    
     INFO1("<%s> timeouted. Kill the process.",rctx->filepos);
     rctx->timeout = 1;
     kill(rctx->pid,SIGTERM);
@@ -408,13 +428,13 @@ void *rctx_wait(void* r) {
   /* 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_os_thread_join(rctx->writer,NULL);
+  xbt_os_thread_join(rctx->reader,NULL);
 
-  /*  xbt_mutex_unlock(rctx->interruption);
+  /*  xbt_os_mutex_unlock(rctx->interruption);
   if (rctx->interrupted)
     return NULL;
-    xbt_mutex_lock(rctx->interruption);*/
+    xbt_os_mutex_lock(rctx->interruption);*/
  
   buff_chomp(rctx->output_got);
   buff_chomp(rctx->output_wanted);
@@ -435,8 +455,11 @@ void *rctx_wait(void* r) {
 	    rctx->filepos);
     ERROR3("Test suite `%s': NOK (<%s> timeout after %d sec)", 
 	   testsuite_name,rctx->filepos,timeout_value);
-    if (!rctx->interrupted)
+    DEBUG2("<%s> Interrupted = %d", rctx->filepos, rctx->interrupted);
+    if (!rctx->interrupted) {
       rctx_armageddon(rctx, 3);
+      return NULL;
+    }
   }
       
   DEBUG2("RCTX=%p (pid=%d)",rctx,rctx->pid);
@@ -491,8 +514,7 @@ void *rctx_wait(void* r) {
 	   || strcmp(rctx->output_got->data, rctx->output_wanted->data))) {
     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);
+       ERROR2("Output of <%s> mismatch:\n%s",rctx->filepos,diff);
        free(diff);
     }     
     ERROR2("Test suite `%s': NOK (<%s> output mismatch)", 
@@ -500,7 +522,7 @@ void *rctx_wait(void* r) {
      
     errcode=2;
   } else if (rctx->output == e_output_ignore) {
-    INFO0("(ignoring the output as requested)");
+    INFO1("(ignoring the output of <%s> as requested)",rctx->filepos);
   } 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||");
@@ -520,11 +542,13 @@ void *rctx_wait(void* r) {
     rctx_empty(rctx);
   }
   if (errcode) {
-    if (!rctx->interrupted)
+    if (!rctx->interrupted) {
       rctx_armageddon(rctx, errcode);
+      return NULL;
+    }
   }
-  xbt_mutex_unlock(rctx->interruption);
 
+  xbt_os_mutex_unlock(rctx->interruption);
   return NULL;
 }