Merge pull request #1183 from urbit/simpler-checkpoint

simplifies loom init/restore, checkpoints during replay

include/noun/manage.h

@@ -26,10 +26,10 @@
         c3_i
         u3m_bail(c3_m how_m) __attribute__((noreturn));
 
-      /* u3m_init(): start the environment, with/without checkpointing.
+      /* u3m_init(): start the environment.
       */
         void
-        u3m_init(c3_o chk_o);
+        u3m_init();
 
       /* u3m_pave(): instantiate or activate image.
       */

noun/events.c

@@ -165,9 +165,9 @@ u3e_fault(void* adr_v, c3_i ser_i)
 /* _ce_image_open(): open or create image.
 */
 static c3_o
-_ce_image_open(u3e_image* img_u, c3_o nuu_o)
+_ce_image_open(u3e_image* img_u)
 {
-  c3_i mod_i = _(nuu_o) ? (O_RDWR | O_CREAT) : O_RDWR;
+  c3_i mod_i = O_RDWR | O_CREAT;
   c3_c ful_c[8193];
 
   snprintf(ful_c, 8192, "%s", u3P.dir_c);
@@ -197,11 +197,7 @@ _ce_image_open(u3e_image* img_u, c3_o nuu_o)
       c3_d pgs_d = (siz_d + (c3_d)((1 << (u3a_page + 2)) - 1)) >> 
                    (c3_d)(u3a_page + 2);
 
-      if ( c3y == nuu_o ) {
-        if ( siz_d ) { 
-          c3_assert(0);
-          return c3n;
-        }
+      if ( !siz_d ) {
         return c3y;
       }
       else { 
@@ -817,7 +813,7 @@ u3e_save(void)
   _ce_patch_free(pat_u);
 }
 
-/* u3e_live(): start the persistence system.
+/* u3e_live(): start the checkpointing system.
 */
 c3_o
 u3e_live(c3_o nuu_o, c3_c* dir_c)
@@ -832,29 +828,17 @@ u3e_live(c3_o nuu_o, c3_c* dir_c)
   } else
 #endif
   {
-    /* Open and apply any patches.
-    */
-    if ( _(nuu_o) ) {
-      if ( (c3n == _ce_image_open(&u3P.nor_u, c3y)) ||
-           (c3n == _ce_image_open(&u3P.sou_u, c3y)) )
-      {
-        printf("boot: image failed\r\n");
-        exit(1);
-      }
+    //  Open image files.
+    //
+    if ( (c3n == _ce_image_open(&u3P.nor_u)) ||
+         (c3n == _ce_image_open(&u3P.sou_u)) )
+    {
+      printf("boot: image failed\r\n");
+      exit(1);
     }
     else {
       u3_ce_patch* pat_u;
 
-      /* Open image files.
-      */
-      {
-        if ( (c3n == _ce_image_open(&u3P.nor_u, c3n)) ||
-             (c3n == _ce_image_open(&u3P.sou_u, c3n)) ) 
-        {
-          fprintf(stderr, "boot: no image\r\n");
-          return u3e_live(c3y, dir_c);
-        }
-      }
       /* Load any patch files; apply them to images.
       */
       if ( 0 != (pat_u = _ce_patch_open()) ) {

noun/manage.c

@@ -1495,10 +1495,10 @@ _cm_signals(void)
   }
 }
 
-/* u3m_init(): start the environment, with/without checkpointing.
+/* u3m_init(): start the environment.
 */
 void
-u3m_init(c3_o chk_o)
+u3m_init(void)
 {
   _cm_limits();
   _cm_signals();
@@ -1515,7 +1515,7 @@ u3m_init(c3_o chk_o)
 
     map_v = mmap((void *)u3_Loom,
                  len_w,
-                 _(chk_o) ? PROT_READ : (PROT_READ | PROT_WRITE),
+                 (PROT_READ | PROT_WRITE),
                  (MAP_ANON | MAP_FIXED | MAP_PRIVATE),
                  -1, 0);
 
@@ -1694,7 +1694,7 @@ u3m_boot(c3_o nuu_o, c3_o bug_o, c3_c* dir_c,
 {
   /* Activate the loom.
   */
-  u3m_init(nuu_o);
+  u3m_init();
 
   /* Activate the storage system.
   */

tests/hash_tests.c

@@ -5,7 +5,7 @@
 static void
 _setup(void)
 {
-  u3m_init(c3y);
+  u3m_init();
   u3m_pave(c3y, c3n);
 }
 

tests/hashtable_tests.c

@@ -26,7 +26,7 @@ main(int argc, char* argv[])
 static void
 _setup(void)
 {
-  u3m_init(c3y);
+  u3m_init();
   u3m_pave(c3y, c3n);
 }
 

tests/test.c

@@ -254,7 +254,7 @@ main(int argc, char *argv[])
   printf("hello, world: len %dMB\n", (1 << U3_OS_LoomBits) >> 18);
   // _test_words();
 
-  u3m_init(c3n);
+  u3m_init();
   u3m_pave(c3y, c3n);
   // u3j_boot();
 

vere/sist.c

@@ -208,31 +208,57 @@ u3_sist_nil(const c3_c* key_c)
 /* _sist_suck(): past failure.
 */
 static void
-_sist_suck(u3_noun ovo, u3_noun gon)
+_sist_suck(c3_d evt_d, u3_noun ovo, u3_noun gon)
 {
-  uL(fprintf(uH, "sing: ovum failed!\n"));
+  fprintf(stderr, "sing: ovum failed!\r\n");
+
   {
     c3_c* hed_c = u3r_string(u3h(u3t(ovo)));
 
-    uL(fprintf(uH, "fail %s\n", hed_c));
+    fprintf(stderr, "sing: fail: %s event %lld mug %u\r\n",
+                    hed_c, evt_d, u3r_mug(ovo));
     free(hed_c);
   }
 
-  u3_lo_punt(2, u3kb_flop(u3k(u3t(gon))));
-  // u3_loom_exit();
-#if 1
-  u3_lo_exit();
+  {
+    u3_noun why;
+    u3x_cell(gon, &why, 0);
+    u3m_p("sist: why", why);
+  }
 
-  exit(1);
-#else
-  u3z(ovo); u3z(gon);
+  u3_lo_punt(2, u3kb_flop(u3k(u3t(gon))));
+
+#if 1
+  {
+    c3_c fil_c[2048];
+    snprintf(fil_c, 2048, "%s/.urb/put/failed-%lld.jam", u3_Host.dir_c, evt_d);
+
+    u3_noun pat = u3qe_jam(ovo);
+
+    fprintf(stderr, "sing: saving failed event\r\n");
+    u3_walk_save(fil_c, u3_nul, pat, 0, u3_nul);
+  }
+#endif
+
+#if 1
+  {
+    fprintf(stderr, "sing: saving checkpoint\r\n");
+    c3_d old_d = u3A->ent_d;
+    u3A->ent_d = evt_d - 1ULL;
+    u3e_save();
+    u3A->ent_d = old_d;
+  }
+#endif
+
+#if 1
+  u3_lo_bail();
 #endif
 }
 
 /* _sist_sing(): replay ovum from the past, time already set.
 */
 static void
-_sist_sing(u3_noun ovo)
+_sist_sing(c3_d evt_d, u3_noun ovo)
 {
   u3t_event_trace("Running", 'b');
   u3_noun gon = u3m_soft(0, u3v_poke, u3k(ovo));
@@ -243,7 +269,7 @@ _sist_sing(u3_noun ovo)
     u3x_cell(gon, &hed, &tal);
 
     if ( u3_blip != hed ) {
-      _sist_suck(ovo, gon);
+      _sist_suck(evt_d, ovo, gon);
     }
     else {
       u3_noun vir, cor;
@@ -1077,7 +1103,7 @@ _sist_rest()
         fprintf(stderr, "replay: skipped veer\n");
       }
       else {
-        _sist_sing(u3k(ovo));
+        _sist_sing(lar_u.ent_d, u3k(ovo));
         fputc('.', stderr);
       }
 
@@ -1085,6 +1111,17 @@ _sist_rest()
 
       xno_w++;
 
+#if 1
+      //  save a checkpoint every 100K events
+      //
+      if ( 0 == (xno_w % 100000) ) {
+        c3_d old_d = u3A->ent_d;
+        u3A->ent_d = lar_u.ent_d;
+        u3e_save();
+        u3A->ent_d = old_d;
+      }
+#endif
+
       if ( 0 == (xno_w % 1000) ) {
         uL(fprintf(uH, "{%d}\n", xno_w));
         // u3_lo_grab("rest", rou, u3_none);