Browse Source

Flatten and compress in-memory debug info (saves ~70%).

Mike Pall 14 years ago
parent
commit
8c32b38ca3
14 changed files with 344 additions and 165 deletions
  1. 4 3
      src/Makefile.dep
  2. 3 3
      src/lib_jit.c
  3. 2 2
      src/lj_api.c
  4. 126 68
      src/lj_debug.c
  5. 20 2
      src/lj_debug.h
  6. 1 1
      src/lj_err.c
  7. 1 1
      src/lj_err.h
  8. 0 4
      src/lj_gc.c
  9. 1 1
      src/lj_gdbjit.c
  10. 1 1
      src/lj_lex.c
  11. 7 0
      src/lj_lex.h
  12. 10 19
      src/lj_obj.h
  13. 167 59
      src/lj_parse.c
  14. 1 1
      src/lj_trace.c

+ 4 - 3
src/Makefile.dep

@@ -85,7 +85,8 @@ lj_crecord.o: lj_crecord.c lj_obj.h lua.h luaconf.h lj_def.h lj_arch.h \
 lj_ctype.o: lj_ctype.c lj_obj.h lua.h luaconf.h lj_def.h lj_arch.h \
  lj_gc.h lj_err.h lj_errmsg.h lj_str.h lj_tab.h lj_ctype.h
 lj_debug.o: lj_debug.c lj_obj.h lua.h luaconf.h lj_def.h lj_arch.h \
- lj_err.h lj_errmsg.h lj_str.h lj_tab.h lj_state.h lj_frame.h lj_bc.h
+ lj_err.h lj_errmsg.h lj_debug.h lj_str.h lj_tab.h lj_state.h lj_frame.h \
+ lj_bc.h
 lj_dispatch.o: lj_dispatch.c lj_obj.h lua.h luaconf.h lj_def.h lj_arch.h \
  lj_err.h lj_errmsg.h lj_debug.h lj_state.h lj_frame.h lj_bc.h lj_ff.h \
  lj_ffdef.h lj_jit.h lj_ir.h lj_trace.h lj_dispatch.h lj_traceerr.h \
@@ -143,8 +144,8 @@ lj_opt_split.o: lj_opt_split.c lj_obj.h lua.h luaconf.h lj_def.h \
  lj_arch.h lj_err.h lj_errmsg.h lj_str.h lj_ir.h lj_jit.h lj_ircall.h \
  lj_iropt.h lj_vm.h
 lj_parse.o: lj_parse.c lj_obj.h lua.h luaconf.h lj_def.h lj_arch.h \
- lj_gc.h lj_err.h lj_errmsg.h lj_str.h lj_tab.h lj_func.h lj_state.h \
- lj_bc.h lj_ctype.h lj_lex.h lj_parse.h lj_vm.h lj_vmevent.h
+ lj_gc.h lj_err.h lj_errmsg.h lj_debug.h lj_str.h lj_tab.h lj_func.h \
+ lj_state.h lj_bc.h lj_ctype.h lj_lex.h lj_parse.h lj_vm.h lj_vmevent.h
 lj_record.o: lj_record.c lj_obj.h lua.h luaconf.h lj_def.h lj_arch.h \
  lj_err.h lj_errmsg.h lj_str.h lj_tab.h lj_meta.h lj_frame.h lj_bc.h \
  lj_ff.h lj_ffdef.h lj_ir.h lj_jit.h lj_ircall.h lj_iropt.h lj_trace.h \

+ 3 - 3
src/lib_jit.c

@@ -181,8 +181,8 @@ LJLIB_CF(jit_util_funcinfo)
     GCtab *t;
     lua_createtable(L, 0, 16);  /* Increment hash size if fields are added. */
     t = tabV(L->top-1);
-    setintfield(L, t, "linedefined", proto_line(pt, 0));
-    setintfield(L, t, "lastlinedefined", pt->lastlinedefined);
+    setintfield(L, t, "linedefined", pt->firstline);
+    setintfield(L, t, "lastlinedefined", pt->firstline + pt->numline);
     setintfield(L, t, "stackslots", pt->framesize);
     setintfield(L, t, "params", pt->numparams);
     setintfield(L, t, "bytecodes", (int32_t)pt->sizebc);
@@ -254,7 +254,7 @@ LJLIB_CF(jit_util_funcuvname)
   GCproto *pt = check_Lproto(L, 0);
   uint32_t idx = (uint32_t)lj_lib_checkint(L, 2);
   if (idx < pt->sizeuv) {
-    setstrV(L, L->top-1, proto_uvname(pt, idx));
+    setstrV(L, L->top-1, lj_str_newz(L, lj_debug_uvname(pt, idx)));
     return 1;
   }
   return 0;

+ 2 - 2
src/lj_api.c

@@ -836,7 +836,7 @@ LUA_API int lua_next(lua_State *L, int idx)
 LUA_API const char *lua_getupvalue(lua_State *L, int idx, int n)
 {
   TValue *val;
-  const char *name = lj_debug_uvname(index2adr(L, idx), (uint32_t)(n-1), &val);
+  const char *name = lj_debug_uvnamev(index2adr(L, idx), (uint32_t)(n-1), &val);
   if (name) {
     copyTV(L, L->top, val);
     incr_top(L);
@@ -991,7 +991,7 @@ LUA_API const char *lua_setupvalue(lua_State *L, int idx, int n)
   TValue *val;
   const char *name;
   api_checknelems(L, 1);
-  name = lj_debug_uvname(f, (uint32_t)(n-1), &val);
+  name = lj_debug_uvnamev(f, (uint32_t)(n-1), &val);
   if (name) {
     L->top--;
     copyTV(L, val, L->top);

+ 126 - 68
src/lj_debug.c

@@ -8,6 +8,7 @@
 
 #include "lj_obj.h"
 #include "lj_err.h"
+#include "lj_debug.h"
 #include "lj_str.h"
 #include "lj_tab.h"
 #include "lj_state.h"
@@ -89,9 +90,17 @@ static BCPos debug_framepc(lua_State *L, GCfunc *fn, cTValue *nextframe)
 /* Get line number for a bytecode position. */
 BCLine LJ_FASTCALL lj_debug_line(GCproto *pt, BCPos pc)
 {
-  BCLine *lineinfo = proto_lineinfo(pt);
-  if (pc < pt->sizebc && lineinfo)
-    return lineinfo[pc];
+  const void *lineinfo = proto_lineinfo(pt);
+  if (pc < pt->sizebc && lineinfo) {
+    BCLine first = pt->firstline;
+    if (pc-- == 0) return first;
+    if (pt->numline < 256)
+      return first + (BCLine)((const uint8_t *)lineinfo)[pc];
+    else if (pt->numline < 65536)
+      return first + (BCLine)((const uint16_t *)lineinfo)[pc];
+    else
+      return first + (BCLine)((const uint32_t *)lineinfo)[pc];
+  }
   return 0;
 }
 
@@ -109,14 +118,49 @@ static BCLine debug_frameline(lua_State *L, GCfunc *fn, cTValue *nextframe)
 
 /* -- Variable names ------------------------------------------------------ */
 
+/* Read ULEB128 value. */
+static uint32_t debug_read_uleb128(const uint8_t **pp)
+{
+  const uint8_t *p = *pp;
+  uint32_t v = *p++;
+  if (LJ_UNLIKELY(v >= 0x80)) {
+    int sh = 0;
+    v &= 0x7f;
+    do { v |= ((*p & 0x7f) << (sh += 7)); } while (*p++ >= 0x80);
+  }
+  *pp = p;
+  return v;
+}
+
 /* Get name of a local variable from slot number and PC. */
 static const char *debug_varname(const GCproto *pt, BCPos pc, BCReg slot)
 {
-  MSize i;
-  VarInfo *vi = proto_varinfo(pt);
-  for (i = 0; i < pt->sizevarinfo && vi[i].startpc <= pc; i++)
-    if (pc < vi[i].endpc && slot-- == 0)
-      return strdata(gco2str(gcref(vi[i].name)));
+  const uint8_t *p = proto_varinfo(pt);
+  if (p) {
+    BCPos lastpc = 0;
+    for (;;) {
+      const char *name = (const char *)p;
+      uint32_t vn = *p++;
+      BCPos startpc, endpc;
+      if (vn < VARNAME__MAX) {
+	if (vn == VARNAME_END) break;  /* End of varinfo. */
+      } else {
+	while (*p++) ;  /* Skip over variable name string. */
+      }
+      lastpc = startpc = lastpc + debug_read_uleb128(&p);
+      if (startpc > pc) break;
+      endpc = startpc + debug_read_uleb128(&p);
+      if (pc < endpc && slot-- == 0) {
+	if (vn < VARNAME__MAX) {
+#define VARNAMESTR(name, str)	str "\0"
+	  name = VARNAMEDEF(VARNAMESTR);
+#undef VARNAMESTR
+	  if (--vn) while (*name++ || --vn) ;
+	}
+	return name;
+      }
+    }
+  }
   return NULL;
 }
 
@@ -141,7 +185,17 @@ static TValue *debug_localname(lua_State *L, const lua_Debug *ar,
 }
 
 /* Get name of upvalue. */
-const char *lj_debug_uvname(cTValue *o, uint32_t idx, TValue **tvp)
+const char *lj_debug_uvname(GCproto *pt, uint32_t idx)
+{
+  const uint8_t *p = proto_uvinfo(pt);
+  lua_assert(idx < pt->sizeuv);
+  if (!p) return "";
+  if (idx) while (*p++ || --idx) ;
+  return (const char *)p;
+}
+
+/* Get name and value of upvalue. */
+const char *lj_debug_uvnamev(cTValue *o, uint32_t idx, TValue **tvp)
 {
   if (tvisfunc(o)) {
     GCfunc *fn = funcV(o);
@@ -149,7 +203,7 @@ const char *lj_debug_uvname(cTValue *o, uint32_t idx, TValue **tvp)
       GCproto *pt = funcproto(fn);
       if (idx < pt->sizeuv) {
 	*tvp = uvval(&gcref(fn->l.uvptr[idx])->uv);
-	return strdata(proto_uvname(pt, idx));
+	return lj_debug_uvname(pt, idx);
       }
     } else {
       if (idx < fn->c.nupvalues) {
@@ -194,7 +248,7 @@ restart:
 	}
 	return "field";
       case BC_UGET:
-	*name = strdata(proto_uvname(pt, bc_d(ins)));
+	*name = lj_debug_uvname(pt, bc_d(ins));
 	return "upvalue";
       default:
 	return NULL;
@@ -234,25 +288,26 @@ const char *lj_debug_funcname(lua_State *L, TValue *frame, const char **name)
 /* -- Source code locations ----------------------------------------------- */
 
 /* Generate shortened source name. */
-void lj_debug_shortname(char *out, const char *src)
+void lj_debug_shortname(char *out, GCstr *str)
 {
+  const char *src = strdata(str);
   if (*src == '=') {
-    strncpy(out, src+1, LUA_IDSIZE);  /* remove first char */
-    out[LUA_IDSIZE-1] = '\0';  /* ensures null termination */
-  } else if (*src == '@') { /* out = "source", or "...source" */
-    size_t l = strlen(++src);  /* skip the `@' */
-    if (l >= LUA_IDSIZE) {
-      src += l-(LUA_IDSIZE-4);  /* get last part of file name */
-      strcpy(out, "...");
-      out += 3;
+    strncpy(out, src+1, LUA_IDSIZE);  /* Remove first char. */
+    out[LUA_IDSIZE-1] = '\0';  /* Ensures null termination. */
+  } else if (*src == '@') {  /* Output "source", or "...source". */
+    size_t len = str->len-1;
+    src++;  /* Skip the `@' */
+    if (len >= LUA_IDSIZE) {
+      src += len-(LUA_IDSIZE-4);  /* Get last part of file name. */
+      *out++ = '.'; *out++ = '.'; *out++ = '.';
     }
     strcpy(out, src);
-  } else {  /* out = [string "string"] */
-    size_t len; /* Length, up to first control char. */
+  } else {  /* Output [string "string"]. */
+    size_t len;  /* Length, up to first control char. */
     for (len = 0; len < LUA_IDSIZE-12; len++)
       if (((const unsigned char *)src)[len] < ' ') break;
     strcpy(out, "[string \""); out += 9;
-    if (src[len] != '\0') {  /* must truncate? */
+    if (src[len] != '\0') {  /* Must truncate? */
       if (len > LUA_IDSIZE-15) len = LUA_IDSIZE-15;
       strncpy(out, src, len); out += len;
       strcpy(out, "..."); out += 3;
@@ -273,7 +328,7 @@ void lj_debug_addloc(lua_State *L, const char *msg,
       BCLine line = debug_frameline(L, fn, nextframe);
       if (line >= 0) {
 	char buf[LUA_IDSIZE];
-	lj_debug_shortname(buf, strdata(proto_chunkname(funcproto(fn))));
+	lj_debug_shortname(buf, proto_chunkname(funcproto(fn)));
 	lj_str_pushf(L, "%s:%d: %s", buf, line, msg);
 	return;
       }
@@ -286,27 +341,23 @@ void lj_debug_addloc(lua_State *L, const char *msg,
 void lj_debug_pushloc(lua_State *L, GCproto *pt, BCPos pc)
 {
   GCstr *name = proto_chunkname(pt);
-  if (name) {
-    const char *s = strdata(name);
-    MSize i, len = name->len;
-    BCLine line = lj_debug_line(pt, pc);
-    if (*s == '@') {
-      s++; len--;
-      for (i = len; i > 0; i--)
-	if (s[i] == '/' || s[i] == '\\') {
-	  s += i+1;
-	  break;
-	}
-      lj_str_pushf(L, "%s:%d", s, line);
-    } else if (len > 40) {
-      lj_str_pushf(L, "%p:%d", pt, line);
-    } else if (*s == '=') {
-      lj_str_pushf(L, "%s:%d", s+1, line);
-    } else {
-      lj_str_pushf(L, "\"%s\":%d", s, line);
-    }
+  const char *s = strdata(name);
+  MSize i, len = name->len;
+  BCLine line = lj_debug_line(pt, pc);
+  if (*s == '@') {
+    s++; len--;
+    for (i = len; i > 0; i--)
+      if (s[i] == '/' || s[i] == '\\') {
+	s += i+1;
+	break;
+      }
+    lj_str_pushf(L, "%s:%d", s, line);
+  } else if (len > 40) {
+    lj_str_pushf(L, "%p:%d", pt, line);
+  } else if (*s == '=') {
+    lj_str_pushf(L, "%s:%d", s+1, line);
   } else {
-    lj_str_pushf(L, "%p:%u", pt, pc);
+    lj_str_pushf(L, "\"%s\":%d", s, line);
   }
 }
 
@@ -359,56 +410,63 @@ LUA_API int lua_getinfo(lua_State *L, const char *what, lua_Debug *ar)
     lua_assert(fn->c.gct == ~LJ_TFUNC);
   }
   for (; *what; what++) {
-    switch (*what) {
-    case 'S':
+    if (*what == 'S') {
       if (isluafunc(fn)) {
 	GCproto *pt = funcproto(fn);
-	ar->source = strdata(proto_chunkname(pt));
-	ar->linedefined = (int)proto_line(pt, 0);
-	ar->lastlinedefined = (int)pt->lastlinedefined;
-	ar->what = ar->linedefined ? "Lua" : "main";
+	BCLine firstline = pt->firstline;
+	GCstr *name = proto_chunkname(pt);
+	ar->source = strdata(name);
+	lj_debug_shortname(ar->short_src, name);
+	ar->linedefined = (int)firstline;
+	ar->lastlinedefined = (int)(firstline + pt->numline);
+	ar->what = firstline ? "Lua" : "main";
       } else {
 	ar->source = "=[C]";
+	ar->short_src[0] = '[';
+	ar->short_src[1] = 'C';
+	ar->short_src[2] = ']';
+	ar->short_src[3] = '\0';
 	ar->linedefined = -1;
 	ar->lastlinedefined = -1;
 	ar->what = "C";
       }
-      lj_debug_shortname(ar->short_src, ar->source);
-      break;
-    case 'l':
+    } else if (*what == 'l') {
       ar->currentline = frame ? debug_frameline(L, fn, nextframe) : -1;
-      break;
-    case 'u':
+    } else if (*what == 'u') {
       ar->nups = fn->c.nupvalues;
-      break;
-    case 'n':
+    } else if (*what == 'n') {
       ar->namewhat = frame ? lj_debug_funcname(L, frame, &ar->name) : NULL;
       if (ar->namewhat == NULL) {
 	ar->namewhat = "";
 	ar->name = NULL;
       }
-      break;
-    case 'f':
+    } else if (*what == 'f') {
       setfuncV(L, L->top, fn);
       incr_top(L);
-      break;
-    case 'L':
+    } else if (*what == 'L') {
       if (isluafunc(fn)) {
 	GCtab *t = lj_tab_new(L, 0, 0);
 	GCproto *pt = funcproto(fn);
-	BCLine *lineinfo = proto_lineinfo(pt);
-	MSize i, szl = pt->sizebc;
-	for (i = 1; i < szl; i++)
-	  setboolV(lj_tab_setint(L, t, lineinfo[i]), 1);
+	const void *lineinfo = proto_lineinfo(pt);
+	if (lineinfo) {
+	  BCLine first = pt->firstline;
+	  int sz = pt->numline < 256 ? 1 : pt->numline < 65536 ? 2 : 4;
+	  MSize i, szl = pt->sizebc-1;
+	  for (i = 0; i < szl; i++) {
+	    BCLine line = first +
+	      (sz == 1 ? (BCLine)((const uint8_t *)lineinfo)[i] :
+	       sz == 2 ? (BCLine)((const uint16_t *)lineinfo)[i] :
+	       (BCLine)((const uint32_t *)lineinfo)[i]);
+	    setboolV(lj_tab_setint(L, t, line), 1);
+	  }
+	}
 	settabV(L, L->top, t);
       } else {
 	setnilV(L->top);
       }
       incr_top(L);
-      break;
-    default:
+    } else {
       status = 0;  /* Bad option. */
-      break;
     }
   }
   return status;

+ 20 - 2
src/lj_debug.h

@@ -10,14 +10,32 @@
 
 LJ_FUNC cTValue *lj_debug_frame(lua_State *L, int level, int *size);
 LJ_FUNC BCLine LJ_FASTCALL lj_debug_line(GCproto *pt, BCPos pc);
-LJ_FUNC const char *lj_debug_uvname(cTValue *o, uint32_t idx, TValue **tvp);
+LJ_FUNC const char *lj_debug_uvname(GCproto *pt, uint32_t idx);
+LJ_FUNC const char *lj_debug_uvnamev(cTValue *o, uint32_t idx, TValue **tvp);
 LJ_FUNC const char *lj_debug_slotname(GCproto *pt, const BCIns *pc,
 				      BCReg slot, const char **name);
 LJ_FUNC const char *lj_debug_funcname(lua_State *L, TValue *frame,
 				      const char **name);
-LJ_FUNC void lj_debug_shortname(char *out, const char *src);
+LJ_FUNC void lj_debug_shortname(char *out, GCstr *str);
 LJ_FUNC void lj_debug_addloc(lua_State *L, const char *msg,
 			     cTValue *frame, cTValue *nextframe);
 LJ_FUNC void lj_debug_pushloc(lua_State *L, GCproto *pt, BCPos pc);
 
+/* Fixed internal variable names. */
+#define VARNAMEDEF(_) \
+  _(FOR_IDX, "(for index)") \
+  _(FOR_STOP, "(for limit)") \
+  _(FOR_STEP, "(for step)") \
+  _(FOR_GEN, "(for generator)") \
+  _(FOR_STATE, "(for state)") \
+  _(FOR_CTL, "(for control)")
+
+enum {
+  VARNAME_END,
+#define VARNAMEENUM(name, str)	VARNAME_##name,
+  VARNAMEDEF(VARNAMEENUM)
+#undef VARNAMEENUM
+  VARNAME__MAX
+};
+
 #endif

+ 1 - 1
src/lj_err.c

@@ -521,7 +521,7 @@ LJ_NOINLINE void lj_err_msg(lua_State *L, ErrMsg em)
 }
 
 /* Lexer error. */
-LJ_NOINLINE void lj_err_lex(lua_State *L, const char *src, const char *tok,
+LJ_NOINLINE void lj_err_lex(lua_State *L, GCstr *src, const char *tok,
 			    BCLine line, ErrMsg em, va_list argp)
 {
   char buff[LUA_IDSIZE];

+ 1 - 1
src/lj_err.h

@@ -25,7 +25,7 @@ LJ_FUNCA_NORET void LJ_FASTCALL lj_err_throw(lua_State *L, int errcode);
 LJ_FUNC_NORET void lj_err_mem(lua_State *L);
 LJ_FUNC_NORET void lj_err_run(lua_State *L);
 LJ_FUNC_NORET void lj_err_msg(lua_State *L, ErrMsg em);
-LJ_FUNC_NORET void lj_err_lex(lua_State *L, const char *src, const char *tok,
+LJ_FUNC_NORET void lj_err_lex(lua_State *L, GCstr *src, const char *tok,
 			      BCLine line, ErrMsg em, va_list argp);
 LJ_FUNC_NORET void lj_err_optype(lua_State *L, cTValue *o, ErrMsg opm);
 LJ_FUNC_NORET void lj_err_comp(lua_State *L, cTValue *o1, cTValue *o2);

+ 0 - 4
src/lj_gc.c

@@ -257,10 +257,6 @@ static void gc_traverse_proto(global_State *g, GCproto *pt)
   gc_mark_str(proto_chunkname(pt));
   for (i = -(ptrdiff_t)pt->sizekgc; i < 0; i++)  /* Mark collectable consts. */
     gc_markobj(g, proto_kgc(pt, i));
-  for (i = 0; i < (ptrdiff_t)pt->sizeuv; i++)  /* Mark upvalue names. */
-    gc_mark_str(proto_uvname(pt, i));
-  for (i = 0; i < (ptrdiff_t)pt->sizevarinfo; i++)  /* Mark names of locals. */
-    gc_mark_str(gco2str(gcref(proto_varinfo(pt)[i].name)));
 #if LJ_HASJIT
   if (pt->trace) gc_marktrace(g, pt->trace);
 #endif

+ 1 - 1
src/lj_gdbjit.c

@@ -731,7 +731,7 @@ void lj_gdbjit_addtrace(jit_State *J, GCtrace *T)
   ctx.spadj = CFRAME_SIZE_JIT + T->spadjust;
   lua_assert(startpc >= proto_bc(pt) && startpc < proto_bc(pt) + pt->sizebc);
   ctx.lineno = lj_debug_line(pt, proto_bcpos(pt, startpc));
-  ctx.filename = strdata(proto_chunkname(pt));
+  ctx.filename = proto_chunknamestr(pt);
   if (*ctx.filename == '@' || *ctx.filename == '=')
     ctx.filename++;
   else

+ 1 - 1
src/lj_lex.c

@@ -494,7 +494,7 @@ void lj_lex_error(LexState *ls, LexToken token, ErrMsg em, ...)
     tok = lj_lex_token2str(ls, token);
   }
   va_start(argp, em);
-  lj_err_lex(ls->L, strdata(ls->chunkname), tok, ls->linenumber, em, argp);
+  lj_err_lex(ls->L, ls->chunkname, tok, ls->linenumber, em, argp);
   va_end(argp);
 }
 

+ 7 - 0
src/lj_lex.h

@@ -38,6 +38,13 @@ typedef struct BCInsLine {
   BCLine line;		/* Line number for this bytecode. */
 } BCInsLine;
 
+/* Info for local variables. Only used during bytecode generation. */
+typedef struct VarInfo {
+  GCRef name;		/* Local variable name. */
+  BCPos startpc;	/* First point where the local variable is active. */
+  BCPos endpc;		/* First point where the local variable is dead. */
+} VarInfo;
+
 /* Lua lexer state. */
 typedef struct LexState {
   struct FuncState *fs;	/* Current FuncState. Defined in lj_parse.c. */

+ 10 - 19
src/lj_obj.h

@@ -286,12 +286,6 @@ typedef struct GCcdataVar {
 #define SCALE_NUM_GCO	((int32_t)sizeof(lua_Number)/sizeof(GCRef))
 #define round_nkgc(n)	(((n) + SCALE_NUM_GCO-1) & ~(SCALE_NUM_GCO-1))
 
-typedef struct VarInfo {
-  GCRef name;		/* Local variable name. */
-  BCPos startpc;	/* First point where the local variable is active. */
-  BCPos endpc;		/* First point where the local variable is dead. */
-} VarInfo;
-
 typedef struct GCproto {
   GCHeader;
   uint8_t numparams;	/* Number of parameters. */
@@ -308,11 +302,11 @@ typedef struct GCproto {
   uint16_t trace;	/* Anchor for chain of root traces. */
   /* ------ The following fields are for debugging/tracebacks only ------ */
   GCRef chunkname;	/* Name of the chunk this function was defined in. */
-  BCLine lastlinedefined;  /* Last line of the function definition. */
-  MSize sizevarinfo;	/* Size of local var info array. */
-  MRef varinfo;		/* Names and extents of local variables. */
-  MRef uvname;		/* Array of upvalue names (GCRef of GCstr). */
-  MRef lineinfo;	/* Map from bytecode instructions to source lines. */
+  BCLine firstline;	/* First line of the function definition. */
+  BCLine numline;	/* Number of lines for the function definition. */
+  MRef lineinfo;	/* Compressed map from bytecode ins. to source line. */
+  MRef uvinfo;		/* Upvalue names. */
+  MRef varinfo;		/* Names and compressed extents of local variables. */
 } GCproto;
 
 #define PROTO_IS_VARARG		0x01
@@ -331,14 +325,11 @@ typedef struct GCproto {
 #define proto_bcpos(pt, pc)	((BCPos)((pc) - proto_bc(pt)))
 #define proto_uv(pt)		(mref((pt)->uv, uint16_t))
 
-#define proto_uvname(pt, idx) \
-  check_exp((uintptr_t)(idx) < (pt)->sizeuv, \
-	    gco2str(gcref(mref((pt)->uvname, GCRef)[(idx)])))
-#define proto_chunkname(pt)	(gco2str(gcref((pt)->chunkname)))
-#define proto_lineinfo(pt)	(mref((pt)->lineinfo, BCLine))
-#define proto_line(pt, pos) \
-  check_exp((uintptr_t)(pos) < (pt)->sizebc, proto_lineinfo(pt)[(pos)])
-#define proto_varinfo(pt)	(mref((pt)->varinfo, VarInfo))
+#define proto_chunkname(pt)	(strref((pt)->chunkname))
+#define proto_chunknamestr(pt)	(strdata(proto_chunkname((pt))))
+#define proto_lineinfo(pt)	(mref((pt)->lineinfo, const void))
+#define proto_uvinfo(pt)	(mref((pt)->uvinfo, const uint8_t))
+#define proto_varinfo(pt)	(mref((pt)->varinfo, const uint8_t))
 
 /* -- Upvalue object ------------------------------------------------------ */
 

+ 167 - 59
src/lj_parse.c

@@ -12,6 +12,7 @@
 #include "lj_obj.h"
 #include "lj_gc.h"
 #include "lj_err.h"
+#include "lj_debug.h"
 #include "lj_str.h"
 #include "lj_tab.h"
 #include "lj_func.h"
@@ -1021,7 +1022,8 @@ static void var_new(LexState *ls, BCReg n, GCstr *name)
       lj_lex_error(ls, 0, LJ_ERR_XLIMC, LJ_MAX_VSTACK);
     lj_mem_growvec(ls->L, ls->vstack, ls->sizevstack, LJ_MAX_VSTACK, VarInfo);
   }
-  lua_assert(lj_tab_getstr(fs->kt, name) != NULL);
+  lua_assert((uintptr_t)name < VARNAME__MAX ||
+	     lj_tab_getstr(fs->kt, name) != NULL);
   /* NOBARRIER: name is anchored in fs->kt and ls->vstack is not a GCobj. */
   setgcref(ls->vstack[vtop].name, obj2gco(name));
   fs->varmap[fs->nactvar+n] = (uint16_t)vtop;
@@ -1031,6 +1033,9 @@ static void var_new(LexState *ls, BCReg n, GCstr *name)
 #define var_new_lit(ls, n, v) \
   var_new(ls, (n), lj_parse_keepstr(ls, "" v, sizeof(v)-1))
 
+#define var_new_fixed(ls, n, vn) \
+  var_new(ls, (n), (GCstr *)(uintptr_t)(vn))
+
 /* Add local variables. */
 static void var_add(LexState *ls, BCReg nvars)
 {
@@ -1053,7 +1058,7 @@ static BCReg var_lookup_local(FuncState *fs, GCstr *n)
 {
   int i;
   for (i = fs->nactvar-1; i >= 0; i--) {
-    if (n == gco2str(gcref(var_get(fs->ls, fs, i).name)))
+    if (n == strref(var_get(fs->ls, fs, i).name))
       return (BCReg)i;
   }
   return (BCReg)-1;  /* Not found. */
@@ -1109,23 +1114,16 @@ static MSize var_lookup_(FuncState *fs, GCstr *name, ExpDesc *e, int first)
 
 /* -- Function state management ------------------------------------------- */
 
-/* NYI: compress debug info. */
-
-/* Fixup bytecode and lineinfo for prototype. */
-static void fs_fixup_bc(FuncState *fs, GCproto *pt, BCIns *bc, BCLine *lineinfo)
+/* Fixup bytecode for prototype. */
+static void fs_fixup_bc(FuncState *fs, GCproto *pt, BCIns *bc, MSize n)
 {
-  MSize i, n = fs->pc;
   BCInsLine *base = fs->bcbase;
-  setmref(pt->lineinfo, lineinfo);
+  MSize i;
   pt->sizebc = n;
-  bc[n] = ~0u;  /* Close potentially uninitialized gap between bc and kgc. */
   bc[0] = BCINS_AD((fs->flags & PROTO_IS_VARARG) ? BC_FUNCV : BC_FUNCF,
 		   fs->framesize, 0);
-  lineinfo[0] = fs->linedefined;
-  for (i = 1; i < n; i++) {
+  for (i = 1; i < n; i++)
     bc[i] = base[i].ins;
-    lineinfo[i] = base[i].line;
-  }
 }
 
 /* Fixup constants for prototype. */
@@ -1189,20 +1187,138 @@ static void fs_fixup_uv(FuncState *fs, GCproto *pt, uint16_t *uv)
     uv[i] = fs->uvloc[i].slot;
 }
 
-/* Fixup debug info for prototype. */
-static void fs_fixup_dbg(FuncState *fs, GCproto *pt, VarInfo *vi, MSize sizevi)
+#ifndef LUAJIT_DISABLE_DEBUGINFO
+/* Prepare lineinfo for prototype. */
+static size_t fs_prep_line(FuncState *fs, BCLine numline)
+{
+  return (fs->pc-1) << (numline < 256 ? 0 : numline < 65536 ? 1 : 2);
+}
+
+/* Fixup lineinfo for prototype. */
+static void fs_fixup_line(FuncState *fs, GCproto *pt,
+			  void *lineinfo, BCLine numline)
+{
+  BCInsLine *base = fs->bcbase + 1;
+  BCLine first = fs->linedefined;
+  MSize i = 0, n = fs->pc-1;
+  pt->firstline = fs->linedefined;
+  pt->numline = numline;
+  setmref(pt->lineinfo, lineinfo);
+  if (LJ_LIKELY(numline < 256)) {
+    uint8_t *li = (uint8_t *)lineinfo;
+    do {
+      BCLine delta = base[i].line - first;
+      lua_assert(delta >= 0 && delta < 256);
+      li[i] = (uint8_t)delta;
+    } while (++i < n);
+  } else if (LJ_LIKELY(numline < 65536)) {
+    uint16_t *li = (uint16_t *)lineinfo;
+    do {
+      BCLine delta = base[i].line - first;
+      lua_assert(delta >= 0 && delta < 65536);
+      li[i] = (uint16_t)delta;
+    } while (++i < n);
+  } else {
+    uint32_t *li = (uint32_t *)lineinfo;
+    do {
+      BCLine delta = base[i].line - first;
+      lua_assert(delta >= 0);
+      li[i] = (uint32_t)delta;
+    } while (++i < n);
+  }
+}
+
+/* Resize buffer if needed. */
+static LJ_NOINLINE void fs_buf_resize(LexState *ls, MSize len)
+{
+  MSize sz = ls->sb.sz * 2;
+  while (ls->sb.n + len > sz) sz = sz * 2;
+  lj_str_resizebuf(ls->L, &ls->sb, sz);
+}
+
+static LJ_AINLINE void fs_buf_need(LexState *ls, MSize len)
+{
+  if (LJ_UNLIKELY(ls->sb.n + len > ls->sb.sz))
+    fs_buf_resize(ls, len);
+}
+
+/* Add string to buffer. */
+static void fs_buf_str(LexState *ls, const char *str, MSize len)
+{
+  char *p = ls->sb.buf + ls->sb.n;
+  MSize i;
+  ls->sb.n += len;
+  for (i = 0; i < len; i++) p[i] = str[i];
+}
+
+/* Add ULEB128 value to buffer. */
+static void fs_buf_uleb128(LexState *ls, uint32_t v)
+{
+  MSize n = ls->sb.n;
+  uint8_t *p = (uint8_t *)ls->sb.buf;
+  for (; v >= 0x80; v >>= 7)
+    p[n++] = (uint8_t)((v & 0x7f) | 0x80);
+  p[n++] = (uint8_t)v;
+  ls->sb.n = n;
+}
+
+/* Prepare variable info for prototype. */
+static size_t fs_prep_var(LexState *ls, FuncState *fs, size_t *ofsvar)
 {
-  MSize i, n = fs->nuv;
-  GCRef *uvname = (GCRef *)((char *)vi + sizevi*sizeof(VarInfo));
   VarInfo *vstack = fs->ls->vstack;
-  setmref(pt->varinfo, vi);
-  setmref(pt->uvname, uvname);
-  pt->sizevarinfo = sizevi;
-  memcpy(vi, &vstack[fs->vbase], sizevi*sizeof(VarInfo));
-  for (i = 0; i < n; i++)
-    setgcref(uvname[i], gcref(vstack[fs->uvloc[i].vidx].name));
+  MSize i, n;
+  BCPos lastpc;
+  lj_str_resetbuf(&ls->sb);  /* Copy to temp. string buffer. */
+  /* Store upvalue names. */
+  for (i = 0, n = fs->nuv; i < n; i++) {
+    GCstr *s = strref(vstack[fs->uvloc[i].vidx].name);
+    MSize len = s->len+1;
+    fs_buf_need(ls, len);
+    fs_buf_str(ls, strdata(s), len);
+  }
+  *ofsvar = ls->sb.n;
+  vstack += fs->vbase;
+  lastpc = 0;
+  /* Store local variable names and compressed ranges. */
+  for (i = 0, n = ls->vtop - fs->vbase; i < n; i++) {
+    GCstr *s = strref(vstack[i].name);
+    BCPos startpc = vstack[i].startpc, endpc = vstack[i].endpc;
+    if ((uintptr_t)s < VARNAME__MAX) {
+      fs_buf_need(ls, 1 + 2*5);
+      ls->sb.buf[ls->sb.n++] = (uint8_t)(uintptr_t)s;
+    } else {
+      MSize len = s->len+1;
+      fs_buf_need(ls, len + 2*5);
+      fs_buf_str(ls, strdata(s), len);
+    }
+    fs_buf_uleb128(ls, startpc-lastpc);
+    fs_buf_uleb128(ls, endpc-startpc);
+    lastpc = startpc;
+  }
+  fs_buf_need(ls, 1);
+  ls->sb.buf[ls->sb.n++] = '\0';  /* Terminator for varinfo. */
+  return ls->sb.n;
 }
 
+/* Fixup variable info for prototype. */
+static void fs_fixup_var(LexState *ls, GCproto *pt, uint8_t *p, size_t ofsvar)
+{
+  setmref(pt->uvinfo, p);
+  setmref(pt->varinfo, (char *)p + ofsvar);
+  memcpy(p, ls->sb.buf, ls->sb.n);  /* Copy from temp. string buffer. */
+}
+#else
+
+/* Initialize with empty debug info, if disabled. */
+#define fs_prep_line(fs, numline)		(UNUSED(numline), 0)
+#define fs_fixup_line(fs, pt, li, numline) \
+  pt->firstline = pt->numline = 0, setmref((pt)->lineinfo, NULL)
+#define fs_prep_var(ls, fs, ofsvar)		(UNUSED(ofsvar), 0)
+#define fs_fixup_var(ls, pt, p, ofsvar) \
+  setmref((pt)->uvinfo, NULL), setmref((pt)->varinfo, NULL)
+
+#endif
+
 /* Check if bytecode op returns. */
 static int bcopisret(BCOp op)
 {
@@ -1253,8 +1369,8 @@ static GCproto *fs_finish(LexState *ls, BCLine line)
 {
   lua_State *L = ls->L;
   FuncState *fs = ls->fs;
-  MSize sizevi;
-  size_t sizept, ofsk, ofsuv, ofsdbg, ofsli;
+  BCLine numline = line - fs->linedefined;
+  size_t sizept, ofsk, ofsuv, ofsli, ofsdbg, ofsvar;
   GCproto *pt;
 
   /* Apply final fixups. */
@@ -1264,33 +1380,29 @@ static GCproto *fs_finish(LexState *ls, BCLine line)
 
   /* Calculate total size of prototype including all colocated arrays. */
   sizept = sizeof(GCproto) + fs->pc*sizeof(BCIns) + fs->nkgc*sizeof(GCRef);
-  sizept = (sizept + sizeof(lua_Number)-1) & ~(sizeof(lua_Number)-1);
-  ofsk = sizept;
-  sizept += fs->nkn*sizeof(lua_Number);
-  ofsuv = sizept;
-  sizept += ((fs->nuv+1)&~1)*2;
-  ofsdbg = sizept;
-  sizevi = ls->vtop - fs->vbase;
-  sizept += sizevi*sizeof(VarInfo) + fs->nuv*sizeof(GCRef);
-  ofsli = sizept;
-  sizept += fs->pc*sizeof(BCLine);
+  sizept = (sizept + sizeof(TValue)-1) & ~(sizeof(TValue)-1);
+  ofsk = sizept; sizept += fs->nkn*sizeof(TValue);
+  ofsuv = sizept; sizept += ((fs->nuv+1)&~1)*2;
+  ofsli = sizept; sizept += fs_prep_line(fs, numline);
+  ofsdbg = sizept; sizept += fs_prep_var(ls, fs, &ofsvar);
 
   /* Allocate prototype and initialize its fields. */
   pt = (GCproto *)lj_mem_newgco(L, (MSize)sizept);
   pt->gct = ~LJ_TPROTO;
   pt->sizept = (MSize)sizept;
-  setgcref(pt->chunkname, obj2gco(ls->chunkname));
   pt->trace = 0;
   pt->flags = fs->flags;
   pt->numparams = fs->numparams;
   pt->framesize = fs->framesize;
-  pt->lastlinedefined = line;
+  setgcref(pt->chunkname, obj2gco(ls->chunkname));
 
-  fs_fixup_bc(fs, pt, (BCIns *)((char *)pt + sizeof(GCproto)),
-		      (BCLine *)((char *)pt + ofsli));
+  /* Close potentially uninitialized gap between bc and kgc. */
+  *(uint32_t *)((char *)pt + ofsk - sizeof(GCRef)*(fs->nkgc+1)) = 0;
+  fs_fixup_bc(fs, pt, (BCIns *)((char *)pt + sizeof(GCproto)), fs->pc);
   fs_fixup_k(fs, pt, (void *)((char *)pt + ofsk));
   fs_fixup_uv(fs, pt, (uint16_t *)((char *)pt + ofsuv));
-  fs_fixup_dbg(fs, pt, (VarInfo *)((char *)pt + ofsdbg), sizevi);
+  fs_fixup_line(fs, pt, (void *)((char *)pt + ofsli), numline);
+  fs_fixup_var(ls, pt, (uint8_t *)((char *)pt + ofsdbg), ofsvar);
 
   lj_vmevent_send(L, BC,
     setprotoV(L, L->top++, pt);
@@ -1300,12 +1412,6 @@ static GCproto *fs_finish(LexState *ls, BCLine line)
   ls->vtop = fs->vbase;  /* Reset variable stack. */
   ls->fs = fs->prev;
   lua_assert(ls->fs != NULL || ls->token == TK_eof);
-  /* Re-anchor last string token to avoid GC. */
-  if (ls->token == TK_name || ls->token == TK_string) {
-    /* NOBARRIER: the key is new or kept alive. */
-    TValue *tv = lj_tab_setstr(ls->L, ls->fs->kt, strV(&ls->tokenval));
-    if (tvisnil(tv)) setboolV(tv, 1);
-  }
   return pt;
 }
 
@@ -1530,8 +1636,6 @@ static void parse_chunk(LexState *ls);
 static void parse_body(LexState *ls, ExpDesc *e, int needself, BCLine line)
 {
   FuncState fs, *pfs = ls->fs;
-  BCReg kidx;
-  BCLine lastline;
   GCproto *pt;
   ptrdiff_t oldbase = pfs->bcbase - ls->bcstack;
   fs_init(ls, &fs);
@@ -1541,19 +1645,19 @@ static void parse_body(LexState *ls, ExpDesc *e, int needself, BCLine line)
   fs.bclim = pfs->bclim - pfs->pc;
   bcemit_AD(&fs, BC_FUNCF, 0, 0);  /* Placeholder. */
   parse_chunk(ls);
-  lastline = ls->linenumber;
-  lex_match(ls, TK_end, TK_function, line);
-  pt = fs_finish(ls, lastline);
+  if (ls->token != TK_end) lex_match(ls, TK_end, TK_function, line);
+  pt = fs_finish(ls, (ls->lastline = ls->linenumber));
   pfs->bcbase = ls->bcstack + oldbase;  /* May have been reallocated. */
   pfs->bclim = (BCPos)(ls->sizebcstack - oldbase);
   /* Store new prototype in the constant array of the parent. */
-  kidx = const_gc(pfs, obj2gco(pt), LJ_TPROTO);
-  expr_init(e, VRELOCABLE, bcemit_AD(pfs, BC_FNEW, 0, kidx));
+  expr_init(e, VRELOCABLE,
+	    bcemit_AD(pfs, BC_FNEW, 0, const_gc(pfs, obj2gco(pt), LJ_TPROTO)));
   if (!(pfs->flags & PROTO_HAS_FNEW)) {
     if (pfs->flags & PROTO_HAS_RETURN)
       pfs->flags |= PROTO_FIXUP_RETURN;
     pfs->flags |= PROTO_HAS_FNEW;
   }
+  lj_lex_next(ls);
 }
 
 /* Parse expression list. Last expression is left open. */
@@ -2146,9 +2250,9 @@ static void parse_for_num(LexState *ls, GCstr *varname, BCLine line)
   FuncScope bl;
   BCPos loop, loopend;
   /* Hidden control variables. */
-  var_new_lit(ls, FORL_IDX, "(for index)");
-  var_new_lit(ls, FORL_STOP, "(for limit)");
-  var_new_lit(ls, FORL_STEP, "(for step)");
+  var_new_fixed(ls, FORL_IDX, VARNAME_FOR_IDX);
+  var_new_fixed(ls, FORL_STOP, VARNAME_FOR_STOP);
+  var_new_fixed(ls, FORL_STEP, VARNAME_FOR_STEP);
   /* Visible copy of index variable. */
   var_new(ls, FORL_EXT, varname);
   lex_check(ls, '=');
@@ -2220,9 +2324,9 @@ static void parse_for_iter(LexState *ls, GCstr *indexname)
   FuncScope bl;
   int isnext;
   /* Hidden control variables. */
-  var_new_lit(ls, nvars++, "(for generator)");
-  var_new_lit(ls, nvars++, "(for state)");
-  var_new_lit(ls, nvars++, "(for control)");
+  var_new_fixed(ls, nvars++, VARNAME_FOR_GEN);
+  var_new_fixed(ls, nvars++, VARNAME_FOR_STATE);
+  var_new_fixed(ls, nvars++, VARNAME_FOR_CTL);
   /* Visible variables returned from iterator. */
   var_new(ls, nvars++, indexname);
   while (lex_opt(ls, ','))
@@ -2374,7 +2478,11 @@ GCproto *lj_parse(LexState *ls)
   FuncState fs;
   GCproto *pt;
   lua_State *L = ls->L;
+#ifdef LUAJIT_DISABLE_DEBUGINFO
+  ls->chunkname = lj_str_newlit(L, "=");
+#else
   ls->chunkname = lj_str_newz(L, ls->chunkarg);
+#endif
   setstrV(L, L->top, ls->chunkname);  /* Anchor chunkname string. */
   incr_top(L);
   ls->level = 0;

+ 1 - 1
src/lj_trace.c

@@ -98,7 +98,7 @@ static void perftools_addtrace(GCtrace *T)
   static FILE *fp;
   GCproto *pt = &gcref(T->startpt)->pt;
   const BCIns *startpc = mref(T->startpc, const BCIns);
-  const char *name = strdata(proto_chunkname(pt));
+  const char *name = proto_chunknamestr(pt);
   BCLine lineno;
   if (name[0] == '@' || name[0] == '=')
     name++;