heap_dump 0.0.27 → 0.0.28

data/README.md

@@ -1,8 +1,9 @@
 # HeapDump
 
-Low-level ruby heap memory dump - including data and code references.
-Originally written across ruby 1.9.2-p290 data structures.
+Low-level ruby heap memory dump - including data and code references, useful for finding leaks.
+Has no performance overhead while not active, so can be used in production environment.
 
+Originally written across ruby 1.9.2-p290 data structures.
 Does work on other 1.9.2s and 1.9.3, but not well-tested yet(output is not proven to be as full etc.).
 
 Currently is under development and output format may differ.
@@ -35,6 +36,10 @@ Json contains one object per line, thus can be easily grepped.
 
 ### Injecting into live process via GDB
 
+For long-running applications common and recommended usecase is to have some kind of custom action or signal handler in app itself that invokes dump.
+
+But also dump can be invoked from gdb.
+
 run gdb
 
 ```bash
@@ -51,7 +56,8 @@ call heapdump_dump("mydump.json")
 
 or `call heapdump_dump(0)`, filename defaults to dump.json.
 
-Note that ruby-internal and yajl-ruby gems should be available to process this being injected into.
+Note that yajl-ruby gem (and heap_dump itself) should be available to process this being injected into.
+Also on rare ocassions process(for example if gdb attached while a signal/gc) may crash after and even during dumping, so safer way is to embed it in advance, there's no performance overhead.
 
 ### Importing dump in MongoDB
 

data/ext/heap_dump/heap_dump.c

@@ -1,4 +1,5 @@
 #include "ruby.h"
+#include "ruby/encoding.h"
 #include <stdlib.h>
 #include <stdio.h>
 
@@ -107,8 +108,8 @@ static void flush_yajl(walk_ctx_t *ctx){
 
 static inline int is_in_heap(void *ptr, void* osp);
 
-static inline const char* rb_builtin_type(VALUE obj){
-  switch(BUILTIN_TYPE(obj)){
+static inline const char* rb_type_str(int type){
+  switch(type){
     #define T(t) case t: return #t;
     T(T_NONE); T(T_NIL);
     T(T_OBJECT); T(T_CLASS); T(T_ICLASS); T(T_MODULE);
@@ -124,9 +125,16 @@ static inline const char* rb_builtin_type(VALUE obj){
     T(T_NODE); // code?
     T(T_ZOMBIE);
     #undef T
+    default:
+      return "_unknown_type_";
   }
 }
 
+static inline const char* rb_builtin_type(VALUE obj){
+  //NOTE: only for heap objects, on embedded use (TYPE(...))
+  return rb_type_str(BUILTIN_TYPE(obj));
+}
+
 #define true 1
 #define false 0
 
@@ -185,7 +193,14 @@ static void yg_id1(VALUE obj, walk_ctx_t* ctx){
   if(BUILTIN_TYPE(obj) == T_STRING && (!(RBASIC(obj)->flags & RSTRING_NOEMBED))){
     //printf("embedded string\n");
     //yajl_gen_null(ctx->yajl);
-    yg_rstring(obj);
+
+    if(rb_enc_get_index(obj) == rb_usascii_encindex())
+      yg_rstring(obj);
+    else{
+      //FIXME: convert encoding/safe syms etc?
+      //yg_cstring("(encoded string)");
+      yg_rstring(obj);
+    }
     return;
   }
 
@@ -444,7 +459,36 @@ static inline void dump_node(NODE* obj, walk_ctx_t *ctx){
 
 static int
 dump_keyvalue(st_data_t key, st_data_t value, walk_ctx_t *ctx){
-    yg_id((VALUE)key);
+    if ((VALUE)key == Qundef){
+      printf("undef key!\n");
+      // return ST_CONTINUE;
+    }
+
+    if(!key || (VALUE)key == Qnil){
+      yg_cstring("___null_key___"); //TODO: just ""?
+    } else {
+      //TODO: keys must be strings
+      const int type = TYPE(key);
+      if(type == T_SYMBOL || type == T_STRING && (!(RBASIC(key)->flags & RSTRING_NOEMBED)))
+        yg_id((VALUE)key);
+      else
+        {
+          char buf[128];
+          buf[sizeof(buf)-1] = 0;
+          switch(type){
+            case T_FIXNUM:
+              snprintf(buf, sizeof(buf)-1, "%d", FIX2INT(key));
+              break;
+            case T_FLOAT:
+              snprintf(buf, sizeof(buf)-1, "%lg", NUM2DBL(key));
+              break;
+            default:
+              snprintf(buf, sizeof(buf)-1, "__id_%ld", NUM2LONG(rb_obj_id(key)));
+              break;
+          }
+          yg_cstring(buf);
+        }
+      }
     yg_id((VALUE)value);
     return ST_CONTINUE;
 }
@@ -512,7 +556,7 @@ static int dump_iv_entry(ID key, VALUE value, walk_ctx_t *ctx){
   if(key_str)
     yg_cstring(key_str);
   else{
-    printf("Null key! %d\n", key);
+    // printf("Null key! %d\n", key);
     yg_cstring("___null_key___");
     //yg_null();
   }
@@ -935,13 +979,32 @@ static inline void walk_live_object(VALUE obj, walk_ctx_t *ctx){
       break;
     case T_STRING:
       //TODO: limit string len!
+      {
+      int enc_i = rb_enc_get_index(obj);
+      rb_encoding* enc = rb_enc_from_index(enc_i);
+      if(enc){
+        ygh_cstring("encoding", enc->name);
+      }
+      //FIXME: convert encoding and dump?
+      //if(enc_i == rb_usascii_encindex())
+      //this produces warnings on dump read, but recoverable
       ygh_string("val", RSTRING_PTR(obj), (unsigned int)RSTRING_LEN(obj));
+      }
       break;
     case T_SYMBOL:
       ygh_cstring("val", rb_id2name(SYM2ID(obj)));
       break;
     case T_REGEXP:
+      {
+      int enc_i = rb_enc_get_index(obj);
+      rb_encoding* enc = rb_enc_from_index(enc_i);
+      if(enc){
+        ygh_cstring("encoding", enc->name);
+      }
+      //FIXME: encodings?
+      // if(enc_i == rb_usascii_encindex())
       ygh_string("val", RREGEXP_SRC_PTR(obj), (unsigned int)RREGEXP_SRC_LEN(obj));
+      }
       break;
     // T(T_MATCH);
 
@@ -1494,7 +1557,7 @@ void heapdump_dump(const char* filename){
 
   dump_machine_context(ctx);
   flush_yajl(ctx);
-  fprintf(ctx->file, "\n");
+  // fprintf(ctx->file, "\n");
 
   struct gc_list *list;
   /* mark protected global variables */
@@ -1521,6 +1584,7 @@ void heapdump_dump(const char* filename){
 
   yajl_gen_map_close(ctx->yajl); //id:roots
   flush_yajl(ctx);
+  fprintf(ctx->file, "\n");
 
   //now dump all live objects
   printf("starting objspace walk\n");

data/ext/heap_dump/ruby_io.h

@@ -11,8 +11,8 @@ typedef struct {
 
 //#include "encoding.h" // and this also (rb_encoding + rb_econv_t)
 //FIXME: nasty:
-typedef int rb_encoding;
-typedef struct rb_econv rb_econv_t;
+// typedef int rb_encoding;
+// typedef struct rb_econv rb_econv_t;
 
 typedef struct rb_io_t {
     int fd;                     /* file descriptor */

data/lib/heap_dump/version.rb

@@ -1,3 +1,3 @@
 module HeapDump
-  VERSION = "0.0.27"
+  VERSION = "0.0.28"
 end

metadata

@@ -1,7 +1,7 @@
 --- !ruby/object:Gem::Specification
 name: heap_dump
 version: !ruby/object:Gem::Version
-  version: 0.0.27
+  version: 0.0.28
   prerelease: 
 platform: ruby
 authors:
@@ -9,7 +9,7 @@ authors:
 autorequire: 
 bindir: bin
 cert_chain: []
-date: 2012-11-10 00:00:00.000000000 Z
+date: 2012-11-12 00:00:00.000000000 Z
 dependencies:
 - !ruby/object:Gem::Dependency
   name: debugger-ruby_core_source
@@ -105,7 +105,7 @@ required_rubygems_version: !ruby/object:Gem::Requirement
       version: '0'
       segments:
       - 0
-      hash: -130654953779884601
+      hash: -3488312299511292318
 requirements: []
 rubyforge_project: 
 rubygems_version: 1.8.24