bleak_house 3.6 → 3.7

data/CHANGELOG

@@ -1,4 +1,6 @@
 
+v3.7. Sample object contents. Restore Rails 1.2.x compatibility.
+
 v3.6. Rails 2.0.2 compatibility.
 
 v3.5.1. Update bundled Ruby to patchlevel 110. Delete your current ruby-bleak-house and reinstall.

data/Manifest

@@ -9,7 +9,7 @@ init.rb
 install.rb
 lib/bleak_house/analyzer/analyzer.rb
 lib/bleak_house/analyzer.rb
-lib/bleak_house/logger/mem_usage.rb
+lib/bleak_house/logger/source.rb
 lib/bleak_house/logger.rb
 lib/bleak_house/rails/action_controller.rb
 lib/bleak_house/rails/bleak_house.rb
@@ -39,7 +39,6 @@ test/integration/app/config/initializers/inflections.rb
 test/integration/app/config/initializers/mime_types.rb
 test/integration/app/config/routes.rb
 test/integration/app/doc/README_FOR_APP
-test/integration/app/log/bleak_house_production.dump
 test/integration/app/public/404.html
 test/integration/app/public/422.html
 test/integration/app/public/500.html

data/README

@@ -5,22 +5,24 @@ A library for finding memory leaks.
 
 == License
 
-Copyright 2007 Cloudburst, LLC. See the included LICENSE file. Portions copyright 2006 Eric Hodel and used with permission. See the included LICENSE_BSD file. 
+Copyright 2007, 2008 Cloudburst, LLC. Licensed under the AFL 3. See the included LICENSE file. Portions copyright 2006 Eric Hodel and used with permission. See the included LICENSE_BSD file. 
 
-The public certificate for this gem is at http://rubyforge.org/frs/download.php/25331/evan_weaver-original-public_cert.pem.
+The public certificate for this gem is here[link:http://rubyforge.org/frs/download.php/25331/evan_weaver-original-public_cert.pem]. 
+
+If you like this software, please {make a donation to charity}[link:http://blog.evanweaver.com/donate/] or {recommend Evan}[link:http://www.workingwithrails.com/person/7739-evan-weaver] at Working with Rails.
 
 == Features
 
 * leak-proof C instrumentation
-* fast logging
-* complete tracking of object and symbol history
+* tracks all objects and symbols
+* inspection of sample leaked objects (optional)
 * easy Rails integration
 
 == Requirements
 
 * A unix-like operating system
 * Ruby 1.8.6
-* Rails 2.0.2 or greater
+* Rails 1.2.x or 2.0.x
 
 = Usage
 
@@ -29,7 +31,7 @@ The public certificate for this gem is at http://rubyforge.org/frs/download.php/
 Install the gem:
   sudo gem install bleak_house
 
-The installation takes a long time because it compiles a patched Ruby binary for you. It is installed as <tt>ruby-bleak-house</tt> alongside your regular <tt>ruby</tt>binary.
+The installation takes a long time because it compiles a patched Ruby binary for you. It is installed as <tt>ruby-bleak-house</tt> alongside your regular <tt>ruby</tt> binary.
 
 Please see the forum ( http://rubyforge.org/forum/forum.php?forum_id=13983 ) if you have installation problems.
 
@@ -40,12 +42,18 @@ To setup a Rails app for profiling, just <tt>require 'bleak_house'</tt> in <tt>c
 Then, to engage the logger (ideally, in a live deployment situation), start a server instance as so:
   RAILS_ENV=production BLEAK_HOUSE=true ruby-bleak-house ./script/server 
 
-Browse around manually, thrash your entire app with a script, target troublesome controllers/actions, etc. The BleakHouse logger will dump a huge amount of data to <tt>log/bleak_house_production.dump</tt>--keep an eye on your disk space.
+Browse around manually, thrash your entire app with a script, target troublesome controllers/actions, etc. The BleakHouse logger will dump a huge amount of data to <tt>log/bleak_house_production.dump</tt>--keep an eye on your disk space. I usually aim for a 3GB dump.
 
 Then, to analyze the dump:
   bleak path/to/log/bleak_house_production.dump
   
-Be patient.
+Be patient; it's quite slow.
+
+== Sampling object contents
+
+BleakHouse can sample object contents, which might help you figure out the source of the leak. Just add <tt>SAMPLE_RATE=0.1</tt> to the environment string when you start your server. 
+
+Note that there is a chance this could _introduce_ leaks if you override <tt>inspect</tt> in leaky ways. Unfortunately, the samping is of dubious usefulness and really increases the memory usage of the analyze task. That's why it's off by default. 
 
 = Extras
 
@@ -61,7 +69,7 @@ You may get library require errors if you install <tt>ruby-bleak-house</tt> 1.8.
 
 == Reporting problems
 
-* http://rubyforge.org/forum/forum.php?forum_id=13983
+The support forum is here[http://rubyforge.org/forum/forum.php?forum_id=13983].
 
 Patches and contributions are very welcome. Please note that contributors are required to assign copyright for their additions to Cloudburst, LLC.
 

data/TODO

@@ -1,2 +1,3 @@
 
 * Better docs on what the tags mean
+* Use per-frame file storage to avoid out-of-memory problems

data/bin/bleak

@@ -8,5 +8,19 @@ if !ARGV[0]
 else
   $LOAD_PATH << "#{File.dirname(__FILE__)}/../lib/"
   require 'bleak_house/analyzer'
+
+  require 'ruby-debug' if ENV['DEBUG']
+
+  if ENV['WATCH_MEM']
+    Thread.new do
+      while sleep(10)
+        free = `free -m`.split("\n")[2].split(" ")[3].to_i
+        if free < 20
+          exit!
+        end
+      end
+    end
+  end
+
   BleakHouse::Analyzer.run(ARGV[0])
 end

data/bleak_house.gemspec

@@ -1,22 +1,22 @@
 
-# Gem::Specification for Bleak_house-3.6
+# Gem::Specification for Bleak_house-3.7
 # Originally generated by Echoe
 
 Gem::Specification.new do |s|
   s.name = %q{bleak_house}
-  s.version = "3.6"
+  s.version = "3.7"
 
   s.specification_version = 2 if s.respond_to? :specification_version=
 
   s.required_rubygems_version = Gem::Requirement.new(">= 0") if s.respond_to? :required_rubygems_version=
   s.authors = ["Evan Weaver"]
-  s.date = %q{2008-01-02}
+  s.date = %q{2008-01-17}
   s.default_executable = %q{bleak}
   s.description = %q{A library for finding memory leaks.}
   s.email = %q{}
   s.executables = ["bleak"]
   s.extensions = ["ext/bleak_house/logger/extconf.rb"]
-  s.files = ["bin/bleak", "CHANGELOG", "ext/bleak_house/logger/build_logger.rb", "ext/bleak_house/logger/build_ruby.rb", "ext/bleak_house/logger/extconf.rb", "ext/bleak_house/logger/snapshot.c", "ext/bleak_house/logger/snapshot.h", "init.rb", "install.rb", "lib/bleak_house/analyzer/analyzer.rb", "lib/bleak_house/analyzer.rb", "lib/bleak_house/logger/mem_usage.rb", "lib/bleak_house/logger.rb", "lib/bleak_house/rails/action_controller.rb", "lib/bleak_house/rails/bleak_house.rb", "lib/bleak_house/rails/dispatcher.rb", "lib/bleak_house/rails.rb", "lib/bleak_house/support/core_extensions.rb", "lib/bleak_house.rb", "LICENSE", "LICENSE_BSD", "Manifest", "README", "ruby/gc.c.patch", "ruby/parse.y.patch", "ruby/ruby-1.8.6-p110.tar.bz2", "test/integration/app/app/controllers/application.rb", "test/integration/app/app/controllers/items_controller.rb", "test/integration/app/app/helpers/application_helper.rb", "test/integration/app/app/helpers/items_helper.rb", "test/integration/app/app/views/items/index.rhtml", "test/integration/app/config/boot.rb", "test/integration/app/config/database.yml", "test/integration/app/config/environment.rb", "test/integration/app/config/environments/development.rb", "test/integration/app/config/environments/production.rb", "test/integration/app/config/environments/test.rb", "test/integration/app/config/initializers/inflections.rb", "test/integration/app/config/initializers/mime_types.rb", "test/integration/app/config/routes.rb", "test/integration/app/doc/README_FOR_APP", "test/integration/app/log/bleak_house_production.dump", "test/integration/app/public/404.html", "test/integration/app/public/422.html", "test/integration/app/public/500.html", "test/integration/app/public/dispatch.cgi", "test/integration/app/public/dispatch.fcgi", "test/integration/app/public/dispatch.rb", "test/integration/app/public/favicon.ico", "test/integration/app/public/images/rails.png", "test/integration/app/public/javascripts/application.js", "test/integration/app/public/javascripts/controls.js", "test/integration/app/public/javascripts/dragdrop.js", "test/integration/app/public/javascripts/effects.js", "test/integration/app/public/javascripts/prototype.js", "test/integration/app/public/robots.txt", "test/integration/app/Rakefile", "test/integration/app/README", "test/integration/app/script/about", "test/integration/app/script/console", "test/integration/app/script/destroy", "test/integration/app/script/generate", "test/integration/app/script/performance/benchmarker", "test/integration/app/script/performance/profiler", "test/integration/app/script/performance/request", "test/integration/app/script/plugin", "test/integration/app/script/process/inspector", "test/integration/app/script/process/reaper", "test/integration/app/script/process/spawner", "test/integration/app/script/runner", "test/integration/app/script/server", "test/integration/app/test/functional/items_controller_test.rb", "test/integration/app/test/test_helper.rb", "test/integration/server_test.rb", "test/misc/direct.rb", "test/test_helper.rb", "test/unit/test_bleak_house.rb", "TODO", "bleak_house.gemspec"]
+  s.files = ["bin/bleak", "CHANGELOG", "ext/bleak_house/logger/build_logger.rb", "ext/bleak_house/logger/build_ruby.rb", "ext/bleak_house/logger/extconf.rb", "ext/bleak_house/logger/snapshot.c", "ext/bleak_house/logger/snapshot.h", "init.rb", "install.rb", "lib/bleak_house/analyzer/analyzer.rb", "lib/bleak_house/analyzer.rb", "lib/bleak_house/logger/source.rb", "lib/bleak_house/logger.rb", "lib/bleak_house/rails/action_controller.rb", "lib/bleak_house/rails/bleak_house.rb", "lib/bleak_house/rails/dispatcher.rb", "lib/bleak_house/rails.rb", "lib/bleak_house/support/core_extensions.rb", "lib/bleak_house.rb", "LICENSE", "LICENSE_BSD", "Manifest", "README", "ruby/gc.c.patch", "ruby/parse.y.patch", "ruby/ruby-1.8.6-p110.tar.bz2", "test/integration/app/app/controllers/application.rb", "test/integration/app/app/controllers/items_controller.rb", "test/integration/app/app/helpers/application_helper.rb", "test/integration/app/app/helpers/items_helper.rb", "test/integration/app/app/views/items/index.rhtml", "test/integration/app/config/boot.rb", "test/integration/app/config/database.yml", "test/integration/app/config/environment.rb", "test/integration/app/config/environments/development.rb", "test/integration/app/config/environments/production.rb", "test/integration/app/config/environments/test.rb", "test/integration/app/config/initializers/inflections.rb", "test/integration/app/config/initializers/mime_types.rb", "test/integration/app/config/routes.rb", "test/integration/app/doc/README_FOR_APP", "test/integration/app/public/404.html", "test/integration/app/public/422.html", "test/integration/app/public/500.html", "test/integration/app/public/dispatch.cgi", "test/integration/app/public/dispatch.fcgi", "test/integration/app/public/dispatch.rb", "test/integration/app/public/favicon.ico", "test/integration/app/public/images/rails.png", "test/integration/app/public/javascripts/application.js", "test/integration/app/public/javascripts/controls.js", "test/integration/app/public/javascripts/dragdrop.js", "test/integration/app/public/javascripts/effects.js", "test/integration/app/public/javascripts/prototype.js", "test/integration/app/public/robots.txt", "test/integration/app/Rakefile", "test/integration/app/README", "test/integration/app/script/about", "test/integration/app/script/console", "test/integration/app/script/destroy", "test/integration/app/script/generate", "test/integration/app/script/performance/benchmarker", "test/integration/app/script/performance/profiler", "test/integration/app/script/performance/request", "test/integration/app/script/plugin", "test/integration/app/script/process/inspector", "test/integration/app/script/process/reaper", "test/integration/app/script/process/spawner", "test/integration/app/script/runner", "test/integration/app/script/server", "test/integration/app/test/functional/items_controller_test.rb", "test/integration/app/test/test_helper.rb", "test/integration/server_test.rb", "test/misc/direct.rb", "test/test_helper.rb", "test/unit/test_bleak_house.rb", "TODO", "bleak_house.gemspec"]
   s.has_rdoc = true
   s.homepage = %q{http://blog.evanweaver.com/files/doc/fauna/bleak_house/}
   s.require_paths = ["lib", "ext"]
@@ -25,7 +25,7 @@ Gem::Specification.new do |s|
   s.summary = %q{A library for finding memory leaks.}
   s.test_files = ["test/integration/server_test.rb", "test/unit/test_bleak_house.rb"]
 
-  s.add_dependency(%q<ccsv>, [">= 0"])
+  s.add_dependency(%q<ccsv>, [">= 0.1"])
 end
 
 
@@ -40,9 +40,10 @@ end
 #   p.summary = "A library for finding memory leaks."  
 #   p.url = "http://blog.evanweaver.com/files/doc/fauna/bleak_house/"  
 #   p.docs_host = 'blog.evanweaver.com:~/www/bax/public/files/doc/'
-#   p.dependencies = ['ccsv']  
+#   p.dependencies = ['ccsv >=0.1'] # 'memory >=0.0.2'
 #   p.require_signed = true
 #   
 #   p.rdoc_pattern = /^ext.*\.c|lib.*logger.*rb|analyzer|rails\/bleak_house|^README|^CHANGELOG|^TODO|^LICENSE|^COPYING$/
-#   p.test_pattern = ["test/integration/*.rb", "test/unit/*.rb"]  
+#   p.test_pattern = ["test/integration/*.rb", "test/unit/*.rb"]
+#   p.clean_pattern << "**/bleak_house*dump*"
 # end

data/ext/bleak_house/logger/build_ruby.rb

@@ -5,10 +5,15 @@ if RUBY_PLATFORM =~ /win32|windows/
   raise "Windows is not supported."
 end
 
+unless RUBY_VERSION == '1.8.6'
+  raise "Wrong Ruby version, you're at '#{RUBY_VERSION}', need 1.8.6"
+end
+
 source_dir = File.expand_path(File.dirname(__FILE__)) + "/../../../ruby"
 tmp = "/tmp/"
 
 require 'fileutils'
+require 'rbconfig'
 
 def which(basename)
   # system('which') is not compatible across Linux and BSD
@@ -17,7 +22,7 @@ def which(basename)
     path if File.exist? path
   end
 end
-  
+
 unless which('ruby-bleak-house')
   
   Dir.chdir(tmp) do
@@ -43,6 +48,29 @@ unless which('ruby-bleak-house')
           system("patch -p0 < \'#{source_dir}/gc.c.patch\' > ../gc.c.patch.log 2>&1")
           system("patch -p0 < \'#{source_dir}/parse.y.patch\' > ../parse.y.patch.log 2>&1")
           system("./configure --prefix=#{binary_dir[0..-5]} > ../configure.log 2>&1") # --with-static-linked-ext
+          
+          # Patch the makefile for arch/sitedir
+          makefile = File.read('Makefile')
+          %w{arch sitearch sitedir}.each do | key |
+            makefile.gsub!(/#{key} = .*/, "#{key} = #{Config::CONFIG[key]}")
+          end
+          File.open('Makefile', 'w'){|f| f.puts(makefile)}
+          
+          # Patch the config.h for constants
+          constants = {
+            'RUBY_LIB' => 'rubylibdir',          #define RUBY_LIB "/usr/lib/ruby/1.8"
+            'RUBY_SITE_LIB' => 'sitedir',        #define RUBY_SITE_LIB "/usr/lib/ruby/site_ruby"
+            'RUBY_SITE_LIB2' => 'sitelibdir',    #define RUBY_SITE_LIB2 "/usr/lib/ruby/site_ruby/1.8"
+            'RUBY_PLATFORM' => 'arch',           #define RUBY_PLATFORM "i686-linux"
+            'RUBY_ARCHLIB' => 'topdir',          #define RUBY_ARCHLIB "/usr/lib/ruby/1.8/i686-linux"
+            'RUBY_SITE_ARCHLIB' => 'sitearchdir' #define RUBY_SITE_ARCHLIB "/usr/lib/ruby/site_ruby/1.8/i686-linux"
+          }
+          config_h = File.read('config.h')
+          constants.each do | const, key |
+            config_h.gsub!(/#define #{const} .*/, "#define #{const} \"#{Config::CONFIG[key]}\"")
+          end
+          File.open('config.h', 'w'){|f| f.puts(config_h)}
+          
           system("make > ../make.log 2>&1")
   
           binary = "#{binary_dir}/ruby-bleak-house"

data/ext/bleak_house/logger/snapshot.c

@@ -15,13 +15,15 @@ static VALUE heaps_length(VALUE self) {
 }
 
 /* Count the live objects on the heap and in the symbol table and write a CSV frame to <tt>_logfile</tt>. Set <tt>_specials = true</tt> if you also want to count AST nodes and var scopes; otherwise, use <tt>false</tt>. Note that common classes in the CSV output are hashed to small integers in order to save space.*/
-static VALUE snapshot(VALUE self, VALUE _logfile, VALUE tag, VALUE _specials) {
+static VALUE snapshot(VALUE self, VALUE _logfile, VALUE tag, VALUE _specials, VALUE _sampler) {
   Check_Type(_logfile, T_STRING);
   Check_Type(tag, T_STRING);
 
   RVALUE *obj, *obj_end;
   st_table_entry *sym; 
   
+  /* printf("Requested: %f\n", rb_num2dbl(_sampler)); */
+  
   struct heaps_slot * heaps = rb_gc_heap_slots();
   struct st_table * sym_tbl = rb_parse_sym_tbl();
 
@@ -42,9 +44,9 @@ static VALUE snapshot(VALUE self, VALUE _logfile, VALUE tag, VALUE _specials) {
   fprintf(logfile, "-1,%li\n", time(0));
   
   /* get and write the memory usage */
-  VALUE mem = rb_funcall(self, rb_intern("mem_usage"), 0);
+  /* VALUE mem = rb_funcall(self, rb_intern("mem_usage"), 0);
   fprintf(logfile, "-2,%li\n", NUM2INT(RARRAY_PTR(mem)[0]));
-  fprintf(logfile, "-3,%li\n", NUM2INT(RARRAY_PTR(mem)[1]));
+  fprintf(logfile, "-3,%li\n", NUM2INT(RARRAY_PTR(mem)[1])); */
   
   int filled_slots = 0;
   int free_slots = 0;
@@ -81,17 +83,32 @@ static VALUE snapshot(VALUE self, VALUE _logfile, VALUE tag, VALUE _specials) {
               hashed = lookup_builtin(rb_obj_classname((VALUE)obj));
             }
         }
+        
         /* write to log */
-        if (hashed < 0) {
-          /* regular classname */
-          fprintf(logfile, "%s,%lu\n", rb_obj_classname((VALUE)obj), FIX2ULONG(rb_obj_id((VALUE)obj)));
-        } else {
-          /* builtins key */
-          if (specials || hashed < BUILTINS_SIZE) {
+        if (specials || hashed < BUILTINS_SIZE) {
+          /* write classname */
+          if (hashed < 0) {
+            /* regular classname */
+            fprintf(logfile, "%s", rb_obj_classname((VALUE)obj));
+          } else {
+            /* builtins key */
             /* 0 is not used for 'hashed' because Ruby's to_i turns any String into 0 */
-            fprintf(logfile, "%i,%lu\n", hashed + 1, FIX2ULONG(rb_obj_id((VALUE)obj)));
+            fprintf(logfile, "%i", hashed + 1); 
           }
-        }
+
+          /* write id */
+          fprintf(logfile, ",%lu", FIX2ULONG(rb_obj_id((VALUE)obj)));
+
+          /* write sample, if requested */
+          if (hashed < BUILTINS_SIZE) {
+            if (rand()/((double)RAND_MAX + 1) < rb_num2dbl(_sampler)) {
+              fprintf(logfile, ",%s", rb_rescue(inspect, (VALUE)obj, handle_exception, Qnil));
+            }
+          }
+          
+          /* write newline */
+          fprintf(logfile, "\n");
+        }        
       } else {
         free_slots ++;
       }
@@ -110,10 +127,45 @@ static VALUE snapshot(VALUE self, VALUE _logfile, VALUE tag, VALUE _specials) {
   fprintf(logfile, "-6,%i\n", free_slots);
   fclose(logfile);
     
-  rb_funcall(rb_mGC, rb_intern("start"), 0); /* request GC run */
+  for (i = 0; i < 3; i++) {
+    /* request GC run */
+    rb_funcall(rb_mGC, rb_intern("start"), 0); 
+  }
+  
   return Qtrue;
 }
 
+char * inspect(VALUE obj) {  
+  VALUE value;
+  char * string;
+  int i, length;
+  
+  value = rb_funcall((VALUE)obj, rb_intern("inspect"), 0);
+  string = StringValueCStr(value);
+  length = strlen(string);
+  
+  if (length > MAX_SAMPLE_LENGTH) {
+    string[MAX_SAMPLE_LENGTH] = '\0';
+    length = MAX_SAMPLE_LENGTH;
+  }
+  
+  /* Replace control characters */
+  for (i = 0; i < length; i++) {
+    if (string[i] == '\n') {
+      string[i] = ' ';
+    } if (string[i] == ',') {
+      string[i] = ' ';
+    }
+  }
+    
+  /* result = rb_funcall(result, rb_intern("gsub"), 2, rb_str_new2(","), rb_str_new2(",")); */
+  return string;
+}
+
+char * handle_exception(VALUE unused) {
+  return "";
+}
+
 int lookup_builtin(char * name) {
   int i;
   for (i = 0; i < BUILTINS_SIZE; i++) {
@@ -146,7 +198,7 @@ Init_snapshot()
 {
   rb_mBleakHouse = rb_define_module("BleakHouse");
   rb_cC = rb_define_class_under(rb_mBleakHouse, "Logger", rb_cObject);
-  rb_define_method(rb_cC, "snapshot", snapshot, 3);
+  rb_define_method(rb_cC, "snapshot", snapshot, 4);
   rb_define_method(rb_cC, "heaps_used", heaps_used, 0);
   rb_define_method(rb_cC, "heaps_length", heaps_length, 0);
 }

data/ext/bleak_house/logger/snapshot.h

@@ -50,6 +50,7 @@ static char * builtins[] = {
 
 #define BUILTINS_SIZE 30
 #define SPECIALS_SIZE 7
+#define MAX_SAMPLE_LENGTH 100
 
 typedef struct RVALUE {
     union {
@@ -96,4 +97,6 @@ struct heaps_slot * rb_gc_heap_slots();
 int rb_gc_heaps_used();
 int rb_gc_heaps_length();
 
+char * inspect(VALUE);
+char * handle_exception(VALUE);
 int lookup_builtin(char *);

data/lib/bleak_house/analyzer/analyzer.rb

@@ -1,247 +1,357 @@
 
 require 'ccsv'
+# require 'memory'
 require 'fileutils'
 require 'yaml'
 require 'pp'
 
 module BleakHouse
 
-  class Analyzer 
-  
-    MAGIC_KEYS = {
-      -1 => 'timestamp',
-      -2 => 'mem usage/swap',
-      -3 => 'mem usage/real',
-      -4 => 'tag',
-      -5 => 'heap/filled',
-      -6 => 'heap/free'
-    }    
-        
-    INITIAL_SKIP = 15 # XXX Might be better as a per-tag skip but that gets kinda complicated
-    
-    CLASS_KEYS = eval('[nil, ' + # Skip 0 so that the output of String#to_s is useful
-      open(
-        File.dirname(__FILE__) + '/../../../ext/bleak_house/logger/snapshot.h'
-      ).read[/\{(.*?)\}/m, 1] + ']')
-
-    def self.backwards_detect(array)
-      i = array.size - 1
-      while i >= 0
-        item = array[i]
-        return item if yield(item)
-        i -= 1
-      end
-    end   
-    
-    def self.calculate!(frame, index, total, obj_count = nil)
-      bsize = frame['births'].size
-      dsize = frame['deaths'].size
-      
-      # Avoid divide by zero errors
-      frame['meta']['ratio'] = ratio = (bsize - dsize) / (bsize + dsize + 1).to_f
-      frame['meta']['impact'] = begin
-        result = Math.log10((bsize - dsize).abs.to_i / 10.0)
-        raise Errno::ERANGE if result.nan? or result.infinite?
-        result
-      rescue Errno::ERANGE
-        0
-      end
-      
-      puts "  F#{index}:#{total} (#{index * 100 / total}%): #{frame['meta']['tag']} (#{obj_count.to_s + ' population, ' if obj_count}#{bsize} births, #{dsize} deaths, ratio #{format('%.2f', frame['meta']['ratio'])}, impact #{format('%.2f', frame['meta']['impact'])})"
+  class Analyzer
+
+    SPECIALS = {
+      -1 => :timestamp,
+      -2 => :'mem usage/swap', # Not used
+      -3 => :'mem usage/real',  # Not used
+      -4 => :tag,
+      -5 => :'heap/filled',
+      -6 => :'heap/free'
+    }
+
+    # Might be better as a per-tag skip but that gets kinda complicated
+    initial_skip = (ENV['INITIAL_SKIP'] || 15).to_i
+    INITIAL_SKIP = initial_skip < 2 ? 2 : initial_skip
+
+    DISPLAY_SAMPLES = (ENV['DISPLAY_SAMPLES'] || 5).to_i
+
+    class_key_source = File.dirname(__FILE__) + '/../../../ext/bleak_house/logger/snapshot.h'
+    class_key_string = open(class_key_source).read[/\{(.*?)\}/m, 1]
+    # Skip 0 so that the output of String#to_s is useful
+    CLASS_KEYS = eval("[nil, #{class_key_string} ]").map do |class_name|
+      class_name.to_sym if class_name
     end
-    
-    # Parses and correlates a BleakHouse::Logger output file.
-    def self.run(logfile)
-      logfile.chomp!(".cache")
-      cachefile = logfile + ".cache"
-
-      unless File.exists? logfile or File.exists? cachefile
-        puts "No data file found: #{logfile}"
-        exit 
+
+    class << self
+
+      def reverse_detect(array)
+        i = array.size - 1
+        while i >= 0
+          item = array[i]
+          return item if yield(item)
+          i -= 1
+        end
       end
-      
-      puts "Working..."
-      
-      frames = []
-      last_population = []
-      frame = nil
-      ix = nil
-      
-      if File.exist?(cachefile) and (!File.exists? logfile or File.stat(cachefile).mtime > File.stat(logfile).mtime)
-        # Cache is fresh
-        puts "Using cache"
-        frames = Marshal.load(File.open(cachefile).read)
-        puts "#{frames.size - 1} frames"        
-        frames[0..-2].each_with_index do |frame, index|
-          calculate!(frame, index + 1, frames.size - 1)
+
+      def calculate!(frame, index, total, population = nil)
+        bsize = frame[:births].size
+        dsize = frame[:deaths].size
+
+        # Avoid divide by zero errors
+        frame[:meta][:ratio] = ratio = (bsize - dsize) / (bsize + dsize + 1).to_f
+        frame[:meta][:impact] = begin
+          result = Math.log10((bsize - dsize).abs.to_i / 10.0)
+          raise Errno::ERANGE if result.nan? or result.infinite?
+          result
+        rescue Errno::ERANGE
+          0
         end
+
+        puts "  F#{index}:#{total} (#{index * 100 / total}%): #{frame[:meta][:tag]} (#{population.to_s + ' population, ' if population}#{bsize} births, #{dsize} deaths, ratio #{format('%.2f', frame[:meta][:ratio])}, impact #{format('%.2f', frame[:meta][:impact])})"
+      end
+
+      # Read a frames object from a cache file.
+      def read_cache(cachefile)
+        frames = Marshal.load(File.open(cachefile).read)[0..-2]        
+        total_frames = frames.size - 1
+        announce_total(total_frames)
         
-      else                        
-        # Rebuild frames
-        total_frames = `grep '^-1' #{logfile} | wc`.to_i - 2
-                
-        puts "#{total_frames} frames"     
+        frames[0..-2].each_with_index do |frame, index|
+          calculate!(frame, index + 1, total_frames)
+        end
+        frames
+      end
+      
+      def announce_total(total_frames)
+        puts "#{total_frames} frames"
 
         if total_frames < INITIAL_SKIP * 3
           puts "Not enough frames for accurate results. Please record at least #{INITIAL_SKIP * 3} frames."
           exit # Should be exit! but that messes up backticks capturing in the tests
         end
+      end      
+
+      # Cache an object to disk.
+      def write_cache(object, cachefile)
+        Thread.exclusive do
+          File.open(cachefile, 'w') do |f|
+            f.write Marshal.dump(object)
+          end
+        end
+      end
+
+      # Rebuild frames
+      def read(logfile, cachefile)
+        total_frames = `grep '^-1' #{logfile} | wc`.to_i - 2
+        announce_total(total_frames)        
         
-        Ccsv.foreach(logfile) do |row|                
-        
-          # Stupid is fast
-          i = row[0].to_i
-          row[0] = i if i != 0
-          i = row[1].to_i
-          row[1] = i if i != 0
-          
-          if row[0].to_i < 0          
+        frames = loop(logfile, cachefile, total_frames)
+      end
+
+      # Convert the class and id columns to Fixnums, if possible, and remove memory
+      # addresses from inspection samples.
+      def normalize_row(row)
+
+        # Broken out for speed (we don't want to generate a closure)
+        if (int = row[0].to_i) != 0
+          row[0] = int
+        else
+          row[0] = row[0].to_sym
+        end
+
+        if (int = row[1].to_i) != 0
+          row[1] = int
+        else
+          row[1] = row[1].to_sym
+        end
+
+        if row[2]
+          row[2] = row[2].gsub(/0x[\da-f]{8,16}/, "0xID").to_sym
+        end
+
+        row
+      end
+
+      # Inner loop of the raw log reader. The implementation is kind of a mess.
+      def loop(logfile, cachefile, total_frames)
+
+        frames = []
+        last_population = []
+        frame = nil
+
+        Ccsv.foreach(logfile) do |row|
+
+          class_index, id_or_tag, sampled_content = normalize_row(row)
+
+          # Check for frame headers
+          if class_index < 0
+
             # Get frame meta-information
-            if MAGIC_KEYS[row[0]] == 'timestamp'
-  
-              # The frame has ended; process the last one            
+            if SPECIALS[class_index] == :timestamp
+
+              # The frame has ended; process the last one
               if frame
-                population = frame['objects'].keys
+                population = frame[:objects].keys
                 births = population - last_population
                 deaths = last_population - population
                 last_population = population
-    
-                # assign births
-                frame['births'] = frame['objects'].slice(births).to_a # Work around a Marshal bug
-                
-                # assign deaths to previous frame
-                if final = frames[-2]
-                  final['deaths'] = final['objects'].slice(deaths).to_a # Work around a Marshal bug
-                  obj_count = final['objects'].size
-                  final.delete 'objects'
-                  calculate!(final, frames.size - 1, total_frames, obj_count)
+
+                # Assign births
+                frame[:births] = [] # Uses an Array to work around a Marshal bug
+                births.each do |key|
+                  frame[:births] << [key, frame[:objects][key]]
+                end
+
+                # Assign deaths to previous frame
+                final = frames[-2]
+                if final
+
+                  final[:deaths] = [] # Uses an Array to work around a Marshal bug
+                  deaths.each do |key|
+                    final[:deaths] << [key, [final[:objects][key].first]] # Don't need the sample content for deaths
+                  end
+
+                  # Try to reduce memory footprint
+                  final.delete :objects
+                  GC.start
+                  sleep 1 # Give the GC thread a chance to do something
+
+                  calculate!(final, frames.size - 1, total_frames, population.size)
                 end
               end
-  
+              
               # Set up a new frame
               frame = {}
               frames << frame
-              frame['objects'] ||= {}
-              frame['meta'] ||= {}
-              
-              #puts "  Frame #{frames.size} opened"
+              frame[:objects] ||= {}
+              frame[:meta] ||= {}
+
+              # Write out an in-process cache, in case you run out of RAM
+              if frames.size % 20 == 0
+                write_cache(frames, cachefile)
+              end              
             end
-            
-            frame['meta'][MAGIC_KEYS[row[0]]] = row[1]
+
+            frame[:meta][SPECIALS[class_index]] = id_or_tag
           else
-            # Assign live objects
-            frame['objects'][row[1]] = row[0]
+            # XXX Critical section
+            if sampled_content
+              # Normally object address strings and convert to a symbol
+              frame[:objects][id_or_tag] = [class_index, sampled_content]
+            else
+              frame[:objects][id_or_tag] = [class_index]
+            end
           end
+
         end
-        
-        frames = frames[0..-2]       
-        frames.last['objects'] = frames.last['objects'].to_a # Work around a Marshal bug x86-64
-        
-        # Cache the result
-        File.open(cachefile, 'w') do |f|
-          f.write Marshal.dump(frames)
-        end
-        
+
+        # Work around for a Marshal/Hash bug on x86_64
+        frames[-2][:objects] = frames[-2][:objects].to_a
+
+        # Write the cache
+        write_cache(frames, cachefile)
+
+        # Junk last frame (read_cache also does this)
+        frames[0..-2]
       end
-      
-      puts "\nRehashing."
-      
-      # Convert births back to hashes, necessary due to the Marshal workaround    
-      frames.each do |frame|
-        frame['births_hash'] = {}
-        frame['births'].each do |key, value|
-          frame['births_hash'][key] = value
+
+      # Convert births back to hashes, necessary due to the Marshal workaround
+      def rehash(frames)
+        frames.each do |frame|
+          frame[:births_hash] = {}
+          frame[:births].each do |key, value|
+            frame[:births_hash][key] = value
+          end
+          frame.delete(:births)
         end
-        frame.delete('births')
+        nil
       end
 
-#      require 'ruby-debug'; Debugger.start
-#      
-#      debugger
-                             
-      # See what objects are still laying around
-      population = frames.last['objects'].reject do |key, value|
-        frames.first['births_hash'][key] == value
-      end
+      # Parses and correlates a BleakHouse::Logger output file.
+      def run(logfile)
+        logfile.chomp!(".cache")
+        cachefile = logfile + ".cache"
 
-      puts "\n#{frames.size - 1} full frames. Removing #{INITIAL_SKIP} frames from each end of the run to account for\nstartup overhead and GC lag."
+        unless File.exists? logfile or File.exists? cachefile
+          puts "No data file found: #{logfile}"
+          exit
+        end
 
-      # Remove border frames
-      frames = frames[INITIAL_SKIP..-INITIAL_SKIP]
-      
-      total_births = frames.inject(0) do |births, frame|
-        births + frame['births_hash'].size
-      end
-      total_deaths = frames.inject(0) do |deaths, frame|
-        deaths + frame['deaths'].size
-      end
-      
-      puts "\n#{total_births} total births, #{total_deaths} total deaths, #{population.size} uncollected objects."
-      
-      leakers = {}
-      
-#      debugger
-      
-      # Find the sources of the leftover objects in the final population
-      population.each do |id, klass|
-        leaker = backwards_detect(frames) do |frame|
-          frame['births_hash'][id] == klass
-        end
-        if leaker
-#          debugger
-          tag = leaker['meta']['tag']
-          klass = CLASS_KEYS[klass] if klass.is_a? Fixnum
-          leakers[tag] ||= Hash.new(0)
-          leakers[tag][klass] += 1
+        puts "Working..."
+
+        frames = []
+
+        if File.exist?(cachefile) and (!File.exists? logfile or File.stat(cachefile).mtime > File.stat(logfile).mtime)
+          puts "Using cache"
+          frames = read_cache(cachefile)
+        else
+          frames = read(logfile, cachefile)
         end
-      end      
-      
-      # Sort
-      leakers = leakers.map do |tag, value| 
-        [tag, value.sort_by do |klass, count| 
-          -count
-        end]
-      end.sort_by do |tag, value|
-        Hash[*value.flatten].values.inject(0) {|i, v| i - v}
-      end
-      
-      if leakers.any?
-        puts "\nTags sorted by persistent uncollected objects. These objects did not exist at\nstartup, were instantiated by the associated tags, and were never garbage\ncollected:"
-        leakers.each do |tag, value|
-          requests = frames.select do |frame|
-            frame['meta']['tag'] == tag
-          end.size
-          puts "  #{tag} leaked (over #{requests} requests):"
-          value.each do |klass, count|
-            puts "    #{count} #{klass}"
+
+        puts "\nRehashing."
+
+        rehash(frames)
+
+        # See what objects are still laying around
+        population = frames.last[:objects].reject do |key, value|
+          frames.first[:births_hash][key] and frames.first[:births_hash][key].first == value.first
+        end
+
+        puts "\n#{frames.size - 1} full frames. Removing #{INITIAL_SKIP} frames from each end of the run to account for\nstartup overhead and GC lag."
+
+        # Remove border frames
+        frames = frames[INITIAL_SKIP..-INITIAL_SKIP]
+
+        # Sum all births
+        total_births = frames.inject(0) do |births, frame|
+          births + frame[:births_hash].size
+        end
+
+        # Sum all deaths
+        total_deaths = frames.inject(0) do |deaths, frame|
+          deaths + frame[:deaths].size
+        end
+
+        puts "\n#{total_births} total births, #{total_deaths} total deaths, #{population.size} uncollected objects."
+
+        leakers = {}
+
+        # Find the sources of the leftover objects in the final population
+        population.each do |id, value|
+          klass = value[0]
+          content = value[1]
+          leaker = reverse_detect(frames) do |frame|
+            frame[:births_hash][id] and frame[:births_hash][id].first == klass
+          end
+          if leaker
+            tag = leaker[:meta][:tag]
+            klass = CLASS_KEYS[klass] if klass.is_a? Fixnum
+            leakers[tag] ||= Hash.new()
+            leakers[tag][klass] ||= {:count => 0, :contents => []}
+            leakers[tag][klass][:count] += 1
+            leakers[tag][klass][:contents] << content if content
           end
         end
-      else
-        puts "\nNo persistent uncollected objects found for any tags."
-      end
-            
-      impacts = {}
-      
-      frames.each do |frame|
-        impacts[frame['meta']['tag']] ||= []
-        impacts[frame['meta']['tag']] << frame['meta']['impact'] * frame['meta']['ratio']
-      end      
-      impacts = impacts.map do |tag, values|
-        [tag, values.inject(0) {|acc, i| acc + i} / values.size.to_f]
-      end.sort_by do |tag, impact|
-        impact.nan? ? 0 : -impact
-      end
-      
-      puts "\nTags sorted by average impact * ratio. Impact is the log10 of the size of the"
-      puts "change in object count for a frame:"
-      
-      impacts.each do |tag, total|
-        puts "  #{format('%.4f', total).rjust(7)}: #{tag}"
-      end
 
-      puts "\nDone"
+        # Sort the leakers
+        leakers = leakers.map do |tag, value|
+          # Sort leakiest classes within each tag
+          [tag, value.sort_by do |klass, hash|
+            -hash[:count]
+          end]
+        end.sort_by do |tag, value|
+          # Sort leakiest tags as a whole
+          Hash[*value.flatten].values.inject(0) {|i, hash| i - hash[:count]}
+        end
+
+        if leakers.any?
+          puts "\nTags sorted by persistent uncollected objects. These objects did not exist at\nstartup, were instantiated by the associated tags during the run, and were\nnever garbage collected:"
+          leakers.each do |tag, value|
+            requests = frames.select do |frame|
+              frame[:meta][:tag] == tag
+            end.size
+            puts "  #{tag} leaked per request (#{requests}):"
+            value.each do |klass, hash|
+              puts "    #{sprintf('%.1f', hash[:count] / requests.to_f)} #{klass}"
+
+              # Extract most common samples
+              contents = begin
+                hist = Hash.new(0)
+                hash[:contents].each do |content|
+                  hist[content] += 1
+                end
+                hist.sort_by do |content, count|
+                  -count
+                end[0..DISPLAY_SAMPLES].select do |content, count|
+                  ENV['DISPLAY_SAMPLES'] or count > 5
+                end
+              end
 
-    end    
-    
+              if contents.any?
+                puts "      Inspection samples:"
+                contents.each do |content, count|
+                  puts "        #{sprintf('%.1f', count / requests.to_f)} #{content}"
+                end
+              end
+
+            end
+          end
+        else
+          puts "\nNo persistent uncollected objects found for any tags."
+        end
+
+        impacts = {}
+
+        frames.each do |frame|
+          impacts[frame[:meta][:tag]] ||= []
+          impacts[frame[:meta][:tag]] << frame[:meta][:impact] * frame[:meta][:ratio]
+        end
+        impacts = impacts.map do |tag, values|
+          [tag, values.inject(0) {|acc, i| acc + i} / values.size.to_f]
+        end.sort_by do |tag, impact|
+          impact.nan? ? 0 : -impact
+        end
+
+        puts "\nTags sorted by average impact * ratio. Impact is the log10 of the size of the"
+        puts "change in object count for a frame:"
+
+        impacts.each do |tag, total|
+          puts "  #{format('%.4f', total).rjust(7)}: #{tag}"
+        end
+
+        puts "\nDone"
+
+      end
+
+    end
   end
 end