stackprof 0.2.14 → 0.2.18

data/ext/stackprof/stackprof.c

@@ -7,20 +7,32 @@
 **********************************************************************/
 
 #include <ruby/ruby.h>
+#include <ruby/version.h>
 #include <ruby/debug.h>
 #include <ruby/st.h>
 #include <ruby/io.h>
 #include <ruby/intern.h>
 #include <signal.h>
 #include <sys/time.h>
+#include <time.h>
 #include <pthread.h>
 
 #define BUF_SIZE 2048
+#define MICROSECONDS_IN_SECOND 1000000
+#define NANOSECONDS_IN_SECOND 1000000000
 
 #define FAKE_FRAME_GC    INT2FIX(0)
 #define FAKE_FRAME_MARK  INT2FIX(1)
 #define FAKE_FRAME_SWEEP INT2FIX(2)
 
+/*
+ * As of Ruby 3.0, it should be safe to read stack frames at any time
+ * See https://github.com/ruby/ruby/commit/0e276dc458f94d9d79a0f7c7669bde84abe80f21
+ */
+#if RUBY_API_VERSION_MAJOR < 3
+  #define USE_POSTPONED_JOB
+#endif
+
 static const char *fake_frame_cstrs[] = {
 	"(garbage collection)",
 	"(marking)",
@@ -29,6 +41,47 @@ static const char *fake_frame_cstrs[] = {
 
 #define TOTAL_FAKE_FRAMES (sizeof(fake_frame_cstrs) / sizeof(char *))
 
+#ifdef _POSIX_MONOTONIC_CLOCK
+  #define timestamp_t timespec
+  typedef struct timestamp_t timestamp_t;
+
+  static void capture_timestamp(timestamp_t *ts) {
+      clock_gettime(CLOCK_MONOTONIC, ts);
+  }
+
+  static int64_t delta_usec(timestamp_t *start, timestamp_t *end) {
+      int64_t result = MICROSECONDS_IN_SECOND * (end->tv_sec - start->tv_sec);
+      if (end->tv_nsec < start->tv_nsec) {
+	  result -= MICROSECONDS_IN_SECOND;
+	  result += (NANOSECONDS_IN_SECOND + end->tv_nsec - start->tv_nsec) / 1000;
+      } else {
+	  result += (end->tv_nsec - start->tv_nsec) / 1000;
+      }
+      return result;
+  }
+
+  static uint64_t timestamp_usec(timestamp_t *ts) {
+      return (MICROSECONDS_IN_SECOND * ts->tv_sec) + (ts->tv_nsec / 1000);
+  }
+#else
+  #define timestamp_t timeval
+  typedef struct timestamp_t timestamp_t;
+
+  static void capture_timestamp(timestamp_t *ts) {
+      gettimeofday(ts, NULL);
+  }
+
+  static int64_t delta_usec(timestamp_t *start, timestamp_t *end) {
+      struct timeval diff;
+      timersub(end, start, &diff);
+      return (MICROSECONDS_IN_SECOND * diff.tv_sec) + diff.tv_usec;
+  }
+
+  static uint64_t timestamp_usec(timestamp_t *ts) {
+      return (MICROSECONDS_IN_SECOND * ts.tv_sec) + diff.tv_usec
+  }
+#endif
+
 typedef struct {
     size_t total_samples;
     size_t caller_samples;
@@ -37,6 +90,11 @@ typedef struct {
     st_table *lines;
 } frame_data_t;
 
+typedef struct {
+    uint64_t timestamp_usec;
+    int64_t delta_usec;
+} sample_time_t;
+
 static struct {
     int running;
     int raw;
@@ -46,16 +104,17 @@ static struct {
     VALUE interval;
     VALUE out;
     VALUE metadata;
+    int ignore_gc;
 
     VALUE *raw_samples;
     size_t raw_samples_len;
     size_t raw_samples_capa;
     size_t raw_sample_index;
 
-    struct timeval last_sample_at;
-    int *raw_timestamp_deltas;
-    size_t raw_timestamp_deltas_len;
-    size_t raw_timestamp_deltas_capa;
+    struct timestamp_t last_sample_at;
+    sample_time_t *raw_sample_times;
+    size_t raw_sample_times_len;
+    size_t raw_sample_times_capa;
 
     size_t overall_signals;
     size_t overall_samples;
@@ -67,14 +126,17 @@ static struct {
 
     VALUE fake_frame_names[TOTAL_FAKE_FRAMES];
     VALUE empty_string;
+
+    int buffer_count;
+    sample_time_t buffer_time;
     VALUE frames_buffer[BUF_SIZE];
     int lines_buffer[BUF_SIZE];
 } _stackprof;
 
 static VALUE sym_object, sym_wall, sym_cpu, sym_custom, sym_name, sym_file, sym_line;
 static VALUE sym_samples, sym_total_samples, sym_missed_samples, sym_edges, sym_lines;
-static VALUE sym_version, sym_mode, sym_interval, sym_raw, sym_metadata, sym_frames, sym_out, sym_aggregate, sym_raw_timestamp_deltas;
-static VALUE sym_state, sym_marking, sym_sweeping;
+static VALUE sym_version, sym_mode, sym_interval, sym_raw, sym_metadata, sym_frames, sym_ignore_gc, sym_out;
+static VALUE sym_aggregate, sym_raw_sample_timestamps, sym_raw_timestamp_deltas, sym_state, sym_marking, sym_sweeping;
 static VALUE sym_gc_samples, objtracer;
 static VALUE gc_hook;
 static VALUE rb_mStackProf;
@@ -88,6 +150,7 @@ stackprof_start(int argc, VALUE *argv, VALUE self)
     struct sigaction sa;
     struct itimerval timer;
     VALUE opts = Qnil, mode = Qnil, interval = Qnil, metadata = rb_hash_new(), out = Qfalse;
+    int ignore_gc = 0;
     int raw = 0, aggregate = 1;
 
     if (_stackprof.running)
@@ -99,6 +162,9 @@ stackprof_start(int argc, VALUE *argv, VALUE self)
 	mode = rb_hash_aref(opts, sym_mode);
 	interval = rb_hash_aref(opts, sym_interval);
 	out = rb_hash_aref(opts, sym_out);
+	if (RTEST(rb_hash_aref(opts, sym_ignore_gc))) {
+	    ignore_gc = 1;
+	}
 
 	VALUE metadata_val = rb_hash_aref(opts, sym_metadata);
 	if (RTEST(metadata_val)) {
@@ -115,6 +181,10 @@ stackprof_start(int argc, VALUE *argv, VALUE self)
     }
     if (!RTEST(mode)) mode = sym_wall;
 
+    if (!NIL_P(interval) && (NUM2INT(interval) < 1 || NUM2INT(interval) >= MICROSECONDS_IN_SECOND)) {
+        rb_raise(rb_eArgError, "interval is a number of microseconds between 1 and 1 million");
+    }
+
     if (!_stackprof.frames) {
 	_stackprof.frames = st_init_numtable();
 	_stackprof.overall_signals = 0;
@@ -151,11 +221,12 @@ stackprof_start(int argc, VALUE *argv, VALUE self)
     _stackprof.aggregate = aggregate;
     _stackprof.mode = mode;
     _stackprof.interval = interval;
+    _stackprof.ignore_gc = ignore_gc;
     _stackprof.metadata = metadata;
     _stackprof.out = out;
 
     if (raw) {
-	gettimeofday(&_stackprof.last_sample_at, NULL);
+	capture_timestamp(&_stackprof.last_sample_at);
     }
 
     return Qtrue;
@@ -190,13 +261,19 @@ stackprof_stop(VALUE self)
     return Qtrue;
 }
 
+#if SIZEOF_VOIDP == SIZEOF_LONG
+#  define PTR2NUM(x) (LONG2NUM((long)(x)))
+#else
+#  define PTR2NUM(x) (LL2NUM((LONG_LONG)(x)))
+#endif
+
 static int
 frame_edges_i(st_data_t key, st_data_t val, st_data_t arg)
 {
     VALUE edges = (VALUE)arg;
 
     intptr_t weight = (intptr_t)val;
-    rb_hash_aset(edges, rb_obj_id((VALUE)key), INT2FIX(weight));
+    rb_hash_aset(edges, PTR2NUM(key), INT2FIX(weight));
     return ST_CONTINUE;
 }
 
@@ -223,7 +300,7 @@ frame_i(st_data_t key, st_data_t val, st_data_t arg)
     VALUE name, file, edges, lines;
     VALUE line;
 
-    rb_hash_aset(results, rb_obj_id(frame), details);
+    rb_hash_aset(results, PTR2NUM(frame), details);
 
     if (FIXNUM_P(frame)) {
 	name = _stackprof.fake_frame_names[FIX2INT(frame)];
@@ -284,6 +361,8 @@ stackprof_results(int argc, VALUE *argv, VALUE self)
     rb_hash_aset(results, sym_missed_samples, SIZET2NUM(_stackprof.overall_signals - _stackprof.overall_samples));
     rb_hash_aset(results, sym_metadata, _stackprof.metadata);
 
+    _stackprof.metadata = Qnil;
+
     frames = rb_hash_new();
     rb_hash_aset(results, sym_frames, frames);
     st_foreach(_stackprof.frames, frame_i, (st_data_t)frames);
@@ -293,7 +372,7 @@ stackprof_results(int argc, VALUE *argv, VALUE self)
 
     if (_stackprof.raw && _stackprof.raw_samples_len) {
 	size_t len, n, o;
-	VALUE raw_timestamp_deltas;
+	VALUE raw_sample_timestamps, raw_timestamp_deltas;
 	VALUE raw_samples = rb_ary_new_capa(_stackprof.raw_samples_len);
 
 	for (n = 0; n < _stackprof.raw_samples_len; n++) {
@@ -301,7 +380,7 @@ stackprof_results(int argc, VALUE *argv, VALUE self)
 	    rb_ary_push(raw_samples, SIZET2NUM(len));
 
 	    for (o = 0, n++; o < len; n++, o++)
-		rb_ary_push(raw_samples, rb_obj_id(_stackprof.raw_samples[n]));
+		rb_ary_push(raw_samples, PTR2NUM(_stackprof.raw_samples[n]));
 	    rb_ary_push(raw_samples, SIZET2NUM((size_t)_stackprof.raw_samples[n]));
 	}
 
@@ -313,17 +392,20 @@ stackprof_results(int argc, VALUE *argv, VALUE self)
 
 	rb_hash_aset(results, sym_raw, raw_samples);
 
-	raw_timestamp_deltas = rb_ary_new_capa(_stackprof.raw_timestamp_deltas_len);
+	raw_sample_timestamps = rb_ary_new_capa(_stackprof.raw_sample_times_len);
+	raw_timestamp_deltas = rb_ary_new_capa(_stackprof.raw_sample_times_len);
 
-	for (n = 0; n < _stackprof.raw_timestamp_deltas_len; n++) {
-	    rb_ary_push(raw_timestamp_deltas, INT2FIX(_stackprof.raw_timestamp_deltas[n]));
+	for (n = 0; n < _stackprof.raw_sample_times_len; n++) {
+	    rb_ary_push(raw_sample_timestamps, ULL2NUM(_stackprof.raw_sample_times[n].timestamp_usec));
+	    rb_ary_push(raw_timestamp_deltas, LL2NUM(_stackprof.raw_sample_times[n].delta_usec));
 	}
 
-	free(_stackprof.raw_timestamp_deltas);
-	_stackprof.raw_timestamp_deltas = NULL;
-	_stackprof.raw_timestamp_deltas_len = 0;
-	_stackprof.raw_timestamp_deltas_capa = 0;
+	free(_stackprof.raw_sample_times);
+	_stackprof.raw_sample_times = NULL;
+	_stackprof.raw_sample_times_len = 0;
+	_stackprof.raw_sample_times_capa = 0;
 
+	rb_hash_aset(results, sym_raw_sample_timestamps, raw_sample_timestamps);
 	rb_hash_aset(results, sym_raw_timestamp_deltas, raw_timestamp_deltas);
 
 	_stackprof.raw = 0;
@@ -403,14 +485,14 @@ st_numtable_increment(st_table *table, st_data_t key, size_t increment)
 }
 
 void
-stackprof_record_sample_for_stack(int num, int timestamp_delta)
+stackprof_record_sample_for_stack(int num, uint64_t sample_timestamp, int64_t timestamp_delta)
 {
     int i, n;
     VALUE prev_frame = Qnil;
 
     _stackprof.overall_samples++;
 
-    if (_stackprof.raw) {
+    if (_stackprof.raw && num > 0) {
 	int found = 0;
 
 	/* If there's no sample buffer allocated, then allocate one.  The buffer
@@ -462,20 +544,23 @@ stackprof_record_sample_for_stack(int num, int timestamp_delta)
 	}
 
 	/* If there's no timestamp delta buffer, allocate one */
-	if (!_stackprof.raw_timestamp_deltas) {
-	    _stackprof.raw_timestamp_deltas_capa = 100;
-	    _stackprof.raw_timestamp_deltas = malloc(sizeof(int) * _stackprof.raw_timestamp_deltas_capa);
-	    _stackprof.raw_timestamp_deltas_len = 0;
+	if (!_stackprof.raw_sample_times) {
+	    _stackprof.raw_sample_times_capa = 100;
+	    _stackprof.raw_sample_times = malloc(sizeof(sample_time_t) * _stackprof.raw_sample_times_capa);
+	    _stackprof.raw_sample_times_len = 0;
 	}
 
 	/* Double the buffer size if it's too small */
-	while (_stackprof.raw_timestamp_deltas_capa <= _stackprof.raw_timestamp_deltas_len + 1) {
-	    _stackprof.raw_timestamp_deltas_capa *= 2;
-	    _stackprof.raw_timestamp_deltas = realloc(_stackprof.raw_timestamp_deltas, sizeof(int) * _stackprof.raw_timestamp_deltas_capa);
+	while (_stackprof.raw_sample_times_capa <= _stackprof.raw_sample_times_len + 1) {
+	    _stackprof.raw_sample_times_capa *= 2;
+	    _stackprof.raw_sample_times = realloc(_stackprof.raw_sample_times, sizeof(sample_time_t) * _stackprof.raw_sample_times_capa);
 	}
 
-	/* Store the time delta (which is the amount of time between samples) */
-	_stackprof.raw_timestamp_deltas[_stackprof.raw_timestamp_deltas_len++] = timestamp_delta;
+	/* Store the time delta (which is the amount of microseconds between samples). */
+	_stackprof.raw_sample_times[_stackprof.raw_sample_times_len++] = (sample_time_t) {
+	    .timestamp_usec = sample_timestamp,
+	    .delta_usec = timestamp_delta,
+        };
     }
 
     for (i = 0; i < num; i++) {
@@ -508,48 +593,59 @@ stackprof_record_sample_for_stack(int num, int timestamp_delta)
     }
 
     if (_stackprof.raw) {
-	gettimeofday(&_stackprof.last_sample_at, NULL);
+	capture_timestamp(&_stackprof.last_sample_at);
     }
 }
 
+// buffer the current profile frames
+// This must be async-signal-safe
+// Returns immediately if another set of frames are already in the buffer
 void
-stackprof_record_sample()
+stackprof_buffer_sample(void)
 {
-    int timestamp_delta = 0;
+    if (_stackprof.buffer_count > 0) {
+	// Another sample is already pending
+	return;
+    }
+
+    uint64_t start_timestamp = 0;
+    int64_t timestamp_delta = 0;
     int num;
     if (_stackprof.raw) {
-	struct timeval t;
-	struct timeval diff;
-	gettimeofday(&t, NULL);
-	timersub(&t, &_stackprof.last_sample_at, &diff);
-	timestamp_delta = (1000 * diff.tv_sec) + diff.tv_usec;
+	struct timestamp_t t;
+	capture_timestamp(&t);
+	start_timestamp = timestamp_usec(&t);
+	timestamp_delta = delta_usec(&t, &_stackprof.last_sample_at);
     }
+
     num = rb_profile_frames(0, sizeof(_stackprof.frames_buffer) / sizeof(VALUE), _stackprof.frames_buffer, _stackprof.lines_buffer);
-    stackprof_record_sample_for_stack(num, timestamp_delta);
+
+    _stackprof.buffer_count = num;
+    _stackprof.buffer_time.timestamp_usec = start_timestamp;
+    _stackprof.buffer_time.delta_usec = timestamp_delta;
 }
 
 void
-stackprof_record_gc_samples()
+stackprof_record_gc_samples(void)
 {
-    int delta_to_first_unrecorded_gc_sample = 0;
-    int i;
+    int64_t delta_to_first_unrecorded_gc_sample = 0;
+    uint64_t start_timestamp = 0;
+    size_t i;
     if (_stackprof.raw) {
-	struct timeval t;
-	struct timeval diff;
-	gettimeofday(&t, NULL);
-	timersub(&t, &_stackprof.last_sample_at, &diff);
+	struct timestamp_t t;
+	capture_timestamp(&t);
+	start_timestamp = timestamp_usec(&t);
 
 	// We don't know when the GC samples were actually marked, so let's
 	// assume that they were marked at a perfectly regular interval.
-	delta_to_first_unrecorded_gc_sample = (1000 * diff.tv_sec + diff.tv_usec) - (_stackprof.unrecorded_gc_samples - 1) * NUM2LONG(_stackprof.interval);
+	delta_to_first_unrecorded_gc_sample = delta_usec(&t, &_stackprof.last_sample_at) - (_stackprof.unrecorded_gc_samples - 1) * NUM2LONG(_stackprof.interval);
 	if (delta_to_first_unrecorded_gc_sample < 0) {
 	    delta_to_first_unrecorded_gc_sample = 0;
 	}
     }
 
-
     for (i = 0; i < _stackprof.unrecorded_gc_samples; i++) {
-	int timestamp_delta = i == 0 ? delta_to_first_unrecorded_gc_sample : NUM2LONG(_stackprof.interval);
+	int64_t timestamp_delta = i == 0 ? delta_to_first_unrecorded_gc_sample : NUM2LONG(_stackprof.interval);
 
       if (_stackprof.unrecorded_gc_marking_samples) {
         _stackprof.frames_buffer[0] = FAKE_FRAME_MARK;
@@ -558,7 +654,7 @@ stackprof_record_gc_samples()
         _stackprof.lines_buffer[1] = 0;
         _stackprof.unrecorded_gc_marking_samples--;
 
-        stackprof_record_sample_for_stack(2, timestamp_delta);
+        stackprof_record_sample_for_stack(2, start_timestamp, timestamp_delta);
       } else if (_stackprof.unrecorded_gc_sweeping_samples) {
         _stackprof.frames_buffer[0] = FAKE_FRAME_SWEEP;
         _stackprof.lines_buffer[0] = 0;
@@ -567,11 +663,11 @@ stackprof_record_gc_samples()
 
         _stackprof.unrecorded_gc_sweeping_samples--;
 
-        stackprof_record_sample_for_stack(2, timestamp_delta);
+        stackprof_record_sample_for_stack(2, start_timestamp, timestamp_delta);
       } else {
         _stackprof.frames_buffer[0] = FAKE_FRAME_GC;
         _stackprof.lines_buffer[0] = 0;
-        stackprof_record_sample_for_stack(1, timestamp_delta);
+        stackprof_record_sample_for_stack(1, start_timestamp, timestamp_delta);
       }
     }
     _stackprof.during_gc += _stackprof.unrecorded_gc_samples;
@@ -580,35 +676,61 @@ stackprof_record_gc_samples()
     _stackprof.unrecorded_gc_sweeping_samples = 0;
 }
 
+// record the sample previously buffered by stackprof_buffer_sample
+static void
+stackprof_record_buffer(void)
+{
+    stackprof_record_sample_for_stack(_stackprof.buffer_count, _stackprof.buffer_time.timestamp_usec, _stackprof.buffer_time.delta_usec);
+
+    // reset the buffer
+    _stackprof.buffer_count = 0;
+}
+
 static void
-stackprof_gc_job_handler(void *data)
+stackprof_sample_and_record(void)
+{
+    stackprof_buffer_sample();
+    stackprof_record_buffer();
+}
+
+static void
+stackprof_job_record_gc(void *data)
 {
-    static int in_signal_handler = 0;
-    if (in_signal_handler) return;
     if (!_stackprof.running) return;
 
-    in_signal_handler++;
     stackprof_record_gc_samples();
-    in_signal_handler--;
 }
 
+#ifdef USE_POSTPONED_JOB
 static void
-stackprof_job_handler(void *data)
+stackprof_job_sample_and_record(void *data)
 {
-    static int in_signal_handler = 0;
-    if (in_signal_handler) return;
     if (!_stackprof.running) return;
 
-    in_signal_handler++;
-    stackprof_record_sample();
-    in_signal_handler--;
+    stackprof_sample_and_record();
+}
+#endif
+
+static void
+stackprof_job_record_buffer(void *data)
+{
+    if (!_stackprof.running) return;
+
+    stackprof_record_buffer();
 }
 
 static void
 stackprof_signal_handler(int sig, siginfo_t *sinfo, void *ucontext)
 {
+    static pthread_mutex_t lock = PTHREAD_MUTEX_INITIALIZER;
+
     _stackprof.overall_signals++;
-    if (rb_during_gc()) {
+
+    if (!_stackprof.running) return;
+    if (!ruby_native_thread_p()) return;
+    if (pthread_mutex_trylock(&lock)) return;
+
+    if (!_stackprof.ignore_gc && rb_during_gc()) {
 	VALUE mode = rb_gc_latest_gc_info(sym_state);
 	if (mode == sym_marking) {
 	    _stackprof.unrecorded_gc_marking_samples++;
@@ -616,10 +738,19 @@ stackprof_signal_handler(int sig, siginfo_t *sinfo, void *ucontext)
 	    _stackprof.unrecorded_gc_sweeping_samples++;
 	}
 	_stackprof.unrecorded_gc_samples++;
-	rb_postponed_job_register_one(0, stackprof_gc_job_handler, (void*)0);
+	rb_postponed_job_register_one(0, stackprof_job_record_gc, (void*)0);
     } else {
-	rb_postponed_job_register_one(0, stackprof_job_handler, (void*)0);
+#ifdef USE_POSTPONED_JOB
+	rb_postponed_job_register_one(0, stackprof_job_sample_and_record, (void*)0);
+#else
+	// Buffer a sample immediately, if an existing sample exists this will
+	// return immediately
+	stackprof_buffer_sample();
+	// Enqueue a job to record the sample
+	rb_postponed_job_register_one(0, stackprof_job_record_buffer, (void*)0);
+#endif
     }
+    pthread_mutex_unlock(&lock);
 }
 
 static void
@@ -628,7 +759,7 @@ stackprof_newobj_handler(VALUE tpval, void *data)
     _stackprof.overall_signals++;
     if (RTEST(_stackprof.interval) && _stackprof.overall_signals % NUM2LONG(_stackprof.interval))
 	return;
-    stackprof_job_handler(0);
+    stackprof_sample_and_record();
 }
 
 static VALUE
@@ -638,7 +769,7 @@ stackprof_sample(VALUE self)
 	return Qfalse;
 
     _stackprof.overall_signals++;
-    stackprof_job_handler(0);
+    stackprof_sample_and_record();
     return Qtrue;
 }
 
@@ -653,6 +784,9 @@ frame_mark_i(st_data_t key, st_data_t val, st_data_t arg)
 static void
 stackprof_gc_mark(void *data)
 {
+    if (RTEST(_stackprof.metadata))
+	rb_gc_mark(_stackprof.metadata);
+
     if (RTEST(_stackprof.out))
 	rb_gc_mark(_stackprof.out);
 
@@ -714,9 +848,11 @@ Init_stackprof(void)
     S(mode);
     S(interval);
     S(raw);
+    S(raw_sample_timestamps);
     S(raw_timestamp_deltas);
     S(out);
     S(metadata);
+    S(ignore_gc);
     S(frames);
     S(aggregate);
     S(state);
@@ -735,9 +871,9 @@ Init_stackprof(void)
     _stackprof.raw_samples_capa = 0;
     _stackprof.raw_sample_index = 0;
 
-    _stackprof.raw_timestamp_deltas = NULL;
-    _stackprof.raw_timestamp_deltas_len = 0;
-    _stackprof.raw_timestamp_deltas_capa = 0;
+    _stackprof.raw_sample_times = NULL;
+    _stackprof.raw_sample_times_len = 0;
+    _stackprof.raw_sample_times_capa = 0;
 
     _stackprof.empty_string = rb_str_new_cstr("");
     rb_global_variable(&_stackprof.empty_string);

data/lib/stackprof/report.rb

@@ -1,3 +1,5 @@
+# frozen_string_literal: true
+
 require 'pp'
 require 'digest/md5'
 
@@ -38,7 +40,7 @@ module StackProf
     end
 
     def max_samples
-      @data[:max_samples] ||= frames.max_by{ |addr, frame| frame[:samples] }.last[:samples]
+      @data[:max_samples] ||= @data[:frames].values.max_by{ |frame| frame[:samples] }[:samples]
     end
 
     def files
@@ -96,15 +98,7 @@ module StackProf
     def print_flamegraph(f, skip_common, alphabetical=false)
       raise "profile does not include raw samples (add `raw: true` to collecting StackProf.run)" unless raw = data[:raw]
 
-      stacks = []
-      max_x = 0
-      max_y = 0
-      while len = raw.shift
-        max_y = len if len > max_y
-        stack = raw.slice!(0, len+1)
-        stacks << stack
-        max_x += stack.last
-      end
+      stacks, max_x, max_y = flamegraph_stacks(raw)
 
       stacks.sort! if alphabetical
 
@@ -156,8 +150,26 @@ module StackProf
       f.puts '])'
     end
 
+    def flamegraph_stacks(raw)
+      stacks = []
+      max_x = 0
+      max_y = 0
+      idx = 0
+
+      while len = raw[idx]
+        idx += 1
+        max_y = len if len > max_y
+        stack = raw.slice(idx, len+1)
+        idx += len+1
+        stacks << stack
+        max_x += stack.last
+      end
+
+      return stacks, max_x, max_y
+    end
+
     def flamegraph_row(f, x, y, weight, addr)
-      frame = frames[addr]
+      frame = @data[:frames][addr]
       f.print ',' if @rows_started
       @rows_started = true
       f.puts %{{"x":#{x},"y":#{y},"width":#{weight},"frame_id":#{addr},"frame":#{frame[:name].dump},"file":#{frame[:file].dump}}}
@@ -178,7 +190,7 @@ module StackProf
             weight += stack.last
           end
         else
-          frame = frames[val]
+          frame = @data[:frames][val]
           child_name = "#{ frame[:name] } : #{ frame[:file] }"
           child_data = convert_to_d3_flame_graph_format(child_name, child_stacks, depth + 1)
           weight += child_data["value"]
@@ -196,15 +208,7 @@ module StackProf
     def print_d3_flamegraph(f=STDOUT, skip_common=true)
       raise "profile does not include raw samples (add `raw: true` to collecting StackProf.run)" unless raw = data[:raw]
 
-      stacks = []
-      max_x = 0
-      max_y = 0
-      while len = raw.shift
-        max_y = len if len > max_y
-        stack = raw.slice!(0, len+1)
-        stacks << stack
-        max_x += stack.last
-      end
+      stacks, * = flamegraph_stacks(raw)
 
       # d3-flame-grpah supports only alphabetical flamegraph
       stacks.sort!
@@ -410,7 +414,7 @@ module StackProf
         call, total = info.values_at(:samples, :total_samples)
         break if total < node_minimum || (limit && index >= limit)
 
-        sample = ''
+        sample = ''.dup
         sample << "#{call} (%2.1f%%)\\rof " % (call*100.0/overall_samples) if call < total
         sample << "#{total} (%2.1f%%)\\r" % (total*100.0/overall_samples)
         fontsize = (1.0 * call / max_samples) * 28 + 10
@@ -654,7 +658,8 @@ module StackProf
           end
         end
       end
+    rescue SystemCallError
+      f.puts "        SOURCE UNAVAILABLE"
     end
-
   end
 end

data/lib/stackprof.rb

@@ -1,7 +1,7 @@
 require "stackprof/stackprof"
 
 module StackProf
-  VERSION = '0.2.14'
+  VERSION = '0.2.18'
 end
 
 StackProf.autoload :Report, "stackprof/report.rb"

data/stackprof.gemspec

@@ -1,11 +1,18 @@
 Gem::Specification.new do |s|
   s.name = 'stackprof'
-  s.version = '0.2.14'
+  s.version = '0.2.18'
   s.homepage = 'http://github.com/tmm1/stackprof'
 
   s.authors = 'Aman Gupta'
   s.email   = 'aman@tmm1.net'
 
+  s.metadata = {
+    'bug_tracker_uri'   => 'https://github.com/tmm1/stackprof/issues',
+    'changelog_uri'     => "https://github.com/tmm1/stackprof/blob/v#{s.version}/CHANGELOG.md",
+    'documentation_uri' => "https://www.rubydoc.info/gems/stackprof/#{s.version}",
+    'source_code_uri'   => "https://github.com/tmm1/stackprof/tree/v#{s.version}"
+  }
+
   s.files = `git ls-files`.split("\n")
   s.extensions = 'ext/stackprof/extconf.rb'