michael@0: // Copyright (c) 2011 The Chromium Authors. All rights reserved.
michael@0: // Use of this source code is governed by a BSD-style license that can be
michael@0: // found in the LICENSE file.
michael@0: 
michael@0: // Histogram is an object that aggregates statistics, and can summarize them in
michael@0: // various forms, including ASCII graphical, HTML, and numerically (as a
michael@0: // vector of numbers corresponding to each of the aggregating buckets).
michael@0: // See header file for details and examples.
michael@0: 
michael@0: #include "base/histogram.h"
michael@0: 
michael@0: #include <math.h>
michael@0: 
michael@0: #include <algorithm>
michael@0: #include <string>
michael@0: 
michael@0: #include "base/logging.h"
michael@0: #include "base/pickle.h"
michael@0: #include "base/string_util.h"
michael@0: #include "base/logging.h"
michael@0: 
michael@0: namespace base {
michael@0: 
michael@0: #define DVLOG(x) CHROMIUM_LOG(ERROR)
michael@0: #define CHECK_GT DCHECK_GT
michael@0: #define CHECK_LT DCHECK_LT
michael@0: typedef ::Lock Lock;
michael@0: typedef ::AutoLock AutoLock;
michael@0: 
michael@0: // Static table of checksums for all possible 8 bit bytes.
michael@0: const uint32_t Histogram::kCrcTable[256] = {0x0, 0x77073096L, 0xee0e612cL,
michael@0: 0x990951baL, 0x76dc419L, 0x706af48fL, 0xe963a535L, 0x9e6495a3L, 0xedb8832L,
michael@0: 0x79dcb8a4L, 0xe0d5e91eL, 0x97d2d988L, 0x9b64c2bL, 0x7eb17cbdL, 0xe7b82d07L,
michael@0: 0x90bf1d91L, 0x1db71064L, 0x6ab020f2L, 0xf3b97148L, 0x84be41deL, 0x1adad47dL,
michael@0: 0x6ddde4ebL, 0xf4d4b551L, 0x83d385c7L, 0x136c9856L, 0x646ba8c0L, 0xfd62f97aL,
michael@0: 0x8a65c9ecL, 0x14015c4fL, 0x63066cd9L, 0xfa0f3d63L, 0x8d080df5L, 0x3b6e20c8L,
michael@0: 0x4c69105eL, 0xd56041e4L, 0xa2677172L, 0x3c03e4d1L, 0x4b04d447L, 0xd20d85fdL,
michael@0: 0xa50ab56bL, 0x35b5a8faL, 0x42b2986cL, 0xdbbbc9d6L, 0xacbcf940L, 0x32d86ce3L,
michael@0: 0x45df5c75L, 0xdcd60dcfL, 0xabd13d59L, 0x26d930acL, 0x51de003aL, 0xc8d75180L,
michael@0: 0xbfd06116L, 0x21b4f4b5L, 0x56b3c423L, 0xcfba9599L, 0xb8bda50fL, 0x2802b89eL,
michael@0: 0x5f058808L, 0xc60cd9b2L, 0xb10be924L, 0x2f6f7c87L, 0x58684c11L, 0xc1611dabL,
michael@0: 0xb6662d3dL, 0x76dc4190L, 0x1db7106L, 0x98d220bcL, 0xefd5102aL, 0x71b18589L,
michael@0: 0x6b6b51fL, 0x9fbfe4a5L, 0xe8b8d433L, 0x7807c9a2L, 0xf00f934L, 0x9609a88eL,
michael@0: 0xe10e9818L, 0x7f6a0dbbL, 0x86d3d2dL, 0x91646c97L, 0xe6635c01L, 0x6b6b51f4L,
michael@0: 0x1c6c6162L, 0x856530d8L, 0xf262004eL, 0x6c0695edL, 0x1b01a57bL, 0x8208f4c1L,
michael@0: 0xf50fc457L, 0x65b0d9c6L, 0x12b7e950L, 0x8bbeb8eaL, 0xfcb9887cL, 0x62dd1ddfL,
michael@0: 0x15da2d49L, 0x8cd37cf3L, 0xfbd44c65L, 0x4db26158L, 0x3ab551ceL, 0xa3bc0074L,
michael@0: 0xd4bb30e2L, 0x4adfa541L, 0x3dd895d7L, 0xa4d1c46dL, 0xd3d6f4fbL, 0x4369e96aL,
michael@0: 0x346ed9fcL, 0xad678846L, 0xda60b8d0L, 0x44042d73L, 0x33031de5L, 0xaa0a4c5fL,
michael@0: 0xdd0d7cc9L, 0x5005713cL, 0x270241aaL, 0xbe0b1010L, 0xc90c2086L, 0x5768b525L,
michael@0: 0x206f85b3L, 0xb966d409L, 0xce61e49fL, 0x5edef90eL, 0x29d9c998L, 0xb0d09822L,
michael@0: 0xc7d7a8b4L, 0x59b33d17L, 0x2eb40d81L, 0xb7bd5c3bL, 0xc0ba6cadL, 0xedb88320L,
michael@0: 0x9abfb3b6L, 0x3b6e20cL, 0x74b1d29aL, 0xead54739L, 0x9dd277afL, 0x4db2615L,
michael@0: 0x73dc1683L, 0xe3630b12L, 0x94643b84L, 0xd6d6a3eL, 0x7a6a5aa8L, 0xe40ecf0bL,
michael@0: 0x9309ff9dL, 0xa00ae27L, 0x7d079eb1L, 0xf00f9344L, 0x8708a3d2L, 0x1e01f268L,
michael@0: 0x6906c2feL, 0xf762575dL, 0x806567cbL, 0x196c3671L, 0x6e6b06e7L, 0xfed41b76L,
michael@0: 0x89d32be0L, 0x10da7a5aL, 0x67dd4accL, 0xf9b9df6fL, 0x8ebeeff9L, 0x17b7be43L,
michael@0: 0x60b08ed5L, 0xd6d6a3e8L, 0xa1d1937eL, 0x38d8c2c4L, 0x4fdff252L, 0xd1bb67f1L,
michael@0: 0xa6bc5767L, 0x3fb506ddL, 0x48b2364bL, 0xd80d2bdaL, 0xaf0a1b4cL, 0x36034af6L,
michael@0: 0x41047a60L, 0xdf60efc3L, 0xa867df55L, 0x316e8eefL, 0x4669be79L, 0xcb61b38cL,
michael@0: 0xbc66831aL, 0x256fd2a0L, 0x5268e236L, 0xcc0c7795L, 0xbb0b4703L, 0x220216b9L,
michael@0: 0x5505262fL, 0xc5ba3bbeL, 0xb2bd0b28L, 0x2bb45a92L, 0x5cb36a04L, 0xc2d7ffa7L,
michael@0: 0xb5d0cf31L, 0x2cd99e8bL, 0x5bdeae1dL, 0x9b64c2b0L, 0xec63f226L, 0x756aa39cL,
michael@0: 0x26d930aL, 0x9c0906a9L, 0xeb0e363fL, 0x72076785L, 0x5005713L, 0x95bf4a82L,
michael@0: 0xe2b87a14L, 0x7bb12baeL, 0xcb61b38L, 0x92d28e9bL, 0xe5d5be0dL, 0x7cdcefb7L,
michael@0: 0xbdbdf21L, 0x86d3d2d4L, 0xf1d4e242L, 0x68ddb3f8L, 0x1fda836eL, 0x81be16cdL,
michael@0: 0xf6b9265bL, 0x6fb077e1L, 0x18b74777L, 0x88085ae6L, 0xff0f6a70L, 0x66063bcaL,
michael@0: 0x11010b5cL, 0x8f659effL, 0xf862ae69L, 0x616bffd3L, 0x166ccf45L, 0xa00ae278L,
michael@0: 0xd70dd2eeL, 0x4e048354L, 0x3903b3c2L, 0xa7672661L, 0xd06016f7L, 0x4969474dL,
michael@0: 0x3e6e77dbL, 0xaed16a4aL, 0xd9d65adcL, 0x40df0b66L, 0x37d83bf0L, 0xa9bcae53L,
michael@0: 0xdebb9ec5L, 0x47b2cf7fL, 0x30b5ffe9L, 0xbdbdf21cL, 0xcabac28aL, 0x53b39330L,
michael@0: 0x24b4a3a6L, 0xbad03605L, 0xcdd70693L, 0x54de5729L, 0x23d967bfL, 0xb3667a2eL,
michael@0: 0xc4614ab8L, 0x5d681b02L, 0x2a6f2b94L, 0xb40bbe37L, 0xc30c8ea1L, 0x5a05df1bL,
michael@0: 0x2d02ef8dL,
michael@0: };
michael@0: 
michael@0: typedef Histogram::Count Count;
michael@0: 
michael@0: // static
michael@0: const size_t Histogram::kBucketCount_MAX = 16384u;
michael@0: 
michael@0: Histogram* Histogram::FactoryGet(const std::string& name,
michael@0:                                  Sample minimum,
michael@0:                                  Sample maximum,
michael@0:                                  size_t bucket_count,
michael@0:                                  Flags flags) {
michael@0:   Histogram* histogram(NULL);
michael@0: 
michael@0:   // Defensive code.
michael@0:   if (minimum < 1)
michael@0:     minimum = 1;
michael@0:   if (maximum > kSampleType_MAX - 1)
michael@0:     maximum = kSampleType_MAX - 1;
michael@0: 
michael@0:   if (!StatisticsRecorder::FindHistogram(name, &histogram)) {
michael@0:     // Extra variable is not needed... but this keeps this section basically
michael@0:     // identical to other derived classes in this file (and compiler will
michael@0:     // optimize away the extra variable.
michael@0:     Histogram* tentative_histogram =
michael@0:         new Histogram(name, minimum, maximum, bucket_count);
michael@0:     tentative_histogram->InitializeBucketRange();
michael@0:     tentative_histogram->SetFlags(flags);
michael@0:     histogram =
michael@0:         StatisticsRecorder::RegisterOrDeleteDuplicate(tentative_histogram);
michael@0:   }
michael@0: 
michael@0:   DCHECK_EQ(HISTOGRAM, histogram->histogram_type());
michael@0:   DCHECK(histogram->HasConstructorArguments(minimum, maximum, bucket_count));
michael@0:   return histogram;
michael@0: }
michael@0: 
michael@0: Histogram* Histogram::FactoryTimeGet(const std::string& name,
michael@0:                                      TimeDelta minimum,
michael@0:                                      TimeDelta maximum,
michael@0:                                      size_t bucket_count,
michael@0:                                      Flags flags) {
michael@0:   return FactoryGet(name, minimum.InMilliseconds(), maximum.InMilliseconds(),
michael@0:                     bucket_count, flags);
michael@0: }
michael@0: 
michael@0: void Histogram::Add(int value) {
michael@0:   if (value > kSampleType_MAX - 1)
michael@0:     value = kSampleType_MAX - 1;
michael@0:   if (value < 0)
michael@0:     value = 0;
michael@0:   size_t index = BucketIndex(value);
michael@0:   DCHECK_GE(value, ranges(index));
michael@0:   DCHECK_LT(value, ranges(index + 1));
michael@0:   Accumulate(value, 1, index);
michael@0: }
michael@0: 
michael@0: void Histogram::Subtract(int value) {
michael@0:   if (value > kSampleType_MAX - 1)
michael@0:     value = kSampleType_MAX - 1;
michael@0:   if (value < 0)
michael@0:     value = 0;
michael@0:   size_t index = BucketIndex(value);
michael@0:   DCHECK_GE(value, ranges(index));
michael@0:   DCHECK_LT(value, ranges(index + 1));
michael@0:   Accumulate(value, -1, index);
michael@0: }
michael@0: 
michael@0: void Histogram::AddBoolean(bool value) {
michael@0:   DCHECK(false);
michael@0: }
michael@0: 
michael@0: void Histogram::AddSampleSet(const SampleSet& sample) {
michael@0:   sample_.Add(sample);
michael@0: }
michael@0: 
michael@0: void Histogram::Clear() {
michael@0:   SampleSet ss;
michael@0:   ss.Resize(*this);
michael@0:   sample_ = ss;
michael@0: }
michael@0: 
michael@0: void Histogram::SetRangeDescriptions(const DescriptionPair descriptions[]) {
michael@0:   DCHECK(false);
michael@0: }
michael@0: 
michael@0: // The following methods provide a graphical histogram display.
michael@0: void Histogram::WriteHTMLGraph(std::string* output) const {
michael@0:   // TBD(jar) Write a nice HTML bar chart, with divs an mouse-overs etc.
michael@0:   output->append("<PRE>");
michael@0:   WriteAscii(true, "<br>", output);
michael@0:   output->append("</PRE>");
michael@0: }
michael@0: 
michael@0: void Histogram::WriteAscii(bool graph_it, const std::string& newline,
michael@0:                            std::string* output) const {
michael@0:   // Get local (stack) copies of all effectively volatile class data so that we
michael@0:   // are consistent across our output activities.
michael@0:   SampleSet snapshot;
michael@0:   SnapshotSample(&snapshot);
michael@0:   Count sample_count = snapshot.TotalCount();
michael@0: 
michael@0:   WriteAsciiHeader(snapshot, sample_count, output);
michael@0:   output->append(newline);
michael@0: 
michael@0:   // Prepare to normalize graphical rendering of bucket contents.
michael@0:   double max_size = 0;
michael@0:   if (graph_it)
michael@0:     max_size = GetPeakBucketSize(snapshot);
michael@0: 
michael@0:   // Calculate space needed to print bucket range numbers.  Leave room to print
michael@0:   // nearly the largest bucket range without sliding over the histogram.
michael@0:   size_t largest_non_empty_bucket = bucket_count() - 1;
michael@0:   while (0 == snapshot.counts(largest_non_empty_bucket)) {
michael@0:     if (0 == largest_non_empty_bucket)
michael@0:       break;  // All buckets are empty.
michael@0:     --largest_non_empty_bucket;
michael@0:   }
michael@0: 
michael@0:   // Calculate largest print width needed for any of our bucket range displays.
michael@0:   size_t print_width = 1;
michael@0:   for (size_t i = 0; i < bucket_count(); ++i) {
michael@0:     if (snapshot.counts(i)) {
michael@0:       size_t width = GetAsciiBucketRange(i).size() + 1;
michael@0:       if (width > print_width)
michael@0:         print_width = width;
michael@0:     }
michael@0:   }
michael@0: 
michael@0:   int64_t remaining = sample_count;
michael@0:   int64_t past = 0;
michael@0:   // Output the actual histogram graph.
michael@0:   for (size_t i = 0; i < bucket_count(); ++i) {
michael@0:     Count current = snapshot.counts(i);
michael@0:     if (!current && !PrintEmptyBucket(i))
michael@0:       continue;
michael@0:     remaining -= current;
michael@0:     std::string range = GetAsciiBucketRange(i);
michael@0:     output->append(range);
michael@0:     for (size_t j = 0; range.size() + j < print_width + 1; ++j)
michael@0:       output->push_back(' ');
michael@0:     if (0 == current && i < bucket_count() - 1 && 0 == snapshot.counts(i + 1)) {
michael@0:       while (i < bucket_count() - 1 && 0 == snapshot.counts(i + 1))
michael@0:         ++i;
michael@0:       output->append("... ");
michael@0:       output->append(newline);
michael@0:       continue;  // No reason to plot emptiness.
michael@0:     }
michael@0:     double current_size = GetBucketSize(current, i);
michael@0:     if (graph_it)
michael@0:       WriteAsciiBucketGraph(current_size, max_size, output);
michael@0:     WriteAsciiBucketContext(past, current, remaining, i, output);
michael@0:     output->append(newline);
michael@0:     past += current;
michael@0:   }
michael@0:   DCHECK_EQ(sample_count, past);
michael@0: }
michael@0: 
michael@0: // static
michael@0: std::string Histogram::SerializeHistogramInfo(const Histogram& histogram,
michael@0:                                               const SampleSet& snapshot) {
michael@0:   DCHECK_NE(NOT_VALID_IN_RENDERER, histogram.histogram_type());
michael@0: 
michael@0:   Pickle pickle;
michael@0:   pickle.WriteString(histogram.histogram_name());
michael@0:   pickle.WriteInt(histogram.declared_min());
michael@0:   pickle.WriteInt(histogram.declared_max());
michael@0:   pickle.WriteSize(histogram.bucket_count());
michael@0:   pickle.WriteUInt32(histogram.range_checksum());
michael@0:   pickle.WriteInt(histogram.histogram_type());
michael@0:   pickle.WriteInt(histogram.flags());
michael@0: 
michael@0:   snapshot.Serialize(&pickle);
michael@0:   return std::string(static_cast<const char*>(pickle.data()), pickle.size());
michael@0: }
michael@0: 
michael@0: // static
michael@0: bool Histogram::DeserializeHistogramInfo(const std::string& histogram_info) {
michael@0:   if (histogram_info.empty()) {
michael@0:       return false;
michael@0:   }
michael@0: 
michael@0:   Pickle pickle(histogram_info.data(),
michael@0:                 static_cast<int>(histogram_info.size()));
michael@0:   std::string histogram_name;
michael@0:   int declared_min;
michael@0:   int declared_max;
michael@0:   size_t bucket_count;
michael@0:   uint32_t range_checksum;
michael@0:   int histogram_type;
michael@0:   int pickle_flags;
michael@0:   SampleSet sample;
michael@0: 
michael@0:   void* iter = NULL;
michael@0:   if (!pickle.ReadString(&iter, &histogram_name) ||
michael@0:       !pickle.ReadInt(&iter, &declared_min) ||
michael@0:       !pickle.ReadInt(&iter, &declared_max) ||
michael@0:       !pickle.ReadSize(&iter, &bucket_count) ||
michael@0:       !pickle.ReadUInt32(&iter, &range_checksum) ||
michael@0:       !pickle.ReadInt(&iter, &histogram_type) ||
michael@0:       !pickle.ReadInt(&iter, &pickle_flags) ||
michael@0:       !sample.Histogram::SampleSet::Deserialize(&iter, pickle)) {
michael@0:     CHROMIUM_LOG(ERROR) << "Pickle error decoding Histogram: " << histogram_name;
michael@0:     return false;
michael@0:   }
michael@0:   DCHECK(pickle_flags & kIPCSerializationSourceFlag);
michael@0:   // Since these fields may have come from an untrusted renderer, do additional
michael@0:   // checks above and beyond those in Histogram::Initialize()
michael@0:   if (declared_max <= 0 || declared_min <= 0 || declared_max < declared_min ||
michael@0:       INT_MAX / sizeof(Count) <= bucket_count || bucket_count < 2) {
michael@0:     CHROMIUM_LOG(ERROR) << "Values error decoding Histogram: " << histogram_name;
michael@0:     return false;
michael@0:   }
michael@0: 
michael@0:   Flags flags = static_cast<Flags>(pickle_flags & ~kIPCSerializationSourceFlag);
michael@0: 
michael@0:   DCHECK_NE(NOT_VALID_IN_RENDERER, histogram_type);
michael@0: 
michael@0:   Histogram* render_histogram(NULL);
michael@0: 
michael@0:   if (histogram_type == HISTOGRAM) {
michael@0:     render_histogram = Histogram::FactoryGet(
michael@0:         histogram_name, declared_min, declared_max, bucket_count, flags);
michael@0:   } else if (histogram_type == LINEAR_HISTOGRAM) {
michael@0:     render_histogram = LinearHistogram::FactoryGet(
michael@0:         histogram_name, declared_min, declared_max, bucket_count, flags);
michael@0:   } else if (histogram_type == BOOLEAN_HISTOGRAM) {
michael@0:     render_histogram = BooleanHistogram::FactoryGet(histogram_name, flags);
michael@0:   } else {
michael@0:     CHROMIUM_LOG(ERROR) << "Error Deserializing Histogram Unknown histogram_type: "
michael@0:                         << histogram_type;
michael@0:     return false;
michael@0:   }
michael@0: 
michael@0:   DCHECK_EQ(render_histogram->declared_min(), declared_min);
michael@0:   DCHECK_EQ(render_histogram->declared_max(), declared_max);
michael@0:   DCHECK_EQ(render_histogram->bucket_count(), bucket_count);
michael@0:   DCHECK_EQ(render_histogram->range_checksum(), range_checksum);
michael@0:   DCHECK_EQ(render_histogram->histogram_type(), histogram_type);
michael@0: 
michael@0:   if (render_histogram->flags() & kIPCSerializationSourceFlag) {
michael@0:     DVLOG(1) << "Single process mode, histogram observed and not copied: "
michael@0:              << histogram_name;
michael@0:   } else {
michael@0:     DCHECK_EQ(flags & render_histogram->flags(), flags);
michael@0:     render_histogram->AddSampleSet(sample);
michael@0:   }
michael@0: 
michael@0:   return true;
michael@0: }
michael@0: 
michael@0: //------------------------------------------------------------------------------
michael@0: // Methods for the validating a sample and a related histogram.
michael@0: //------------------------------------------------------------------------------
michael@0: 
michael@0: Histogram::Inconsistencies Histogram::FindCorruption(
michael@0:     const SampleSet& snapshot) const {
michael@0:   int inconsistencies = NO_INCONSISTENCIES;
michael@0:   Sample previous_range = -1;  // Bottom range is always 0.
michael@0:   int64_t count = 0;
michael@0:   for (size_t index = 0; index < bucket_count(); ++index) {
michael@0:     count += snapshot.counts(index);
michael@0:     int new_range = ranges(index);
michael@0:     if (previous_range >= new_range)
michael@0:       inconsistencies |= BUCKET_ORDER_ERROR;
michael@0:     previous_range = new_range;
michael@0:   }
michael@0: 
michael@0:   if (!HasValidRangeChecksum())
michael@0:     inconsistencies |= RANGE_CHECKSUM_ERROR;
michael@0: 
michael@0:   int64_t delta64 = snapshot.redundant_count() - count;
michael@0:   if (delta64 != 0) {
michael@0:     int delta = static_cast<int>(delta64);
michael@0:     if (delta != delta64)
michael@0:       delta = INT_MAX;  // Flag all giant errors as INT_MAX.
michael@0:     // Since snapshots of histograms are taken asynchronously relative to
michael@0:     // sampling (and snapped from different threads), it is pretty likely that
michael@0:     // we'll catch a redundant count that doesn't match the sample count.  We
michael@0:     // allow for a certain amount of slop before flagging this as an
michael@0:     // inconsistency.  Even with an inconsistency, we'll snapshot it again (for
michael@0:     // UMA in about a half hour, so we'll eventually get the data, if it was
michael@0:     // not the result of a corruption.  If histograms show that 1 is "too tight"
michael@0:     // then we may try to use 2 or 3 for this slop value.
michael@0:     const int kCommonRaceBasedCountMismatch = 1;
michael@0:     if (delta > 0) {
michael@0:       UMA_HISTOGRAM_COUNTS("Histogram.InconsistentCountHigh", delta);
michael@0:       if (delta > kCommonRaceBasedCountMismatch)
michael@0:         inconsistencies |= COUNT_HIGH_ERROR;
michael@0:     } else {
michael@0:       DCHECK_GT(0, delta);
michael@0:       UMA_HISTOGRAM_COUNTS("Histogram.InconsistentCountLow", -delta);
michael@0:       if (-delta > kCommonRaceBasedCountMismatch)
michael@0:         inconsistencies |= COUNT_LOW_ERROR;
michael@0:     }
michael@0:   }
michael@0:   return static_cast<Inconsistencies>(inconsistencies);
michael@0: }
michael@0: 
michael@0: Histogram::ClassType Histogram::histogram_type() const {
michael@0:   return HISTOGRAM;
michael@0: }
michael@0: 
michael@0: Histogram::Sample Histogram::ranges(size_t i) const {
michael@0:   return ranges_[i];
michael@0: }
michael@0: 
michael@0: size_t Histogram::bucket_count() const {
michael@0:   return bucket_count_;
michael@0: }
michael@0: 
michael@0: // Do a safe atomic snapshot of sample data.
michael@0: // This implementation assumes we are on a safe single thread.
michael@0: void Histogram::SnapshotSample(SampleSet* sample) const {
michael@0:   // Note locking not done in this version!!!
michael@0:   *sample = sample_;
michael@0: }
michael@0: 
michael@0: bool Histogram::HasConstructorArguments(Sample minimum,
michael@0:                                         Sample maximum,
michael@0:                                         size_t bucket_count) {
michael@0:   return ((minimum == declared_min_) && (maximum == declared_max_) &&
michael@0:           (bucket_count == bucket_count_));
michael@0: }
michael@0: 
michael@0: bool Histogram::HasConstructorTimeDeltaArguments(TimeDelta minimum,
michael@0:                                                  TimeDelta maximum,
michael@0:                                                  size_t bucket_count) {
michael@0:   return ((minimum.InMilliseconds() == declared_min_) &&
michael@0:           (maximum.InMilliseconds() == declared_max_) &&
michael@0:           (bucket_count == bucket_count_));
michael@0: }
michael@0: 
michael@0: bool Histogram::HasValidRangeChecksum() const {
michael@0:   return CalculateRangeChecksum() == range_checksum_;
michael@0: }
michael@0: 
michael@0: size_t Histogram::SizeOfIncludingThis(mozilla::MallocSizeOf aMallocSizeOf)
michael@0: {
michael@0:   size_t n = 0;
michael@0:   n += aMallocSizeOf(this);
michael@0:   // We're not allowed to do deep dives into STL data structures.  This
michael@0:   // is as close as we can get to measuring this array.
michael@0:   n += aMallocSizeOf(&ranges_[0]);
michael@0:   n += sample_.SizeOfExcludingThis(aMallocSizeOf);
michael@0:   return n;
michael@0: }
michael@0: 
michael@0: size_t Histogram::SampleSet::SizeOfExcludingThis(mozilla::MallocSizeOf aMallocSizeOf)
michael@0: {
michael@0:   // We're not allowed to do deep dives into STL data structures.  This
michael@0:   // is as close as we can get to measuring this array.
michael@0:   return aMallocSizeOf(&counts_[0]);
michael@0: }
michael@0: 
michael@0: Histogram::Histogram(const std::string& name, Sample minimum,
michael@0:                      Sample maximum, size_t bucket_count)
michael@0:   : sample_(),
michael@0:     histogram_name_(name),
michael@0:     declared_min_(minimum),
michael@0:     declared_max_(maximum),
michael@0:     bucket_count_(bucket_count),
michael@0:     flags_(kNoFlags),
michael@0:     ranges_(bucket_count + 1, 0),
michael@0:     range_checksum_(0) {
michael@0:   Initialize();
michael@0: }
michael@0: 
michael@0: Histogram::Histogram(const std::string& name, TimeDelta minimum,
michael@0:                      TimeDelta maximum, size_t bucket_count)
michael@0:   : sample_(),
michael@0:     histogram_name_(name),
michael@0:     declared_min_(static_cast<int> (minimum.InMilliseconds())),
michael@0:     declared_max_(static_cast<int> (maximum.InMilliseconds())),
michael@0:     bucket_count_(bucket_count),
michael@0:     flags_(kNoFlags),
michael@0:     ranges_(bucket_count + 1, 0),
michael@0:     range_checksum_(0) {
michael@0:   Initialize();
michael@0: }
michael@0: 
michael@0: Histogram::~Histogram() {
michael@0:   if (StatisticsRecorder::dump_on_exit()) {
michael@0:     std::string output;
michael@0:     WriteAscii(true, "\n", &output);
michael@0:     CHROMIUM_LOG(INFO) << output;
michael@0:   }
michael@0: 
michael@0:   // Just to make sure most derived class did this properly...
michael@0:   DCHECK(ValidateBucketRanges());
michael@0: }
michael@0: 
michael@0: // Calculate what range of values are held in each bucket.
michael@0: // We have to be careful that we don't pick a ratio between starting points in
michael@0: // consecutive buckets that is sooo small, that the integer bounds are the same
michael@0: // (effectively making one bucket get no values).  We need to avoid:
michael@0: //   ranges_[i] == ranges_[i + 1]
michael@0: // To avoid that, we just do a fine-grained bucket width as far as we need to
michael@0: // until we get a ratio that moves us along at least 2 units at a time.  From
michael@0: // that bucket onward we do use the exponential growth of buckets.
michael@0: void Histogram::InitializeBucketRange() {
michael@0:   double log_max = log(static_cast<double>(declared_max()));
michael@0:   double log_ratio;
michael@0:   double log_next;
michael@0:   size_t bucket_index = 1;
michael@0:   Sample current = declared_min();
michael@0:   SetBucketRange(bucket_index, current);
michael@0:   while (bucket_count() > ++bucket_index) {
michael@0:     double log_current;
michael@0:     log_current = log(static_cast<double>(current));
michael@0:     // Calculate the count'th root of the range.
michael@0:     log_ratio = (log_max - log_current) / (bucket_count() - bucket_index);
michael@0:     // See where the next bucket would start.
michael@0:     log_next = log_current + log_ratio;
michael@0:     int next;
michael@0:     next = static_cast<int>(floor(exp(log_next) + 0.5));
michael@0:     if (next > current)
michael@0:       current = next;
michael@0:     else
michael@0:       ++current;  // Just do a narrow bucket, and keep trying.
michael@0:     SetBucketRange(bucket_index, current);
michael@0:   }
michael@0:   ResetRangeChecksum();
michael@0: 
michael@0:   DCHECK_EQ(bucket_count(), bucket_index);
michael@0: }
michael@0: 
michael@0: bool Histogram::PrintEmptyBucket(size_t index) const {
michael@0:   return true;
michael@0: }
michael@0: 
michael@0: size_t Histogram::BucketIndex(Sample value) const {
michael@0:   // Use simple binary search.  This is very general, but there are better
michael@0:   // approaches if we knew that the buckets were linearly distributed.
michael@0:   DCHECK_LE(ranges(0), value);
michael@0:   DCHECK_GT(ranges(bucket_count()), value);
michael@0:   size_t under = 0;
michael@0:   size_t over = bucket_count();
michael@0:   size_t mid;
michael@0: 
michael@0:   do {
michael@0:     DCHECK_GE(over, under);
michael@0:     mid = under + (over - under)/2;
michael@0:     if (mid == under)
michael@0:       break;
michael@0:     if (ranges(mid) <= value)
michael@0:       under = mid;
michael@0:     else
michael@0:       over = mid;
michael@0:   } while (true);
michael@0: 
michael@0:   DCHECK_LE(ranges(mid), value);
michael@0:   CHECK_GT(ranges(mid+1), value);
michael@0:   return mid;
michael@0: }
michael@0: 
michael@0: // Use the actual bucket widths (like a linear histogram) until the widths get
michael@0: // over some transition value, and then use that transition width.  Exponentials
michael@0: // get so big so fast (and we don't expect to see a lot of entries in the large
michael@0: // buckets), so we need this to make it possible to see what is going on and
michael@0: // not have 0-graphical-height buckets.
michael@0: double Histogram::GetBucketSize(Count current, size_t i) const {
michael@0:   DCHECK_GT(ranges(i + 1), ranges(i));
michael@0:   static const double kTransitionWidth = 5;
michael@0:   double denominator = ranges(i + 1) - ranges(i);
michael@0:   if (denominator > kTransitionWidth)
michael@0:     denominator = kTransitionWidth;  // Stop trying to normalize.
michael@0:   return current/denominator;
michael@0: }
michael@0: 
michael@0: void Histogram::ResetRangeChecksum() {
michael@0:   range_checksum_ = CalculateRangeChecksum();
michael@0: }
michael@0: 
michael@0: const std::string Histogram::GetAsciiBucketRange(size_t i) const {
michael@0:   std::string result;
michael@0:   if (kHexRangePrintingFlag & flags_)
michael@0:     StringAppendF(&result, "%#x", ranges(i));
michael@0:   else
michael@0:     StringAppendF(&result, "%d", ranges(i));
michael@0:   return result;
michael@0: }
michael@0: 
michael@0: // Update histogram data with new sample.
michael@0: void Histogram::Accumulate(Sample value, Count count, size_t index) {
michael@0:   // Note locking not done in this version!!!
michael@0:   sample_.AccumulateWithExponentialStats(value, count, index,
michael@0: 					 flags_ & kExtendedStatisticsFlag);
michael@0: }
michael@0: 
michael@0: void Histogram::SetBucketRange(size_t i, Sample value) {
michael@0:   DCHECK_GT(bucket_count_, i);
michael@0:   ranges_[i] = value;
michael@0: }
michael@0: 
michael@0: bool Histogram::ValidateBucketRanges() const {
michael@0:   // Standard assertions that all bucket ranges should satisfy.
michael@0:   DCHECK_EQ(bucket_count_ + 1, ranges_.size());
michael@0:   DCHECK_EQ(0, ranges_[0]);
michael@0:   DCHECK_EQ(declared_min(), ranges_[1]);
michael@0:   DCHECK_EQ(declared_max(), ranges_[bucket_count_ - 1]);
michael@0:   DCHECK_EQ(kSampleType_MAX, ranges_[bucket_count_]);
michael@0:   return true;
michael@0: }
michael@0: 
michael@0: uint32_t Histogram::CalculateRangeChecksum() const {
michael@0:   DCHECK_EQ(ranges_.size(), bucket_count() + 1);
michael@0:   uint32_t checksum = static_cast<uint32_t>(ranges_.size());  // Seed checksum.
michael@0:   for (size_t index = 0; index < bucket_count(); ++index)
michael@0:     checksum = Crc32(checksum, ranges(index));
michael@0:   return checksum;
michael@0: }
michael@0: 
michael@0: void Histogram::Initialize() {
michael@0:   sample_.Resize(*this);
michael@0:   if (declared_min_ < 1)
michael@0:     declared_min_ = 1;
michael@0:   if (declared_max_ > kSampleType_MAX - 1)
michael@0:     declared_max_ = kSampleType_MAX - 1;
michael@0:   DCHECK_LE(declared_min_, declared_max_);
michael@0:   DCHECK_GT(bucket_count_, 1u);
michael@0:   CHECK_LT(bucket_count_, kBucketCount_MAX);
michael@0:   size_t maximal_bucket_count = declared_max_ - declared_min_ + 2;
michael@0:   DCHECK_LE(bucket_count_, maximal_bucket_count);
michael@0:   DCHECK_EQ(0, ranges_[0]);
michael@0:   ranges_[bucket_count_] = kSampleType_MAX;
michael@0: }
michael@0: 
michael@0: // We generate the CRC-32 using the low order bits to select whether to XOR in
michael@0: // the reversed polynomial 0xedb88320L.  This is nice and simple, and allows us
michael@0: // to keep the quotient in a uint32_t.  Since we're not concerned about the nature
michael@0: // of corruptions (i.e., we don't care about bit sequencing, since we are
michael@0: // handling memory changes, which are more grotesque) so we don't bother to
michael@0: // get the CRC correct for big-endian vs little-ending calculations.  All we
michael@0: // need is a nice hash, that tends to depend on all the bits of the sample, with
michael@0: // very little chance of changes in one place impacting changes in another
michael@0: // place.
michael@0: uint32_t Histogram::Crc32(uint32_t sum, Histogram::Sample range) {
michael@0:   const bool kUseRealCrc = true;  // TODO(jar): Switch to false and watch stats.
michael@0:   if (kUseRealCrc) {
michael@0:     union {
michael@0:       Histogram::Sample range;
michael@0:       unsigned char bytes[sizeof(Histogram::Sample)];
michael@0:     } converter;
michael@0:     converter.range = range;
michael@0:     for (size_t i = 0; i < sizeof(converter); ++i)
michael@0:       sum = kCrcTable[(sum & 0xff) ^ converter.bytes[i]] ^ (sum >> 8);
michael@0:   } else {
michael@0:     // Use hash techniques provided in ReallyFastHash, except we don't care
michael@0:     // about "avalanching" (which would worsten the hash, and add collisions),
michael@0:     // and we don't care about edge cases since we have an even number of bytes.
michael@0:     union {
michael@0:       Histogram::Sample range;
michael@0:       uint16_t ints[sizeof(Histogram::Sample) / 2];
michael@0:     } converter;
michael@0:     DCHECK_EQ(sizeof(Histogram::Sample), sizeof(converter));
michael@0:     converter.range = range;
michael@0:     sum += converter.ints[0];
michael@0:     sum = (sum << 16) ^ sum ^ (static_cast<uint32_t>(converter.ints[1]) << 11);
michael@0:     sum += sum >> 11;
michael@0:   }
michael@0:   return sum;
michael@0: }
michael@0: 
michael@0: //------------------------------------------------------------------------------
michael@0: // Private methods
michael@0: 
michael@0: double Histogram::GetPeakBucketSize(const SampleSet& snapshot) const {
michael@0:   double max = 0;
michael@0:   for (size_t i = 0; i < bucket_count() ; ++i) {
michael@0:     double current_size = GetBucketSize(snapshot.counts(i), i);
michael@0:     if (current_size > max)
michael@0:       max = current_size;
michael@0:   }
michael@0:   return max;
michael@0: }
michael@0: 
michael@0: void Histogram::WriteAsciiHeader(const SampleSet& snapshot,
michael@0:                                  Count sample_count,
michael@0:                                  std::string* output) const {
michael@0:   StringAppendF(output,
michael@0:                 "Histogram: %s recorded %d samples",
michael@0:                 histogram_name().c_str(),
michael@0:                 sample_count);
michael@0:   if (0 == sample_count) {
michael@0:     DCHECK_EQ(snapshot.sum(), 0);
michael@0:   } else {
michael@0:     double average = static_cast<float>(snapshot.sum()) / sample_count;
michael@0: 
michael@0:     StringAppendF(output, ", average = %.1f", average);
michael@0:   }
michael@0:   if (flags_ & ~kHexRangePrintingFlag)
michael@0:     StringAppendF(output, " (flags = 0x%x)", flags_ & ~kHexRangePrintingFlag);
michael@0: }
michael@0: 
michael@0: void Histogram::WriteAsciiBucketContext(const int64_t past,
michael@0:                                         const Count current,
michael@0:                                         const int64_t remaining,
michael@0:                                         const size_t i,
michael@0:                                         std::string* output) const {
michael@0:   double scaled_sum = (past + current + remaining) / 100.0;
michael@0:   WriteAsciiBucketValue(current, scaled_sum, output);
michael@0:   if (0 < i) {
michael@0:     double percentage = past / scaled_sum;
michael@0:     StringAppendF(output, " {%3.1f%%}", percentage);
michael@0:   }
michael@0: }
michael@0: 
michael@0: void Histogram::WriteAsciiBucketValue(Count current, double scaled_sum,
michael@0:                                       std::string* output) const {
michael@0:   StringAppendF(output, " (%d = %3.1f%%)", current, current/scaled_sum);
michael@0: }
michael@0: 
michael@0: void Histogram::WriteAsciiBucketGraph(double current_size, double max_size,
michael@0:                                       std::string* output) const {
michael@0:   const int k_line_length = 72;  // Maximal horizontal width of graph.
michael@0:   int x_count = static_cast<int>(k_line_length * (current_size / max_size)
michael@0:                                  + 0.5);
michael@0:   int x_remainder = k_line_length - x_count;
michael@0: 
michael@0:   while (0 < x_count--)
michael@0:     output->append("-");
michael@0:   output->append("O");
michael@0:   while (0 < x_remainder--)
michael@0:     output->append(" ");
michael@0: }
michael@0: 
michael@0: //------------------------------------------------------------------------------
michael@0: // Methods for the Histogram::SampleSet class
michael@0: //------------------------------------------------------------------------------
michael@0: 
michael@0: Histogram::SampleSet::SampleSet()
michael@0:     : counts_(),
michael@0:       sum_(0),
michael@0:       sum_squares_(0),
michael@0:       log_sum_(0),
michael@0:       log_sum_squares_(0),
michael@0:       redundant_count_(0) {
michael@0: }
michael@0: 
michael@0: Histogram::SampleSet::~SampleSet() {
michael@0: }
michael@0: 
michael@0: void Histogram::SampleSet::Resize(const Histogram& histogram) {
michael@0:   counts_.resize(histogram.bucket_count(), 0);
michael@0: }
michael@0: 
michael@0: void Histogram::SampleSet::CheckSize(const Histogram& histogram) const {
michael@0:   DCHECK_EQ(histogram.bucket_count(), counts_.size());
michael@0: }
michael@0: 
michael@0: void Histogram::SampleSet::Accumulate(Sample value, Count count,
michael@0: 				      size_t index) {
michael@0:   DCHECK(count == 1 || count == -1);
michael@0:   counts_[index] += count;
michael@0:   redundant_count_ += count;
michael@0:   sum_ += static_cast<int64_t>(count) * value;
michael@0:   DCHECK_GE(counts_[index], 0);
michael@0:   DCHECK_GE(sum_, 0);
michael@0:   DCHECK_GE(redundant_count_, 0);
michael@0: }
michael@0: 
michael@0: void Histogram::SampleSet::AccumulateWithLinearStats(Sample value,
michael@0:                                                      Count count,
michael@0:                                                      size_t index) {
michael@0:   Accumulate(value, count, index);
michael@0:   sum_squares_ += static_cast<int64_t>(count) * value * value;
michael@0: }
michael@0: 
michael@0: void Histogram::SampleSet::AccumulateWithExponentialStats(Sample value,
michael@0:                                                           Count count,
michael@0:                                                           size_t index,
michael@0: 							  bool computeExtendedStatistics) {
michael@0:   Accumulate(value, count, index);
michael@0:   if (computeExtendedStatistics) {
michael@0:     DCHECK_GE(value, 0);
michael@0:     float value_log = logf(static_cast<float>(value) + 1.0f);
michael@0:     log_sum_ += count * value_log;
michael@0:     log_sum_squares_ += count * value_log * value_log;
michael@0:   }
michael@0: }
michael@0: 
michael@0: Count Histogram::SampleSet::TotalCount() const {
michael@0:   Count total = 0;
michael@0:   for (Counts::const_iterator it = counts_.begin();
michael@0:        it != counts_.end();
michael@0:        ++it) {
michael@0:     total += *it;
michael@0:   }
michael@0:   return total;
michael@0: }
michael@0: 
michael@0: void Histogram::SampleSet::Add(const SampleSet& other) {
michael@0:   DCHECK_EQ(counts_.size(), other.counts_.size());
michael@0:   sum_ += other.sum_;
michael@0:   sum_squares_ += other.sum_squares_;
michael@0:   log_sum_ += other.log_sum_;
michael@0:   log_sum_squares_ += other.log_sum_squares_;
michael@0:   redundant_count_ += other.redundant_count_;
michael@0:   for (size_t index = 0; index < counts_.size(); ++index)
michael@0:     counts_[index] += other.counts_[index];
michael@0: }
michael@0: 
michael@0: void Histogram::SampleSet::Subtract(const SampleSet& other) {
michael@0:   DCHECK_EQ(counts_.size(), other.counts_.size());
michael@0:   // Note: Race conditions in snapshotting a sum may lead to (temporary)
michael@0:   // negative values when snapshots are later combined (and deltas calculated).
michael@0:   // As a result, we don't currently CHCEK() for positive values.
michael@0:   sum_ -= other.sum_;
michael@0:   sum_squares_ -= other.sum_squares_;
michael@0:   log_sum_ -= other.log_sum_;
michael@0:   log_sum_squares_ -= other.log_sum_squares_;
michael@0:   redundant_count_ -= other.redundant_count_;
michael@0:   for (size_t index = 0; index < counts_.size(); ++index) {
michael@0:     counts_[index] -= other.counts_[index];
michael@0:     DCHECK_GE(counts_[index], 0);
michael@0:   }
michael@0: }
michael@0: 
michael@0: bool Histogram::SampleSet::Serialize(Pickle* pickle) const {
michael@0:   pickle->WriteInt64(sum_);
michael@0:   pickle->WriteInt64(redundant_count_);
michael@0:   pickle->WriteSize(counts_.size());
michael@0: 
michael@0:   for (size_t index = 0; index < counts_.size(); ++index) {
michael@0:     pickle->WriteInt(counts_[index]);
michael@0:   }
michael@0: 
michael@0:   return true;
michael@0: }
michael@0: 
michael@0: bool Histogram::SampleSet::Deserialize(void** iter, const Pickle& pickle) {
michael@0:   DCHECK_EQ(counts_.size(), 0u);
michael@0:   DCHECK_EQ(sum_, 0);
michael@0:   DCHECK_EQ(redundant_count_, 0);
michael@0: 
michael@0:   size_t counts_size;
michael@0: 
michael@0:   if (!pickle.ReadInt64(iter, &sum_) ||
michael@0:       !pickle.ReadInt64(iter, &redundant_count_) ||
michael@0:       !pickle.ReadSize(iter, &counts_size)) {
michael@0:     return false;
michael@0:   }
michael@0: 
michael@0:   if (counts_size == 0)
michael@0:     return false;
michael@0: 
michael@0:   int count = 0;
michael@0:   for (size_t index = 0; index < counts_size; ++index) {
michael@0:     int i;
michael@0:     if (!pickle.ReadInt(iter, &i))
michael@0:       return false;
michael@0:     counts_.push_back(i);
michael@0:     count += i;
michael@0:   }
michael@0: 
michael@0:   return true;
michael@0: }
michael@0: 
michael@0: //------------------------------------------------------------------------------
michael@0: // LinearHistogram: This histogram uses a traditional set of evenly spaced
michael@0: // buckets.
michael@0: //------------------------------------------------------------------------------
michael@0: 
michael@0: LinearHistogram::~LinearHistogram() {
michael@0: }
michael@0: 
michael@0: Histogram* LinearHistogram::FactoryGet(const std::string& name,
michael@0:                                        Sample minimum,
michael@0:                                        Sample maximum,
michael@0:                                        size_t bucket_count,
michael@0:                                        Flags flags) {
michael@0:   Histogram* histogram(NULL);
michael@0: 
michael@0:   if (minimum < 1)
michael@0:     minimum = 1;
michael@0:   if (maximum > kSampleType_MAX - 1)
michael@0:     maximum = kSampleType_MAX - 1;
michael@0: 
michael@0:   if (!StatisticsRecorder::FindHistogram(name, &histogram)) {
michael@0:     LinearHistogram* tentative_histogram =
michael@0:         new LinearHistogram(name, minimum, maximum, bucket_count);
michael@0:     tentative_histogram->InitializeBucketRange();
michael@0:     tentative_histogram->SetFlags(flags);
michael@0:     histogram =
michael@0:         StatisticsRecorder::RegisterOrDeleteDuplicate(tentative_histogram);
michael@0:   }
michael@0: 
michael@0:   DCHECK_EQ(LINEAR_HISTOGRAM, histogram->histogram_type());
michael@0:   DCHECK(histogram->HasConstructorArguments(minimum, maximum, bucket_count));
michael@0:   return histogram;
michael@0: }
michael@0: 
michael@0: Histogram* LinearHistogram::FactoryTimeGet(const std::string& name,
michael@0:                                            TimeDelta minimum,
michael@0:                                            TimeDelta maximum,
michael@0:                                            size_t bucket_count,
michael@0:                                            Flags flags) {
michael@0:   return FactoryGet(name, minimum.InMilliseconds(), maximum.InMilliseconds(),
michael@0:                     bucket_count, flags);
michael@0: }
michael@0: 
michael@0: Histogram::ClassType LinearHistogram::histogram_type() const {
michael@0:   return LINEAR_HISTOGRAM;
michael@0: }
michael@0: 
michael@0: void LinearHistogram::Accumulate(Sample value, Count count, size_t index) {
michael@0:   sample_.AccumulateWithLinearStats(value, count, index);
michael@0: }
michael@0: 
michael@0: void LinearHistogram::SetRangeDescriptions(
michael@0:     const DescriptionPair descriptions[]) {
michael@0:   for (int i =0; descriptions[i].description; ++i) {
michael@0:     bucket_description_[descriptions[i].sample] = descriptions[i].description;
michael@0:   }
michael@0: }
michael@0: 
michael@0: LinearHistogram::LinearHistogram(const std::string& name,
michael@0:                                  Sample minimum,
michael@0:                                  Sample maximum,
michael@0:                                  size_t bucket_count)
michael@0:     : Histogram(name, minimum >= 1 ? minimum : 1, maximum, bucket_count) {
michael@0: }
michael@0: 
michael@0: LinearHistogram::LinearHistogram(const std::string& name,
michael@0:                                  TimeDelta minimum,
michael@0:                                  TimeDelta maximum,
michael@0:                                  size_t bucket_count)
michael@0:     : Histogram(name, minimum >= TimeDelta::FromMilliseconds(1) ?
michael@0:                                  minimum : TimeDelta::FromMilliseconds(1),
michael@0:                 maximum, bucket_count) {
michael@0: }
michael@0: 
michael@0: void LinearHistogram::InitializeBucketRange() {
michael@0:   DCHECK_GT(declared_min(), 0);  // 0 is the underflow bucket here.
michael@0:   double min = declared_min();
michael@0:   double max = declared_max();
michael@0:   size_t i;
michael@0:   for (i = 1; i < bucket_count(); ++i) {
michael@0:     double linear_range = (min * (bucket_count() -1 - i) + max * (i - 1)) /
michael@0:                           (bucket_count() - 2);
michael@0:     SetBucketRange(i, static_cast<int> (linear_range + 0.5));
michael@0:   }
michael@0:   ResetRangeChecksum();
michael@0: }
michael@0: 
michael@0: double LinearHistogram::GetBucketSize(Count current, size_t i) const {
michael@0:   DCHECK_GT(ranges(i + 1), ranges(i));
michael@0:   // Adjacent buckets with different widths would have "surprisingly" many (few)
michael@0:   // samples in a histogram if we didn't normalize this way.
michael@0:   double denominator = ranges(i + 1) - ranges(i);
michael@0:   return current/denominator;
michael@0: }
michael@0: 
michael@0: const std::string LinearHistogram::GetAsciiBucketRange(size_t i) const {
michael@0:   int range = ranges(i);
michael@0:   BucketDescriptionMap::const_iterator it = bucket_description_.find(range);
michael@0:   if (it == bucket_description_.end())
michael@0:     return Histogram::GetAsciiBucketRange(i);
michael@0:   return it->second;
michael@0: }
michael@0: 
michael@0: bool LinearHistogram::PrintEmptyBucket(size_t index) const {
michael@0:   return bucket_description_.find(ranges(index)) == bucket_description_.end();
michael@0: }
michael@0: 
michael@0: 
michael@0: //------------------------------------------------------------------------------
michael@0: // This section provides implementation for BooleanHistogram.
michael@0: //------------------------------------------------------------------------------
michael@0: 
michael@0: Histogram* BooleanHistogram::FactoryGet(const std::string& name, Flags flags) {
michael@0:   Histogram* histogram(NULL);
michael@0: 
michael@0:   if (!StatisticsRecorder::FindHistogram(name, &histogram)) {
michael@0:     BooleanHistogram* tentative_histogram = new BooleanHistogram(name);
michael@0:     tentative_histogram->InitializeBucketRange();
michael@0:     tentative_histogram->SetFlags(flags);
michael@0:     histogram =
michael@0:         StatisticsRecorder::RegisterOrDeleteDuplicate(tentative_histogram);
michael@0:   }
michael@0: 
michael@0:   DCHECK_EQ(BOOLEAN_HISTOGRAM, histogram->histogram_type());
michael@0:   return histogram;
michael@0: }
michael@0: 
michael@0: Histogram::ClassType BooleanHistogram::histogram_type() const {
michael@0:   return BOOLEAN_HISTOGRAM;
michael@0: }
michael@0: 
michael@0: void BooleanHistogram::AddBoolean(bool value) {
michael@0:   Add(value ? 1 : 0);
michael@0: }
michael@0: 
michael@0: BooleanHistogram::BooleanHistogram(const std::string& name)
michael@0:     : LinearHistogram(name, 1, 2, 3) {
michael@0: }
michael@0: 
michael@0: void
michael@0: BooleanHistogram::Accumulate(Sample value, Count count, size_t index)
michael@0: {
michael@0:   // Callers will have computed index based on the non-booleanified value.
michael@0:   // So we need to adjust the index manually.
michael@0:   LinearHistogram::Accumulate(!!value, count, value ? 1 : 0);
michael@0: }
michael@0: 
michael@0: //------------------------------------------------------------------------------
michael@0: // FlagHistogram:
michael@0: //------------------------------------------------------------------------------
michael@0: 
michael@0: Histogram *
michael@0: FlagHistogram::FactoryGet(const std::string &name, Flags flags)
michael@0: {
michael@0:   Histogram *h(nullptr);
michael@0: 
michael@0:   if (!StatisticsRecorder::FindHistogram(name, &h)) {
michael@0:     FlagHistogram *fh = new FlagHistogram(name);
michael@0:     fh->InitializeBucketRange();
michael@0:     fh->SetFlags(flags);
michael@0:     size_t zero_index = fh->BucketIndex(0);
michael@0:     fh->LinearHistogram::Accumulate(0, 1, zero_index);
michael@0:     h = StatisticsRecorder::RegisterOrDeleteDuplicate(fh);
michael@0:   }
michael@0: 
michael@0:   return h;
michael@0: }
michael@0: 
michael@0: FlagHistogram::FlagHistogram(const std::string &name)
michael@0:   : BooleanHistogram(name), mSwitched(false) {
michael@0: }
michael@0: 
michael@0: Histogram::ClassType
michael@0: FlagHistogram::histogram_type() const
michael@0: {
michael@0:   return FLAG_HISTOGRAM;
michael@0: }
michael@0: 
michael@0: void
michael@0: FlagHistogram::Accumulate(Sample value, Count count, size_t index)
michael@0: {
michael@0:   if (mSwitched) {
michael@0:     return;
michael@0:   }
michael@0: 
michael@0:   mSwitched = true;
michael@0:   DCHECK_EQ(value, 1);
michael@0:   LinearHistogram::Accumulate(value, 1, index);
michael@0:   size_t zero_index = BucketIndex(0);
michael@0:   LinearHistogram::Accumulate(0, -1, zero_index);
michael@0: }
michael@0: 
michael@0: void
michael@0: FlagHistogram::AddSampleSet(const SampleSet& sample) {
michael@0:   DCHECK_EQ(bucket_count(), sample.size());
michael@0:   // We can't be sure the SampleSet provided came from another FlagHistogram,
michael@0:   // so we take the following steps:
michael@0:   //  - If our flag has already been set do nothing.
michael@0:   //  - Set our flag if the following hold:
michael@0:   //      - The sum of the counts in the provided SampleSet is 1.
michael@0:   //      - The bucket index for that single value is the same as the index where we
michael@0:   //        would place our set flag.
michael@0:   //  - Otherwise, take no action.
michael@0: 
michael@0:   if (mSwitched) {
michael@0:     return;
michael@0:   }
michael@0: 
michael@0:   if (sample.sum() != 1) {
michael@0:     return;
michael@0:   }
michael@0: 
michael@0:   size_t one_index = BucketIndex(1);
michael@0:   if (sample.counts(one_index) == 1) {
michael@0:     Accumulate(1, 1, one_index);
michael@0:   }
michael@0: }
michael@0: //------------------------------------------------------------------------------
michael@0: // CustomHistogram:
michael@0: //------------------------------------------------------------------------------
michael@0: 
michael@0: Histogram* CustomHistogram::FactoryGet(const std::string& name,
michael@0:                                        const std::vector<Sample>& custom_ranges,
michael@0:                                        Flags flags) {
michael@0:   Histogram* histogram(NULL);
michael@0: 
michael@0:   // Remove the duplicates in the custom ranges array.
michael@0:   std::vector<int> ranges = custom_ranges;
michael@0:   ranges.push_back(0);  // Ensure we have a zero value.
michael@0:   std::sort(ranges.begin(), ranges.end());
michael@0:   ranges.erase(std::unique(ranges.begin(), ranges.end()), ranges.end());
michael@0:   if (ranges.size() <= 1) {
michael@0:     DCHECK(false);
michael@0:     // Note that we pushed a 0 in above, so for defensive code....
michael@0:     ranges.push_back(1);  // Put in some data so we can index to [1].
michael@0:   }
michael@0: 
michael@0:   DCHECK_LT(ranges.back(), kSampleType_MAX);
michael@0: 
michael@0:   if (!StatisticsRecorder::FindHistogram(name, &histogram)) {
michael@0:     CustomHistogram* tentative_histogram = new CustomHistogram(name, ranges);
michael@0:     tentative_histogram->InitializedCustomBucketRange(ranges);
michael@0:     tentative_histogram->SetFlags(flags);
michael@0:     histogram =
michael@0:         StatisticsRecorder::RegisterOrDeleteDuplicate(tentative_histogram);
michael@0:   }
michael@0: 
michael@0:   DCHECK_EQ(histogram->histogram_type(), CUSTOM_HISTOGRAM);
michael@0:   DCHECK(histogram->HasConstructorArguments(ranges[1], ranges.back(),
michael@0:                                             ranges.size()));
michael@0:   return histogram;
michael@0: }
michael@0: 
michael@0: Histogram::ClassType CustomHistogram::histogram_type() const {
michael@0:   return CUSTOM_HISTOGRAM;
michael@0: }
michael@0: 
michael@0: CustomHistogram::CustomHistogram(const std::string& name,
michael@0:                                  const std::vector<Sample>& custom_ranges)
michael@0:     : Histogram(name, custom_ranges[1], custom_ranges.back(),
michael@0:                 custom_ranges.size()) {
michael@0:   DCHECK_GT(custom_ranges.size(), 1u);
michael@0:   DCHECK_EQ(custom_ranges[0], 0);
michael@0: }
michael@0: 
michael@0: void CustomHistogram::InitializedCustomBucketRange(
michael@0:     const std::vector<Sample>& custom_ranges) {
michael@0:   DCHECK_GT(custom_ranges.size(), 1u);
michael@0:   DCHECK_EQ(custom_ranges[0], 0);
michael@0:   DCHECK_LE(custom_ranges.size(), bucket_count());
michael@0:   for (size_t index = 0; index < custom_ranges.size(); ++index)
michael@0:     SetBucketRange(index, custom_ranges[index]);
michael@0:   ResetRangeChecksum();
michael@0: }
michael@0: 
michael@0: double CustomHistogram::GetBucketSize(Count current, size_t i) const {
michael@0:   return 1;
michael@0: }
michael@0: 
michael@0: //------------------------------------------------------------------------------
michael@0: // The next section handles global (central) support for all histograms, as well
michael@0: // as startup/teardown of this service.
michael@0: //------------------------------------------------------------------------------
michael@0: 
michael@0: // This singleton instance should be started during the single threaded portion
michael@0: // of main(), and hence it is not thread safe.  It initializes globals to
michael@0: // provide support for all future calls.
michael@0: StatisticsRecorder::StatisticsRecorder() {
michael@0:   DCHECK(!histograms_);
michael@0:   if (lock_ == NULL) {
michael@0:     // This will leak on purpose. It's the only way to make sure we won't race
michael@0:     // against the static uninitialization of the module while one of our
michael@0:     // static methods relying on the lock get called at an inappropriate time
michael@0:     // during the termination phase. Since it's a static data member, we will
michael@0:     // leak one per process, which would be similar to the instance allocated
michael@0:     // during static initialization and released only on  process termination.
michael@0:     lock_ = new base::Lock;
michael@0:   }
michael@0:   base::AutoLock auto_lock(*lock_);
michael@0:   histograms_ = new HistogramMap;
michael@0: }
michael@0: 
michael@0: StatisticsRecorder::~StatisticsRecorder() {
michael@0:   DCHECK(histograms_ && lock_);
michael@0: 
michael@0:   if (dump_on_exit_) {
michael@0:     std::string output;
michael@0:     WriteGraph("", &output);
michael@0:     CHROMIUM_LOG(INFO) << output;
michael@0:   }
michael@0:   // Clean up.
michael@0:   HistogramMap* histograms = NULL;
michael@0:   {
michael@0:     base::AutoLock auto_lock(*lock_);
michael@0:     histograms = histograms_;
michael@0:     histograms_ = NULL;
michael@0:     for (HistogramMap::iterator it = histograms->begin();
michael@0:          histograms->end() != it;
michael@0:          ++it) {
michael@0:       // No other clients permanently hold Histogram references, so we
michael@0:       // have the only one and it is safe to delete it.
michael@0:       delete it->second;
michael@0:     }
michael@0:   }
michael@0:   delete histograms;
michael@0:   // We don't delete lock_ on purpose to avoid having to properly protect
michael@0:   // against it going away after we checked for NULL in the static methods.
michael@0: }
michael@0: 
michael@0: // static
michael@0: bool StatisticsRecorder::IsActive() {
michael@0:   if (lock_ == NULL)
michael@0:     return false;
michael@0:   base::AutoLock auto_lock(*lock_);
michael@0:   return NULL != histograms_;
michael@0: }
michael@0: 
michael@0: Histogram* StatisticsRecorder::RegisterOrDeleteDuplicate(Histogram* histogram) {
michael@0:   DCHECK(histogram->HasValidRangeChecksum());
michael@0:   if (lock_ == NULL)
michael@0:     return histogram;
michael@0:   base::AutoLock auto_lock(*lock_);
michael@0:   if (!histograms_)
michael@0:     return histogram;
michael@0:   const std::string name = histogram->histogram_name();
michael@0:   HistogramMap::iterator it = histograms_->find(name);
michael@0:   // Avoid overwriting a previous registration.
michael@0:   if (histograms_->end() == it) {
michael@0:     (*histograms_)[name] = histogram;
michael@0:   } else {
michael@0:     delete histogram;  // We already have one by this name.
michael@0:     histogram = it->second;
michael@0:   }
michael@0:   return histogram;
michael@0: }
michael@0: 
michael@0: // static
michael@0: void StatisticsRecorder::WriteHTMLGraph(const std::string& query,
michael@0:                                         std::string* output) {
michael@0:   if (!IsActive())
michael@0:     return;
michael@0:   output->append("<html><head><title>About Histograms");
michael@0:   if (!query.empty())
michael@0:     output->append(" - " + query);
michael@0:   output->append("</title>"
michael@0:                  // We'd like the following no-cache... but it doesn't work.
michael@0:                  // "<META HTTP-EQUIV=\"Pragma\" CONTENT=\"no-cache\">"
michael@0:                  "</head><body>");
michael@0: 
michael@0:   Histograms snapshot;
michael@0:   GetSnapshot(query, &snapshot);
michael@0:   for (Histograms::iterator it = snapshot.begin();
michael@0:        it != snapshot.end();
michael@0:        ++it) {
michael@0:     (*it)->WriteHTMLGraph(output);
michael@0:     output->append("<br><hr><br>");
michael@0:   }
michael@0:   output->append("</body></html>");
michael@0: }
michael@0: 
michael@0: // static
michael@0: void StatisticsRecorder::WriteGraph(const std::string& query,
michael@0:                                     std::string* output) {
michael@0:   if (!IsActive())
michael@0:     return;
michael@0:   if (query.length())
michael@0:     StringAppendF(output, "Collections of histograms for %s\n", query.c_str());
michael@0:   else
michael@0:     output->append("Collections of all histograms\n");
michael@0: 
michael@0:   Histograms snapshot;
michael@0:   GetSnapshot(query, &snapshot);
michael@0:   for (Histograms::iterator it = snapshot.begin();
michael@0:        it != snapshot.end();
michael@0:        ++it) {
michael@0:     (*it)->WriteAscii(true, "\n", output);
michael@0:     output->append("\n");
michael@0:   }
michael@0: }
michael@0: 
michael@0: // static
michael@0: void StatisticsRecorder::GetHistograms(Histograms* output) {
michael@0:   if (lock_ == NULL)
michael@0:     return;
michael@0:   base::AutoLock auto_lock(*lock_);
michael@0:   if (!histograms_)
michael@0:     return;
michael@0:   for (HistogramMap::iterator it = histograms_->begin();
michael@0:        histograms_->end() != it;
michael@0:        ++it) {
michael@0:     DCHECK_EQ(it->first, it->second->histogram_name());
michael@0:     output->push_back(it->second);
michael@0:   }
michael@0: }
michael@0: 
michael@0: bool StatisticsRecorder::FindHistogram(const std::string& name,
michael@0:                                        Histogram** histogram) {
michael@0:   if (lock_ == NULL)
michael@0:     return false;
michael@0:   base::AutoLock auto_lock(*lock_);
michael@0:   if (!histograms_)
michael@0:     return false;
michael@0:   HistogramMap::iterator it = histograms_->find(name);
michael@0:   if (histograms_->end() == it)
michael@0:     return false;
michael@0:   *histogram = it->second;
michael@0:   return true;
michael@0: }
michael@0: 
michael@0: // private static
michael@0: void StatisticsRecorder::GetSnapshot(const std::string& query,
michael@0:                                      Histograms* snapshot) {
michael@0:   if (lock_ == NULL)
michael@0:     return;
michael@0:   base::AutoLock auto_lock(*lock_);
michael@0:   if (!histograms_)
michael@0:     return;
michael@0:   for (HistogramMap::iterator it = histograms_->begin();
michael@0:        histograms_->end() != it;
michael@0:        ++it) {
michael@0:     if (it->first.find(query) != std::string::npos)
michael@0:       snapshot->push_back(it->second);
michael@0:   }
michael@0: }
michael@0: 
michael@0: // static
michael@0: StatisticsRecorder::HistogramMap* StatisticsRecorder::histograms_ = NULL;
michael@0: // static
michael@0: base::Lock* StatisticsRecorder::lock_ = NULL;
michael@0: // static
michael@0: bool StatisticsRecorder::dump_on_exit_ = false;
michael@0: 
michael@0: }  // namespace base