Home | History | Annotate | Download | only in metrics
      1 // Copyright (c) 2012 The Chromium Authors. All rights reserved.
      2 // Use of this source code is governed by a BSD-style license that can be
      3 // found in the LICENSE file.
      4 
      5 // Histogram is an object that aggregates statistics, and can summarize them in
      6 // various forms, including ASCII graphical, HTML, and numerically (as a
      7 // vector of numbers corresponding to each of the aggregating buckets).
      8 
      9 // It supports calls to accumulate either time intervals (which are processed
     10 // as integral number of milliseconds), or arbitrary integral units.
     11 
     12 // For Histogram(exponential histogram), LinearHistogram and CustomHistogram,
     13 // the minimum for a declared range is 1 (instead of 0), while the maximum is
     14 // (HistogramBase::kSampleType_MAX - 1). Currently you can declare histograms
     15 // with ranges exceeding those limits (e.g. 0 as minimal or
     16 // HistogramBase::kSampleType_MAX as maximal), but those excesses will be
     17 // silently clamped to those limits (for backwards compatibility with existing
     18 // code). Best practice is to not exceed the limits.
     19 
     20 // Each use of a histogram with the same name will reference the same underlying
     21 // data, so it is safe to record to the same histogram from multiple locations
     22 // in the code. It is a runtime error if all uses of the same histogram do not
     23 // agree exactly in type, bucket size and range.
     24 
     25 // For Histogram and LinearHistogram, the maximum for a declared range should
     26 // always be larger (not equal) than minmal range. Zero and
     27 // HistogramBase::kSampleType_MAX are implicitly added as first and last ranges,
     28 // so the smallest legal bucket_count is 3. However CustomHistogram can have
     29 // bucket count as 2 (when you give a custom ranges vector containing only 1
     30 // range).
     31 // For these 3 kinds of histograms, the max bucket count is always
     32 // (Histogram::kBucketCount_MAX - 1).
     33 
     34 // The buckets layout of class Histogram is exponential. For example, buckets
     35 // might contain (sequentially) the count of values in the following intervals:
     36 // [0,1), [1,2), [2,4), [4,8), [8,16), [16,32), [32,64), [64,infinity)
     37 // That bucket allocation would actually result from construction of a histogram
     38 // for values between 1 and 64, with 8 buckets, such as:
     39 // Histogram count("some name", 1, 64, 8);
     40 // Note that the underflow bucket [0,1) and the overflow bucket [64,infinity)
     41 // are also counted by the constructor in the user supplied "bucket_count"
     42 // argument.
     43 // The above example has an exponential ratio of 2 (doubling the bucket width
     44 // in each consecutive bucket.  The Histogram class automatically calculates
     45 // the smallest ratio that it can use to construct the number of buckets
     46 // selected in the constructor.  An another example, if you had 50 buckets,
     47 // and millisecond time values from 1 to 10000, then the ratio between
     48 // consecutive bucket widths will be approximately somewhere around the 50th
     49 // root of 10000.  This approach provides very fine grain (narrow) buckets
     50 // at the low end of the histogram scale, but allows the histogram to cover a
     51 // gigantic range with the addition of very few buckets.
     52 
     53 // Usually we use macros to define and use a histogram. These macros use a
     54 // pattern involving a function static variable, that is a pointer to a
     55 // histogram.  This static is explicitly initialized on any thread
     56 // that detects a uninitialized (NULL) pointer.  The potentially racy
     57 // initialization is not a problem as it is always set to point to the same
     58 // value (i.e., the FactoryGet always returns the same value).  FactoryGet
     59 // is also completely thread safe, which results in a completely thread safe,
     60 // and relatively fast, set of counters.  To avoid races at shutdown, the static
     61 // pointer is NOT deleted, and we leak the histograms at process termination.
     62 
     63 #ifndef BASE_METRICS_HISTOGRAM_H_
     64 #define BASE_METRICS_HISTOGRAM_H_
     65 
     66 #include <map>
     67 #include <string>
     68 #include <vector>
     69 
     70 #include "base/atomicops.h"
     71 #include "base/base_export.h"
     72 #include "base/basictypes.h"
     73 #include "base/compiler_specific.h"
     74 #include "base/gtest_prod_util.h"
     75 #include "base/logging.h"
     76 #include "base/memory/scoped_ptr.h"
     77 #include "base/metrics/bucket_ranges.h"
     78 #include "base/metrics/histogram_base.h"
     79 #include "base/metrics/histogram_samples.h"
     80 #include "base/time/time.h"
     81 
     82 class Pickle;
     83 class PickleIterator;
     84 
     85 namespace base {
     86 
     87 class Lock;
     88 //------------------------------------------------------------------------------
     89 // Histograms are often put in areas where they are called many many times, and
     90 // performance is critical.  As a result, they are designed to have a very low
     91 // recurring cost of executing (adding additional samples).  Toward that end,
     92 // the macros declare a static pointer to the histogram in question, and only
     93 // take a "slow path" to construct (or find) the histogram on the first run
     94 // through the macro.  We leak the histograms at shutdown time so that we don't
     95 // have to validate using the pointers at any time during the running of the
     96 // process.
     97 
     98 // The following code is generally what a thread-safe static pointer
     99 // initializaion looks like for a histogram (after a macro is expanded).  This
    100 // sample is an expansion (with comments) of the code for
    101 // HISTOGRAM_CUSTOM_COUNTS().
    102 
    103 /*
    104   do {
    105     // The pointer's presence indicates the initialization is complete.
    106     // Initialization is idempotent, so it can safely be atomically repeated.
    107     static base::subtle::AtomicWord atomic_histogram_pointer = 0;
    108 
    109     // Acquire_Load() ensures that we acquire visibility to the pointed-to data
    110     // in the histogrom.
    111     base::Histogram* histogram_pointer(reinterpret_cast<base::Histogram*>(
    112         base::subtle::Acquire_Load(&atomic_histogram_pointer)));
    113 
    114     if (!histogram_pointer) {
    115       // This is the slow path, which will construct OR find the matching
    116       // histogram.  FactoryGet includes locks on a global histogram name map
    117       // and is completely thread safe.
    118       histogram_pointer = base::Histogram::FactoryGet(
    119           name, min, max, bucket_count, base::HistogramBase::kNoFlags);
    120 
    121       // Use Release_Store to ensure that the histogram data is made available
    122       // globally before we make the pointer visible.
    123       // Several threads may perform this store, but the same value will be
    124       // stored in all cases (for a given named/spec'ed histogram).
    125       // We could do this without any barrier, since FactoryGet entered and
    126       // exited a lock after construction, but this barrier makes things clear.
    127       base::subtle::Release_Store(&atomic_histogram_pointer,
    128           reinterpret_cast<base::subtle::AtomicWord>(histogram_pointer));
    129     }
    130 
    131     // Ensure calling contract is upheld, and the name does NOT vary.
    132     DCHECK(histogram_pointer->histogram_name() == constant_histogram_name);
    133 
    134     histogram_pointer->Add(sample);
    135   } while (0);
    136 */
    137 
    138 // The above pattern is repeated in several macros.  The only elements that
    139 // vary are the invocation of the Add(sample) vs AddTime(sample), and the choice
    140 // of which FactoryGet method to use.  The different FactoryGet methods have
    141 // various argument lists, so the function with its argument list is provided as
    142 // a macro argument here.  The name is only used in a DCHECK, to assure that
    143 // callers don't try to vary the name of the histogram (which would tend to be
    144 // ignored by the one-time initialization of the histogtram_pointer).
    145 #define STATIC_HISTOGRAM_POINTER_BLOCK(constant_histogram_name, \
    146                                        histogram_add_method_invocation, \
    147                                        histogram_factory_get_invocation) \
    148   do { \
    149     static base::subtle::AtomicWord atomic_histogram_pointer = 0; \
    150     base::HistogramBase* histogram_pointer( \
    151         reinterpret_cast<base::HistogramBase*>( \
    152             base::subtle::Acquire_Load(&atomic_histogram_pointer))); \
    153     if (!histogram_pointer) { \
    154       histogram_pointer = histogram_factory_get_invocation; \
    155       base::subtle::Release_Store(&atomic_histogram_pointer, \
    156           reinterpret_cast<base::subtle::AtomicWord>(histogram_pointer)); \
    157     } \
    158     DCHECK_EQ(histogram_pointer->histogram_name(), \
    159               std::string(constant_histogram_name)); \
    160     histogram_pointer->histogram_add_method_invocation; \
    161   } while (0)
    162 
    163 
    164 //------------------------------------------------------------------------------
    165 // Provide easy general purpose histogram in a macro, just like stats counters.
    166 // The first four macros use 50 buckets.
    167 
    168 #define HISTOGRAM_TIMES(name, sample) HISTOGRAM_CUSTOM_TIMES( \
    169     name, sample, base::TimeDelta::FromMilliseconds(1), \
    170     base::TimeDelta::FromSeconds(10), 50)
    171 
    172 // For folks that need real specific times, use this to select a precise range
    173 // of times you want plotted, and the number of buckets you want used.
    174 #define HISTOGRAM_CUSTOM_TIMES(name, sample, min, max, bucket_count) \
    175     STATIC_HISTOGRAM_POINTER_BLOCK(name, AddTime(sample), \
    176         base::Histogram::FactoryTimeGet(name, min, max, bucket_count, \
    177                                         base::HistogramBase::kNoFlags))
    178 
    179 #define HISTOGRAM_COUNTS(name, sample) HISTOGRAM_CUSTOM_COUNTS( \
    180     name, sample, 1, 1000000, 50)
    181 
    182 #define HISTOGRAM_COUNTS_100(name, sample) HISTOGRAM_CUSTOM_COUNTS( \
    183     name, sample, 1, 100, 50)
    184 
    185 #define HISTOGRAM_COUNTS_10000(name, sample) HISTOGRAM_CUSTOM_COUNTS( \
    186     name, sample, 1, 10000, 50)
    187 
    188 #define HISTOGRAM_CUSTOM_COUNTS(name, sample, min, max, bucket_count) \
    189     STATIC_HISTOGRAM_POINTER_BLOCK(name, Add(sample), \
    190         base::Histogram::FactoryGet(name, min, max, bucket_count, \
    191                                     base::HistogramBase::kNoFlags))
    192 
    193 #define HISTOGRAM_PERCENTAGE(name, under_one_hundred) \
    194     HISTOGRAM_ENUMERATION(name, under_one_hundred, 101)
    195 
    196 #define HISTOGRAM_BOOLEAN(name, sample) \
    197     STATIC_HISTOGRAM_POINTER_BLOCK(name, AddBoolean(sample), \
    198         base::BooleanHistogram::FactoryGet(name, base::Histogram::kNoFlags))
    199 
    200 // Support histograming of an enumerated value.  The samples should always be
    201 // strictly less than |boundary_value| -- this prevents you from running into
    202 // problems down the line if you add additional buckets to the histogram.  Note
    203 // also that, despite explicitly setting the minimum bucket value to |1| below,
    204 // it is fine for enumerated histograms to be 0-indexed -- this is because
    205 // enumerated histograms should never have underflow.
    206 #define HISTOGRAM_ENUMERATION(name, sample, boundary_value) \
    207     STATIC_HISTOGRAM_POINTER_BLOCK(name, Add(sample), \
    208         base::LinearHistogram::FactoryGet(name, 1, boundary_value, \
    209             boundary_value + 1, base::HistogramBase::kNoFlags))
    210 
    211 // Support histograming of an enumerated value. Samples should be one of the
    212 // std::vector<int> list provided via |custom_ranges|. See comments above
    213 // CustomRanges::FactoryGet about the requirement of |custom_ranges|.
    214 // You can use the helper function CustomHistogram::ArrayToCustomRanges to
    215 // transform a C-style array of valid sample values to a std::vector<int>.
    216 #define HISTOGRAM_CUSTOM_ENUMERATION(name, sample, custom_ranges) \
    217     STATIC_HISTOGRAM_POINTER_BLOCK(name, Add(sample), \
    218         base::CustomHistogram::FactoryGet(name, custom_ranges, \
    219                                           base::HistogramBase::kNoFlags))
    220 
    221 #define HISTOGRAM_MEMORY_KB(name, sample) HISTOGRAM_CUSTOM_COUNTS( \
    222     name, sample, 1000, 500000, 50)
    223 
    224 //------------------------------------------------------------------------------
    225 // Define Debug vs non-debug flavors of macros.
    226 #ifndef NDEBUG
    227 
    228 #define DHISTOGRAM_TIMES(name, sample) HISTOGRAM_TIMES(name, sample)
    229 #define DHISTOGRAM_COUNTS(name, sample) HISTOGRAM_COUNTS(name, sample)
    230 #define DHISTOGRAM_PERCENTAGE(name, under_one_hundred) HISTOGRAM_PERCENTAGE(\
    231     name, under_one_hundred)
    232 #define DHISTOGRAM_CUSTOM_TIMES(name, sample, min, max, bucket_count) \
    233     HISTOGRAM_CUSTOM_TIMES(name, sample, min, max, bucket_count)
    234 #define DHISTOGRAM_CLIPPED_TIMES(name, sample, min, max, bucket_count) \
    235     HISTOGRAM_CLIPPED_TIMES(name, sample, min, max, bucket_count)
    236 #define DHISTOGRAM_CUSTOM_COUNTS(name, sample, min, max, bucket_count) \
    237     HISTOGRAM_CUSTOM_COUNTS(name, sample, min, max, bucket_count)
    238 #define DHISTOGRAM_ENUMERATION(name, sample, boundary_value) \
    239     HISTOGRAM_ENUMERATION(name, sample, boundary_value)
    240 #define DHISTOGRAM_CUSTOM_ENUMERATION(name, sample, custom_ranges) \
    241     HISTOGRAM_CUSTOM_ENUMERATION(name, sample, custom_ranges)
    242 
    243 #else  // NDEBUG
    244 // Keep a mention of passed variables to avoid unused variable warnings in
    245 // release build if these variables are only used in macros.
    246 #define DISCARD_2_ARGUMENTS(a, b) \
    247   while (0) { \
    248     static_cast<void>(a); \
    249     static_cast<void>(b); \
    250  }
    251 #define DISCARD_3_ARGUMENTS(a, b, c) \
    252   while (0) { \
    253     static_cast<void>(a); \
    254     static_cast<void>(b); \
    255     static_cast<void>(c); \
    256  }
    257 #define DISCARD_5_ARGUMENTS(a, b, c, d ,e) \
    258   while (0) { \
    259     static_cast<void>(a); \
    260     static_cast<void>(b); \
    261     static_cast<void>(c); \
    262     static_cast<void>(d); \
    263     static_cast<void>(e); \
    264  }
    265 #define DHISTOGRAM_TIMES(name, sample) \
    266     DISCARD_2_ARGUMENTS(name, sample)
    267 
    268 #define DHISTOGRAM_COUNTS(name, sample) \
    269     DISCARD_2_ARGUMENTS(name, sample)
    270 
    271 #define DHISTOGRAM_PERCENTAGE(name, under_one_hundred) \
    272     DISCARD_2_ARGUMENTS(name, under_one_hundred)
    273 
    274 #define DHISTOGRAM_CUSTOM_TIMES(name, sample, min, max, bucket_count) \
    275     DISCARD_5_ARGUMENTS(name, sample, min, max, bucket_count)
    276 
    277 #define DHISTOGRAM_CLIPPED_TIMES(name, sample, min, max, bucket_count) \
    278     DISCARD_5_ARGUMENTS(name, sample, min, max, bucket_count)
    279 
    280 #define DHISTOGRAM_CUSTOM_COUNTS(name, sample, min, max, bucket_count) \
    281     DISCARD_5_ARGUMENTS(name, sample, min, max, bucket_count)
    282 
    283 #define DHISTOGRAM_ENUMERATION(name, sample, boundary_value) \
    284     DISCARD_3_ARGUMENTS(name, sample, boundary_value)
    285 
    286 #define DHISTOGRAM_CUSTOM_ENUMERATION(name, sample, custom_ranges) \
    287     DISCARD_3_ARGUMENTS(name, sample, custom_ranges)
    288 
    289 #endif  // NDEBUG
    290 
    291 //------------------------------------------------------------------------------
    292 // The following macros provide typical usage scenarios for callers that wish
    293 // to record histogram data, and have the data submitted/uploaded via UMA.
    294 // Not all systems support such UMA, but if they do, the following macros
    295 // should work with the service.
    296 
    297 #define UMA_HISTOGRAM_TIMES(name, sample) UMA_HISTOGRAM_CUSTOM_TIMES( \
    298     name, sample, base::TimeDelta::FromMilliseconds(1), \
    299     base::TimeDelta::FromSeconds(10), 50)
    300 
    301 #define UMA_HISTOGRAM_MEDIUM_TIMES(name, sample) UMA_HISTOGRAM_CUSTOM_TIMES( \
    302     name, sample, base::TimeDelta::FromMilliseconds(10), \
    303     base::TimeDelta::FromMinutes(3), 50)
    304 
    305 // Use this macro when times can routinely be much longer than 10 seconds.
    306 #define UMA_HISTOGRAM_LONG_TIMES(name, sample) UMA_HISTOGRAM_CUSTOM_TIMES( \
    307     name, sample, base::TimeDelta::FromMilliseconds(1), \
    308     base::TimeDelta::FromHours(1), 50)
    309 
    310 #define UMA_HISTOGRAM_CUSTOM_TIMES(name, sample, min, max, bucket_count) \
    311     STATIC_HISTOGRAM_POINTER_BLOCK(name, AddTime(sample), \
    312         base::Histogram::FactoryTimeGet(name, min, max, bucket_count, \
    313             base::HistogramBase::kUmaTargetedHistogramFlag))
    314 
    315 #define UMA_HISTOGRAM_COUNTS(name, sample) UMA_HISTOGRAM_CUSTOM_COUNTS( \
    316     name, sample, 1, 1000000, 50)
    317 
    318 #define UMA_HISTOGRAM_COUNTS_100(name, sample) UMA_HISTOGRAM_CUSTOM_COUNTS( \
    319     name, sample, 1, 100, 50)
    320 
    321 #define UMA_HISTOGRAM_COUNTS_10000(name, sample) UMA_HISTOGRAM_CUSTOM_COUNTS( \
    322     name, sample, 1, 10000, 50)
    323 
    324 #define UMA_HISTOGRAM_CUSTOM_COUNTS(name, sample, min, max, bucket_count) \
    325     STATIC_HISTOGRAM_POINTER_BLOCK(name, Add(sample), \
    326         base::Histogram::FactoryGet(name, min, max, bucket_count, \
    327             base::HistogramBase::kUmaTargetedHistogramFlag))
    328 
    329 #define UMA_HISTOGRAM_MEMORY_KB(name, sample) UMA_HISTOGRAM_CUSTOM_COUNTS( \
    330     name, sample, 1000, 500000, 50)
    331 
    332 #define UMA_HISTOGRAM_MEMORY_MB(name, sample) UMA_HISTOGRAM_CUSTOM_COUNTS( \
    333     name, sample, 1, 1000, 50)
    334 
    335 #define UMA_HISTOGRAM_PERCENTAGE(name, under_one_hundred) \
    336     UMA_HISTOGRAM_ENUMERATION(name, under_one_hundred, 101)
    337 
    338 #define UMA_HISTOGRAM_BOOLEAN(name, sample) \
    339     STATIC_HISTOGRAM_POINTER_BLOCK(name, AddBoolean(sample), \
    340         base::BooleanHistogram::FactoryGet(name, \
    341             base::HistogramBase::kUmaTargetedHistogramFlag))
    342 
    343 // The samples should always be strictly less than |boundary_value|.  For more
    344 // details, see the comment for the |HISTOGRAM_ENUMERATION| macro, above.
    345 #define UMA_HISTOGRAM_ENUMERATION(name, sample, boundary_value) \
    346     STATIC_HISTOGRAM_POINTER_BLOCK(name, Add(sample), \
    347         base::LinearHistogram::FactoryGet(name, 1, boundary_value, \
    348             boundary_value + 1, base::HistogramBase::kUmaTargetedHistogramFlag))
    349 
    350 #define UMA_HISTOGRAM_CUSTOM_ENUMERATION(name, sample, custom_ranges) \
    351     STATIC_HISTOGRAM_POINTER_BLOCK(name, Add(sample), \
    352         base::CustomHistogram::FactoryGet(name, custom_ranges, \
    353             base::HistogramBase::kUmaTargetedHistogramFlag))
    354 
    355 //------------------------------------------------------------------------------
    356 
    357 class BucketRanges;
    358 class SampleVector;
    359 
    360 class BooleanHistogram;
    361 class CustomHistogram;
    362 class Histogram;
    363 class LinearHistogram;
    364 
    365 class BASE_EXPORT Histogram : public HistogramBase {
    366  public:
    367   // Initialize maximum number of buckets in histograms as 16,384.
    368   static const size_t kBucketCount_MAX;
    369 
    370   typedef std::vector<Count> Counts;
    371 
    372   //----------------------------------------------------------------------------
    373   // For a valid histogram, input should follow these restrictions:
    374   // minimum > 0 (if a minimum below 1 is specified, it will implicitly be
    375   //              normalized up to 1)
    376   // maximum > minimum
    377   // buckets > 2 [minimum buckets needed: underflow, overflow and the range]
    378   // Additionally,
    379   // buckets <= (maximum - minimum + 2) - this is to ensure that we don't have
    380   // more buckets than the range of numbers; having more buckets than 1 per
    381   // value in the range would be nonsensical.
    382   static HistogramBase* FactoryGet(const std::string& name,
    383                                    Sample minimum,
    384                                    Sample maximum,
    385                                    size_t bucket_count,
    386                                    int32 flags);
    387   static HistogramBase* FactoryTimeGet(const std::string& name,
    388                                        base::TimeDelta minimum,
    389                                        base::TimeDelta maximum,
    390                                        size_t bucket_count,
    391                                        int32 flags);
    392 
    393   // Time call for use with DHISTOGRAM*.
    394   // Returns TimeTicks::Now() in debug and TimeTicks() in release build.
    395   static TimeTicks DebugNow();
    396 
    397   static void InitializeBucketRanges(Sample minimum,
    398                                      Sample maximum,
    399                                      BucketRanges* ranges);
    400 
    401   // This constant if for FindCorruption. Since snapshots of histograms are
    402   // taken asynchronously relative to sampling, and our counting code currently
    403   // does not prevent race conditions, it is pretty likely that we'll catch a
    404   // redundant count that doesn't match the sample count.  We allow for a
    405   // certain amount of slop before flagging this as an inconsistency. Even with
    406   // an inconsistency, we'll snapshot it again (for UMA in about a half hour),
    407   // so we'll eventually get the data, if it was not the result of a corruption.
    408   static const int kCommonRaceBasedCountMismatch;
    409 
    410   // Check to see if bucket ranges, counts and tallies in the snapshot are
    411   // consistent with the bucket ranges and checksums in our histogram.  This can
    412   // produce a false-alarm if a race occurred in the reading of the data during
    413   // a SnapShot process, but should otherwise be false at all times (unless we
    414   // have memory over-writes, or DRAM failures).
    415   virtual int FindCorruption(const HistogramSamples& samples) const OVERRIDE;
    416 
    417   //----------------------------------------------------------------------------
    418   // Accessors for factory constuction, serialization and testing.
    419   //----------------------------------------------------------------------------
    420   Sample declared_min() const { return declared_min_; }
    421   Sample declared_max() const { return declared_max_; }
    422   virtual Sample ranges(size_t i) const;
    423   virtual size_t bucket_count() const;
    424   const BucketRanges* bucket_ranges() const { return bucket_ranges_; }
    425 
    426   // This function validates histogram construction arguments. It returns false
    427   // if some of the arguments are totally bad.
    428   // Note. Currently it allow some bad input, e.g. 0 as minimum, but silently
    429   // converts it to good input: 1.
    430   // TODO(kaiwang): Be more restrict and return false for any bad input, and
    431   // make this a readonly validating function.
    432   static bool InspectConstructionArguments(const std::string& name,
    433                                            Sample* minimum,
    434                                            Sample* maximum,
    435                                            size_t* bucket_count);
    436 
    437   // HistogramBase implementation:
    438   virtual HistogramType GetHistogramType() const OVERRIDE;
    439   virtual bool HasConstructionArguments(
    440       Sample expected_minimum,
    441       Sample expected_maximum,
    442       size_t expected_bucket_count) const OVERRIDE;
    443   virtual void Add(Sample value) OVERRIDE;
    444   virtual scoped_ptr<HistogramSamples> SnapshotSamples() const OVERRIDE;
    445   virtual void AddSamples(const HistogramSamples& samples) OVERRIDE;
    446   virtual bool AddSamplesFromPickle(PickleIterator* iter) OVERRIDE;
    447   virtual void WriteHTMLGraph(std::string* output) const OVERRIDE;
    448   virtual void WriteAscii(std::string* output) const OVERRIDE;
    449 
    450  protected:
    451   // |ranges| should contain the underflow and overflow buckets. See top
    452   // comments for example.
    453   Histogram(const std::string& name,
    454             Sample minimum,
    455             Sample maximum,
    456             const BucketRanges* ranges);
    457 
    458   virtual ~Histogram();
    459 
    460   // HistogramBase implementation:
    461   virtual bool SerializeInfoImpl(Pickle* pickle) const OVERRIDE;
    462 
    463   // Method to override to skip the display of the i'th bucket if it's empty.
    464   virtual bool PrintEmptyBucket(size_t index) const;
    465 
    466   // Get normalized size, relative to the ranges(i).
    467   virtual double GetBucketSize(Count current, size_t i) const;
    468 
    469   // Return a string description of what goes in a given bucket.
    470   // Most commonly this is the numeric value, but in derived classes it may
    471   // be a name (or string description) given to the bucket.
    472   virtual const std::string GetAsciiBucketRange(size_t it) const;
    473 
    474  private:
    475   // Allow tests to corrupt our innards for testing purposes.
    476   FRIEND_TEST_ALL_PREFIXES(HistogramTest, BoundsTest);
    477   FRIEND_TEST_ALL_PREFIXES(HistogramTest, BucketPlacementTest);
    478   FRIEND_TEST_ALL_PREFIXES(HistogramTest, CorruptBucketBounds);
    479   FRIEND_TEST_ALL_PREFIXES(HistogramTest, CorruptSampleCounts);
    480   FRIEND_TEST_ALL_PREFIXES(HistogramTest, NameMatchTest);
    481 
    482   friend class StatisticsRecorder;  // To allow it to delete duplicates.
    483   friend class StatisticsRecorderTest;
    484 
    485   friend BASE_EXPORT_PRIVATE HistogramBase* DeserializeHistogramInfo(
    486       PickleIterator* iter);
    487   static HistogramBase* DeserializeInfoImpl(PickleIterator* iter);
    488 
    489   // Implementation of SnapshotSamples function.
    490   scoped_ptr<SampleVector> SnapshotSampleVector() const;
    491 
    492   //----------------------------------------------------------------------------
    493   // Helpers for emitting Ascii graphic.  Each method appends data to output.
    494 
    495   void WriteAsciiImpl(bool graph_it,
    496                       const std::string& newline,
    497                       std::string* output) const;
    498 
    499   // Find out how large (graphically) the largest bucket will appear to be.
    500   double GetPeakBucketSize(const SampleVector& samples) const;
    501 
    502   // Write a common header message describing this histogram.
    503   void WriteAsciiHeader(const SampleVector& samples,
    504                         Count sample_count,
    505                         std::string* output) const;
    506 
    507   // Write information about previous, current, and next buckets.
    508   // Information such as cumulative percentage, etc.
    509   void WriteAsciiBucketContext(const int64 past, const Count current,
    510                                const int64 remaining, const size_t i,
    511                                std::string* output) const;
    512 
    513   // WriteJSON calls these.
    514   virtual void GetParameters(DictionaryValue* params) const OVERRIDE;
    515 
    516   virtual void GetCountAndBucketData(Count* count,
    517                                      int64* sum,
    518                                      ListValue* buckets) const OVERRIDE;
    519 
    520   // Does not own this object. Should get from StatisticsRecorder.
    521   const BucketRanges* bucket_ranges_;
    522 
    523   Sample declared_min_;  // Less than this goes into the first bucket.
    524   Sample declared_max_;  // Over this goes into the last bucket.
    525 
    526   // Finally, provide the state that changes with the addition of each new
    527   // sample.
    528   scoped_ptr<SampleVector> samples_;
    529 
    530   DISALLOW_COPY_AND_ASSIGN(Histogram);
    531 };
    532 
    533 //------------------------------------------------------------------------------
    534 
    535 // LinearHistogram is a more traditional histogram, with evenly spaced
    536 // buckets.
    537 class BASE_EXPORT LinearHistogram : public Histogram {
    538  public:
    539   virtual ~LinearHistogram();
    540 
    541   /* minimum should start from 1. 0 is as minimum is invalid. 0 is an implicit
    542      default underflow bucket. */
    543   static HistogramBase* FactoryGet(const std::string& name,
    544                                    Sample minimum,
    545                                    Sample maximum,
    546                                    size_t bucket_count,
    547                                    int32 flags);
    548   static HistogramBase* FactoryTimeGet(const std::string& name,
    549                                        TimeDelta minimum,
    550                                        TimeDelta maximum,
    551                                        size_t bucket_count,
    552                                        int32 flags);
    553 
    554   struct DescriptionPair {
    555     Sample sample;
    556     const char* description;  // Null means end of a list of pairs.
    557   };
    558 
    559   // Create a LinearHistogram and store a list of number/text values for use in
    560   // writing the histogram graph.
    561   // |descriptions| can be NULL, which means no special descriptions to set. If
    562   // it's not NULL, the last element in the array must has a NULL in its
    563   // "description" field.
    564   static HistogramBase* FactoryGetWithRangeDescription(
    565       const std::string& name,
    566       Sample minimum,
    567       Sample maximum,
    568       size_t bucket_count,
    569       int32 flags,
    570       const DescriptionPair descriptions[]);
    571 
    572   static void InitializeBucketRanges(Sample minimum,
    573                                      Sample maximum,
    574                                      BucketRanges* ranges);
    575 
    576   // Overridden from Histogram:
    577   virtual HistogramType GetHistogramType() const OVERRIDE;
    578 
    579  protected:
    580   LinearHistogram(const std::string& name,
    581                   Sample minimum,
    582                   Sample maximum,
    583                   const BucketRanges* ranges);
    584 
    585   virtual double GetBucketSize(Count current, size_t i) const OVERRIDE;
    586 
    587   // If we have a description for a bucket, then return that.  Otherwise
    588   // let parent class provide a (numeric) description.
    589   virtual const std::string GetAsciiBucketRange(size_t i) const OVERRIDE;
    590 
    591   // Skip printing of name for numeric range if we have a name (and if this is
    592   // an empty bucket).
    593   virtual bool PrintEmptyBucket(size_t index) const OVERRIDE;
    594 
    595  private:
    596   friend BASE_EXPORT_PRIVATE HistogramBase* DeserializeHistogramInfo(
    597       PickleIterator* iter);
    598   static HistogramBase* DeserializeInfoImpl(PickleIterator* iter);
    599 
    600   // For some ranges, we store a printable description of a bucket range.
    601   // If there is no desciption, then GetAsciiBucketRange() uses parent class
    602   // to provide a description.
    603   typedef std::map<Sample, std::string> BucketDescriptionMap;
    604   BucketDescriptionMap bucket_description_;
    605 
    606   DISALLOW_COPY_AND_ASSIGN(LinearHistogram);
    607 };
    608 
    609 //------------------------------------------------------------------------------
    610 
    611 // BooleanHistogram is a histogram for booleans.
    612 class BASE_EXPORT BooleanHistogram : public LinearHistogram {
    613  public:
    614   static HistogramBase* FactoryGet(const std::string& name, int32 flags);
    615 
    616   virtual HistogramType GetHistogramType() const OVERRIDE;
    617 
    618  private:
    619   BooleanHistogram(const std::string& name, const BucketRanges* ranges);
    620 
    621   friend BASE_EXPORT_PRIVATE HistogramBase* DeserializeHistogramInfo(
    622       PickleIterator* iter);
    623   static HistogramBase* DeserializeInfoImpl(PickleIterator* iter);
    624 
    625   DISALLOW_COPY_AND_ASSIGN(BooleanHistogram);
    626 };
    627 
    628 //------------------------------------------------------------------------------
    629 
    630 // CustomHistogram is a histogram for a set of custom integers.
    631 class BASE_EXPORT CustomHistogram : public Histogram {
    632  public:
    633   // |custom_ranges| contains a vector of limits on ranges. Each limit should be
    634   // > 0 and < kSampleType_MAX. (Currently 0 is still accepted for backward
    635   // compatibility). The limits can be unordered or contain duplication, but
    636   // client should not depend on this.
    637   static HistogramBase* FactoryGet(const std::string& name,
    638                                    const std::vector<Sample>& custom_ranges,
    639                                    int32 flags);
    640 
    641   // Overridden from Histogram:
    642   virtual HistogramType GetHistogramType() const OVERRIDE;
    643 
    644   // Helper method for transforming an array of valid enumeration values
    645   // to the std::vector<int> expected by HISTOGRAM_CUSTOM_ENUMERATION.
    646   // This function ensures that a guard bucket exists right after any
    647   // valid sample value (unless the next higher sample is also a valid value),
    648   // so that invalid samples never fall into the same bucket as valid samples.
    649   // TODO(kaiwang): Change name to ArrayToCustomEnumRanges.
    650   static std::vector<Sample> ArrayToCustomRanges(const Sample* values,
    651                                                  size_t num_values);
    652  protected:
    653   CustomHistogram(const std::string& name,
    654                   const BucketRanges* ranges);
    655 
    656   // HistogramBase implementation:
    657   virtual bool SerializeInfoImpl(Pickle* pickle) const OVERRIDE;
    658 
    659   virtual double GetBucketSize(Count current, size_t i) const OVERRIDE;
    660 
    661  private:
    662   friend BASE_EXPORT_PRIVATE HistogramBase* DeserializeHistogramInfo(
    663       PickleIterator* iter);
    664   static HistogramBase* DeserializeInfoImpl(PickleIterator* iter);
    665 
    666   static bool ValidateCustomRanges(const std::vector<Sample>& custom_ranges);
    667   static BucketRanges* CreateBucketRangesFromCustomRanges(
    668       const std::vector<Sample>& custom_ranges);
    669 
    670   DISALLOW_COPY_AND_ASSIGN(CustomHistogram);
    671 };
    672 
    673 }  // namespace base
    674 
    675 #endif  // BASE_METRICS_HISTOGRAM_H_
    676