MySQL++

Check-in [dd35735f4f]
Login

Many hyperlinks are disabled.
Use anonymous login to enable hyperlinks.

Overview
Comment:Removed Wishlist file: a subset of its contents is now in the Fossil wiki and ticket trackers. Dropped items are those we probably will never get to. If they come up again, they can be added as tickets.
Downloads: Tarball | ZIP archive | SQL archive
Timelines: family | ancestors | descendants | both | trunk
Files: files | file ages | folders
SHA3-256:dd35735f4f335b06675b7e07d483a6ed7230ad98b619955cda88d565d3289347
User & Date: tangent 2018-07-26 07:13:23
Context
2018-07-26
15:25
Converted the plain-text ChangeLog file to Markdown format, adding only as much formatting as needed to avoid rendering problems in Vim's markdown mode and in Fossil's web rendering. Updated all source code and documentation input file references to the file to now use "ChangeLog.md". (References from Fossil wiki documents will be updated next.) check-in: 7955e40246 user: tangent tags: trunk
07:13
Removed Wishlist file: a subset of its contents is now in the Fossil wiki and ticket trackers. Dropped items are those we probably will never get to. If they come up again, they can be added as tickets. check-in: dd35735f4f user: tangent tags: trunk
2018-07-18
21:01
DBDriver::shutdown() now executes a SHUTDOWN query on MySQL 8.0.0 and higher due to removal of the C API function mysql_shutdown() in that release. They put it back in 8.0.1, but only due to end-user complaints, and they say it'll go away for good eventually, so we prefer to deploy and test this workaround before then. check-in: b238cd34c2 user: tangent tags: trunk
Changes
Hide Diffs Unified Diffs Ignore Whitespace Patch

Deleted Wishlist.

1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
316
317
318
319
320
321
322
323
324
325
326
327
328
329
330
331
332
333
334
335
336
337
338
339
340
341
342
343
344
345
346
347
348
349
350
351
352
353
354
355
356
357
358
359
360
361
362
363
364
365
366
367
368
369
370
371
372
373
374
375
376
377
378
379
380
381
382
383
384
385
386
387
388
389
390
391
392
393
394
395
396
397
398
399
400
401
402
403
404
405
406
407
408
409
410
411
412
413
414
415
416
417
418
419
420
421
422
423
424
425
426
427
428
429
430
431
432
433
434
435
436
437
438
439
440
441
442
443
444
445
446
447
448
449
450
451
452
453
454
455
456
457
458
459
460
461
462
463
464
465
466
467
468
469
470
471
472
473
474
475
476
477
478
479
480
481
482
483
484
485
486
487
488
489
490
491
492
493
494
495
496
497
498
499
500
501
502
503
504
505
506
507
508
509
510
511
512
513
514
515
516
517
518
519
520
521
522
523
524
525
Patches for any of these thoughtfully considered!  See the HACKERS.txt
file for instructions on sending patches.

Any Version
-----------
    o The DateTime family's time_t conversion operators should detect
      DATETIME("0") type values and return time_t(0) rather than
      attempt to stuff an out-of-range value into a smaller box.

    o Any time you must hand-roll some SQL code in your program,
      consider whether it could be generalized to a widely-useful
      API feature.

    o Suppress DOS line-ending related diffs from examples/cgi_jpeg
      output when running dtest on Windows.  Check for -D?

    o Need to link statically to connect to MySQL Embedded?

      http://stackoverflow.com/questions/672451/

    o When Bakefile allows, bring examples/vstudio/* into the top-level
      build system.  This will let us generate separate project
      files for each VC++ version we support, let us use MYSQL_WIN_DIR
      variable instead of hard-coded paths, and build against the local
      version of MySQL++ instead of requiring it to be installed first.
      Should probably drop the MFC example project, to avoid requiring
      Visual C++ Professional.

    o Query::storein(slist<T>&) is unusable.  As a template method, it
      must be defined in a header file; we cannot #include config.h from
      a header, thus the proper HAVE macro that would let us define this
      template method is never defined.

      One solution is to create lib/slist.h.in, parameterized by the
      detected slist type name and the header file defining it.  This
      will #include the proper header file, define SList<T>::Type (a
      template typdedef: http://www.gotw.ca/gotw/079.htm), and #define
      MYSQLPP_HAVE_SLIST.  Then Query::storein() can be wrapped by an
      ifdef checking for MYSQLPP_HAVE_SLIST, only defined when the
      configure script found a suitable slist type.

      Also create a default lib/slist.h file, checked into svn as
      lib/slistdef.h and copied to slist.h on tarball creation.  Remove
      this file early in configure script run, so we're forced to
      overwrite its contents with detected values.  Default version
      contains ifdefs for non-autoconf platforms where we know what
      slist definition is available on that platform.  Xcode, for
      instance, will let us use <ext/slist>.

    o Add ctor and assignment operator that takes mysqlpp::null.  You
      now have to call a 4-argument version of either the ctor or
      assign() to get a NULL SQL string.

    o The escape_q() functions aren't returning the right value when
      dealing with Null<T> wrapped types, such as sql_blob_null.

    o The current method SSQLS uses to compare floating point numbers
      is highly dubious.  It just subtracts them, and checks that
      the absolute difference is under some threshold.  The manual
      warns that this is fine for "human scale" applications, but even
      that's not actually true.  It means that if Larry Ellison loses
      a hundredth of a penny in his couch, it is somehow significant.
      I have no idea how much money Larry Ellison is comfortable losing
      to his couch cushions, but it's probably closer to 1 ppm than
      the current threshold, which is 100 parts per quadrillion on
      the scale of $1 bn.

      For backards compatibility, we should keep this method, but we
      should add these two more mathematically sound methods:

      - Percentage: Divide the smaller number into the larger, then
        compare against a threshold.  The default should be
        something like 1.000001 (1 ppm), which lets us make much
        finer distinctions without running out of precision, even
        with single-precision numbers counting Bill Gates' losses to
        his couch cushions.

      - Logarithmic, or "Bels": Same as percentage, but on a log10
        scale so it works better for FP numbers, which are based on
        powers of 10.  Logarithms are more costly than division,
        and we still need a division for this, so it shouldn't be
        the new default.

        1 ppm is ~4.3e-7, which is below what single-precision FP
        can distinguish.  Increasing the threshold to a value you
        *can* distinghish with a 32-bit IEEE float makes it ignore
        significant amounts of money in Carlos Slim's couch cusions.
        (Hundreds of dollars.)  Therefore, we should use something
        like 1e-7 or 1e-8 anyway, and make it clear that the default
        threshold is only suitable for doubles.
        
        Someone using single precision FP should increase the threshold
        to 1e-5 or so.  Such a person would be assumed to know what
        they're doing.

        It's probably more efficient to change the algorithm from:

           double diff = log10(a > b ? a / b : b / a);

        to:

          double diff = fabs(log10(a / b));

        Logarithms give the same magnitude result for a/b as b/a,
        differing only in sign.  fabs() is probably implemented as
        an intrinsic that just clears a single bit, which should be
        cheaper than a floating point comparison followed by a jump.

        With suitable tuning, this method would allow you to
        distinguish the change lost by a single Pentagon contractor's
        lobbyist to a single couch, on a single occasion, as compared
        to the combined net worth of all Pentagon contractors and
        their employees, assigns, mistresses, and, ah, hired help.
        If you use doubles, anyway.


v3.3 Plan: Finish SSQLS v2
--------------------------
    See http://lists.mysql.com/plusplus/6929 for high-level plan.

    o Goal: Restore VC++ 2003 compatibility with SSQLS, lost with v1

    o C++ code generator, from walking DSL parse tree:

      examples/stock.ssqls gives ssqls_stock.h containing:

          class SQLStock : public mysqlpp::SsqlsBase
          {
          public:
              SQLStock(Connection* conn = 0); // default ctor
              SQLStock(const SQLStock& other);
              SQLStock(const mysqlpp::Row& row,
                      Connection* conn = 0); // full init from query
              SQLStock(mysqlpp::sql_bigint key1); // exemplar creation
              SQLStock(Connection* conn,
                      mysqlpp::sql_bigint key1) // calls load();
              SQLStock(
                  mysqlpp::sql_bigint f1,
                  mysqlpp::sql_bigint f2,
                  mysqlpp::sql_double f3,
                  mysqlpp::sql_double f4,
                  const mysqlpp::sql_date& f5,
                  const mysqlpp::sql_mediumtext& f6); // full init
              SQLStock(Connection* conn,
                  mysqlpp::sql_bigint f2,
                  mysqlpp::sql_double f3,
                  mysqlpp::sql_double f4,
                  const mysqlpp::sql_date& f5,
                  const mysqlpp::sql_mediumtext& f6); // calls create()
              SQLStock(Connection* conn,
                  mysqlpp::sql_bigint f1,
                  mysqlpp::sql_bigint f2,
                  mysqlpp::sql_double f3,
                  mysqlpp::sql_double f4,
                  const mysqlpp::sql_date& f5,
                  const mysqlpp::sql_mediumtext& f6); // calls save()

              bool createTable(Connection* conn = 0) const;
              const char* getTableName() const
                      { return class_table_name_ || table(); }
              void setTableName(const char* name)
                      { instance_table(name); }
              static void setTableName(const char* name)
                      { class_table_name_ = name; }

              std::ostream& equal_list(std::ostream& os) const;
              std::ostream& json(std::ostream& os) const;
              std::ostream& name_list(std::ostream& os) const;
              std::ostream& value_list(std::ostream& os) const;
              std::ostream& xml(std::ostream& os) const;

              mysqlpp::sql_bigint getId() const;
              mysqlpp::sql_bigint getNum() const;
              mysqlpp::sql_double getWeight() const;
              mysqlpp::sql_double getPrice() const;
              const mysqlpp::sql_date& getSdate() const;
              const mysqlpp::sql_mediumtext& getDescription() const;

              void setId(mysqlpp::sql_bigint value);
              void setNum(mysqlpp::sql_bigint value);
              void setWeight(mysqlpp::sql_double value);
              void setPrice(mysqlpp::sql_double value);
              void setSdate(const mysqlpp::sql_date& value);
              void setDescripion(const mysqlpp::sql_mediumtext& value);

              bool operator<(const SQLStock& rhs) const;
              SQLStock& operator=(const SQLStock& rhs);

          protected:
              mysqlpp::sql_bigint id_;
              mysqlpp::sql_bigint num_;
              mysqlpp::sql_double weight_;
              mysqlpp::sql_double price_;
              mysqlpp::sql_date sdate_;
              mysqlpp::sql_mediumtext description_;

              std::bitset<6> set_fields_;

          private:
              static const char* class_table_name_;
          };

      ...and ssqls_stock.cc, containing implementation for same.

    o Ensure we're using case-insensitive SQL column to C++ field name
      matching.  Column names aren't case-sensitive in SQL.

    o SQL table updater/builder, -T option, taking .ssqls and
      creating or updating the DB table to match.

    o If using accessors, generate "std::bitset<num_fields> is_set_",
      and set the appropriate bit when calling each setFoo() so we
      can intuit which fields were set.  Probably also need an enum:

          enum FieldIndices {
              id_field_index_,
              num_field_index_,
              ...
          };

      This will allow Query::select(), for instance, to figure out that
      we want it to select by a non-key field, returning all matches.

    o Define operator<< for SSQLS and a set of manipulators which
      govern whether the operator calls equal_list(), json(),
      name_list(), value_list(), or xml() to do the actual insertion.

    o Define operator<< for sequence_container<SsqlsBase> and similar
      for associative containers.  Give mysql(1)-like ASCII grid or
      some other table format.

    o Define operator>> for SSQLS, taking XML as input, in the form
      emitted via xml().  expat uses the new BSD license, so maybe
      we can just drop it in the tree, with an option to use the
      platform expat on autoconf systems.

    o MySQL table metadata to SSQLSv2 data structure translator.
      (-s, -u, -p and -t flag support.)  Add this to dtest, extracting
      stock.ssqls definition from DB.

    o Replace Query's template methods taking SSQLSes with concrete
      methods taking const SsqlsBase&.

    o Create Query::remove(const SsqlsBase&)

    o Try to remove of 'explicit' from Date, DateTime and Time
      ctors taking stringish types.  (Can't do it for real until v4,
      but we can lay the ground work here.)

    o Auto-create() ctor: if there is an auto_increment field,
      populate it on query success.

    o Detect ctor conflicts corresponding to sql_create_N(N, N...)
      in SSQLSv1.  ssqlsxlat can be smart enough to just not emit
      duplicate ctors.

    o Special case of ctor suppression: if the number of key fields
      equals the total number of fields, you get an auto-load() ctor,
      not auto-update().

    o Replace CREATE TABLE SQL in resetdb with create_table() calls.

    o Option to use integer indices into Row when populating?  Only
      bother if it gives a speed advantage we can actually see,
      because it removes all the dynamic typing advantages we got
      with the change to field name indices in v3.0.

    o Try to add Query::storein(container, ssqls), which generates
      SELECT * from {ssqls.table()} and stores the result.  May not be
      possible due to existing overloads, but try.  If it works, use
      this form in the userman Overview section, saving one LOC.

    o Convert SSQLS v1 examples to v2.  SSQLS v2 isn't done until
      there's a straightformward conversion path for all examples.

    o Add #warning to generated ssqls.h saying that you should now use
      SSQLS v2.  Wrap it in a check for MYSQLPP_ALLOW_SSQLS_V1,
      so people can disable the warning.


v3.4 Tentative Plan
-------------------
    o Add Query::storein<Container, T>(container), getting table
      name from container::value_type.table() instead.

    o Define operator<< for Fields, Row, StoreQueryResult, etc., giving
      CSV format.

    o Remove libexcommon.  Between above and SSQLSv2, we should have
      everything we need to get equivalent output without special
      purpose code.  There should be no ad hoc data dumping code in
      the examples.

    o Bring back mandatory quoting for manipulators?  If someone says
      os << mysqlpp::escape << foo; do they not really really mean
      escape foo?  Automatic quoting and escaping is different.  See

      http://lists.mysql.com/plusplus/7999

    o Configure script should try to get MySQL C API directories
      from mysql_config.

    o If pkg-config is available, register ourselves with it using
      information discovered by configure.  Also, write out a
      mysql++-config script, which either wraps pkg-config or
      reinvents it, poorly, for systems that don't have it.

    o Add String::operator==(const mysqlpp::null_type&).  Needed to
      allow comparison of row[x] returns to SQL null.  Change one of
      the examples to show it?

    o Memory "leak" and C API library init fixes:
    
      - Add DBDriver::library_begin() and library_end(), wrapping
        similarly named functions in the C API.

      - Create Process class, which you create at the top of main()
        on the stack, purely to call these automatically.

      - Update userman to recommend creating Process object in
        ConnectionPool derivatives, instead.

      - Create Thread class to call existing DBDriver::thread_start()
        and thread_end(), similar to Process, created on the stack
        of the thread entry function.

      - Move memory leak FAQ into userman, rewriting it to cover
        all this.

    o mysqlpp::execute manipulator.  Immediately executes built
      query string.  Works best with exceptions, as that's the only way
      to detect failures.

    o Chris Frey's packarray class

    o Create adaptors for std::bitset, for storing binary data in a
      MySQL table.  Make two options available, one for storing the
      return from bitset::to_ulong() in an UNSIGNED INTEGER column,
      and another for storing a larger set of bits in a more flexible
      way, perhaps as a BLOB.

    o Create a backtick manipulator for use by field_list() in row.h
      and ssqls.h.  These currently use do_nothing0, but that prevents
      use of SQL reserved words as identifiers.

    o Has experience with new thread awareness changed our mind on
      atomic inc/dec of reference counts in RefCounted*?

    o Create a fixed-point data type for use with SQL's DECIMAL and
      related types.  Right now, sql_decimal is a typedef for double,
      so you lose accuracy in the fractional part.  Don't forget to
      include an "is_null" flag to cope with conversion from infinite
      or NaN float values; that's how MySQL stores these.

    o Optional checked conversions in String for numerics: throw
      BadConversion on range overflow?

    o Add Query::storein_if(), mirroring store_if()

    o Add a method to mysqlpp::String to return a widened version of the
      string.  Probably make return type templatized so we can return
      wstring, C++/CLI native strings, etc.  Then convert examples that
      do this conversion to use this new mechanism.

    o Try to add operator std::string to String.  If it doesn't work,
      explain why not in the userman, and in Row::operator[] refman.
      
    o Wrap LOAD DATA INFILE:

      bool Query::load_file(
          const char* path,
          bool local_path,
          const char* table,
          const char* field_terminator = 0,   // default \t
          const char* field_encloser = 0,     // default none
          const char* field_escape = 0,       // default \
          const char* line_terminator = 0,    // \n on *ix, \r\n on Windows
          const char* line_starter = 0,       // default none
          const char* character_set = 0,      // default UTF-8
          const char* comment_prefix = 0,     // ignore no lines
          bool replace_existing = false,
          bool ignore_duplicates = false,
          bool low_priority = false,
          int skip_lines = 0);

    o Wrappers for above: load_local_tab_file(), load_local_csv_file(),
      load_remote_*()...

    o Query::save_file() interfaces, wrapping SELECT ... INTO FILE,
      modeled on above.


v4.0 or Later
-------------
    o Database independence:

      - Use libdbi or similar?  http://libdbi.sf.net/

      - Make DBDriver class purely abstract; move its entire functional
        contents to new MysqlDriver.

      - Must create at least two other DBDriver subclasses to
        ensure base class is reusable before releasing v4.0.
        PostgresDriver and SqlLiteDriver?

      - Templatize all classes that use DBDriver interface with the
        DB driver type.  This lets you specify the driver type to use
        with a Connection and all its children without modifying the
        existing method parameter lists.  This also lets us worry less
        about C API types, as they can be hidden away behind typedefs:

        class MysqlDriver : public DBDriver { ...
           typedef MYSQL_ROW row_type;
           ...
        }

        template <class DBD = MysqlDriver>
        class Connection ... { ...
           Query<DBD> query();
           ...
        }

        template <class DBD = MysqlDriver>
        class UseQueryResult { ...
            DBD::row_type fetch_raw_row();
        }

      - Tricky bits:
    
        - Initializing result set objects.

        - type_info module.  Extremely closely tied to MySQL C API
          right now.  Will probably have to turn it into a parallel
          class hierarchy to DBDriver, or fold it in with same.

        - Building MySQL++ on systems without autoconf.  How to
          specify what DB engines are available?  Probably default to
          supporting MySQL only, and let people turn things on manually
          as they need them.  Or, maybe make them use Bakefile so they
          can fiddle with the options if they want something atypical.

    o Fork mysqlpp::String into mysqlpp::Blob, which differs only in
      that it knows that it should be automatically quoted and escaped
      when inserted into a SQL query.  Could do this in 3.x, but it's a
      pretty serious API breakage.

    o Some sort of support for prepared statements.  Can we hijack
      the template query mechanism?

    o If SSQLSv2 does use a common base class, change Query template
      methods taking SSQLS into concrete methods taking SsqlsBase&.

    o Make Query::insert(), replace() and update() execute their
      queries immediately.  Requires an ABI break, because they'll
      have to return SimpleResult.

    o Switch Query's safe bool to overload basic_ios<>::operator
      void*() instead.  We create an ambiguous conversion in bool
      context with some C++ standard libraries otherwise.

    o Templatize mysqlpp::String on value_type so it can be used to
      hold wide characters.  Then the method that converts UTF-8 to the
      platform's best wide character type can just return a different
      variant of mysqlpp::String.

    o Add wrapper functions to Null<> like length() that call the
      corresponding function on data member, if present, to make it
      more transparent.  At minimum, mirror the std::string API.

    o Transaction class should check an "in transaction" flag on
      Connection (or DBDriver) before sending BEGIN, defaulting to
      false.  If set, the Transaction object does nothing.  If not
      set, set it and send the query.  This prevents it from trying
      to set up nested queries, which MySQL doesn't support.

    o Remove throw-spec for std::out_of_range from SQLTypeAdapter::at().
      It no longer throws this, and throw-specs are passee' anyway.

    o Store failed query string in BadQuery exception object, to make
      logging and debugging easier.  One could have a try block wrapping
      many queries, and be able to recover the failed query string from
      the exception object, instead of somehow keeping track yourself.

      Patch: http://lists.mysql.com/plusplus/8374
 
    o Query and SQLStream could have a common base class that would
      allow the stream manipulator functions to catch and modify
      strings based on only one dynamic_cast instead of requiring
      two as it does since the addition of the SQLStream class.

    o Make internal call chain steps like Query::execute(SQLQueryParms&)
      protected?  No good reason for end users to call it, and making
      it part of the public API causes people to try calling it, and
      discovering that it's not a very elegant interface, compared to
      the ones taking SQLStrings.

    o SQL time type allows +/- 839 hours of range.  v3.0 code doesn't
      cope with negative times, and if we change it to use signed
      integers, we'll still only get +/-127 hours instead of +255.
      Need to switch the hour field to a short to get the full range.

    o Create a thread-safe message queue for separating DB access and
      data use into multiple threads.  Something like ConnectionPool,
      optional and with no ties to the internals of MySQL++.  There
      could be an adapter between one end of the queue and a Connection
      object, which creates Queries to handle standardized messages,
      delivering the results back to the queue.

    o Get rid of two-step create in DBDriver, requiring a connection to
      be established in ctor for object to be valid?  RAII.  The
      DB-specific functions that don't require a connection can be
      static methods.  Tricky bit: a failed Connection::connect() call
      will likely be followed by an indirect call to DBDriver::err*().
      Does Connection cache the error value and message?  If we can pull
      this off, we can drop the DBDriver::is_connected_ flag and change
      Connection::connected() to "return driver_ != 0".

    o Add STL-like custom Allocator template parameters to memory-hungry
      classes like Row?  Useful in apps that process lots of data over
      long periods, causing heap fragmentation with the default C++
      allocator.

    o Audit library for non-virtual methods that could be profitably
      made virtual.  Many are non-overridable now purely because
      changing them would break the ABI, so now's the time to fix that.
<
<
<
<
<
<
<
<
<
<
<
<
<
<
<
<
<
<
<
<
<
<
<
<
<
<
<
<
<
<
<
<
<
<
<
<
<
<
<
<
<
<
<
<
<
<
<
<
<
<
<
<
<
<
<
<
<
<
<
<
<
<
<
<
<
<
<
<
<
<
<
<
<
<
<
<
<
<
<
<
<
<
<
<
<
<
<
<
<
<
<
<
<
<
<
<
<
<
<
<
<
<
<
<
<
<
<
<
<
<
<
<
<
<
<
<
<
<
<
<
<
<
<
<
<
<
<
<
<
<
<
<
<
<
<
<
<
<
<
<
<
<
<
<
<
<
<
<
<
<
<
<
<
<
<
<
<
<
<
<
<
<
<
<
<
<
<
<
<
<
<
<
<
<
<
<
<
<
<
<
<
<
<
<
<
<
<
<
<
<
<
<
<
<
<
<
<
<
<
<
<
<
<
<
<
<
<
<
<
<
<
<
<
<
<
<
<
<
<
<
<
<
<
<
<
<
<
<
<
<
<
<
<
<
<
<
<
<
<
<
<
<
<
<
<
<
<
<
<
<
<
<
<
<
<
<
<
<
<
<
<
<
<
<
<
<
<
<
<
<
<
<
<
<
<
<
<
<
<
<
<
<
<
<
<
<
<
<
<
<
<
<
<
<
<
<
<
<
<
<
<
<
<
<
<
<
<
<
<
<
<
<
<
<
<
<
<
<
<
<
<
<
<
<
<
<
<
<
<
<
<
<
<
<
<
<
<
<
<
<
<
<
<
<
<
<
<
<
<
<
<
<
<
<
<
<
<
<
<
<
<
<
<
<
<
<
<
<
<
<
<
<
<
<
<
<
<
<
<
<
<
<
<
<
<
<
<
<
<
<
<
<
<
<
<
<
<
<
<
<
<
<
<
<
<
<
<
<
<
<
<
<
<
<
<
<
<
<
<
<
<
<
<
<
<
<
<
<
<
<
<
<
<
<
<
<
<
<
<
<
<
<
<
<
<
<
<
<
<
<
<
<
<
<
<
<
<
<
<
<
<
<
<
<
<
<
<
<
<
<
<
<
<
<
<
<
<
<
<
<
<
<
<
<
<
<
<
<
<
<
<
<
<
<
<
<
<
<
<
<
<
<
<
<
<
<
<
<
<
<
<
<
<
<
<
<
<
<
<
<
<
<
<
<
<