[20181229]关于字符串的分配问题.txt

[20181229]关于字符串的分配问题.txt

--//链接：http://www.itpub.net/thread-2107534-1-1.html提到的问题，里面一段英文读起来很绕口：
--//百度找到如下内容：https://docs.oracle.com/cd/b19306_01/appdev.102/b14261/datatypes.htm
varchar2 datatype

you use the varchar2 datatype to store variable-length character data. how the data is represented internally depends on
the database character set. the varchar2 datatype takes a required parameter that specifies a maximum size up to 32767
bytes. the syntax follows:

varchar2(maximum_size [char  |  byte])

you cannot use a symbolic constant or variable to specify the maximum size; you must use an integer literal in the range
1 .. 32767.

small varchar2 variables are optimized for performance, and larger ones are optimized for efficient memory use. the
cutoff point is 2000 bytes. for a varchar2 that is 2000 bytes or longer, pl/sql dynamically allocates only enough memory
to hold the actual value. for a varchar2 variable that is shorter than 2000 bytes, pl/sql preallocates the full declared
length of the variable. for example, if you assign the same 500-byte value to a varchar2(2000 byte) variable and to a
varchar2(1999 byte) variable, the former takes up 500 bytes and the latter takes up 1999 bytes.
--//里面的内容来之这里.简单点就是小的varchar2变量优化为性能,大的varchar2变量优化内存使用.分界点在2000字节.
--//贴一段金山词霸的翻译:
小的varchar 2变量是为了性能而优化的，较大的变量是为了高效的内存使用而优化的。截止点是2000字节。对于2000字节或更长的
varchar 2，pl/sql动态分配的内存仅足以容纳实际值。对于小于2000字节的varchar 2变量，pl/sql将分配变量的完整声明长度。例如，
如果将相同的500字节值分配给varchar 2(2000字节)变量和varchar 2(1999字节)变量，则前者占500个字节，后者占1999年字节。

if you specify the maximum size in bytes rather than characters, a varchar2(n) variable might be too small to hold n
multibyte characters. to avoid this possibility, use the notation varchar2(n char) so that the variable can hold n
characters in the database character set, even if some of those characters contain multiple bytes. when you specify the
length in characters, the upper limit is still 32767 bytes. so for double-byte and multibyte character sets, you can
only specify 1/2 or 1/3 as many characters as with a single-byte character set.

although pl/sql character variables can be relatively long, you cannot insert varchar2 values longer than 4000 bytes
into a varchar2 database column.

you can insert any varchar2(n) value into a long database column because the maximum width of a long column is
2147483648 bytes or two gigabytes. however, you cannot retrieve a value longer than 32767 bytes from a long column into
a varchar2(n) variable. note that the long datatype is supported only for backward compatibility; see "long and long raw
datatypes" more information.

when you do not use the char or byte qualifiers, the default is determined by the setting of the nls_length_semantics
initialization parameter. when a pl/sql procedure is compiled, the setting of this parameter is recorded, so that the
same setting is used when the procedure is recompiled after being invalidated.

--//注意该文档是10gr2下的.
--//突然想起我以前的测试,可以验证11g改动分界点,实际上是1001个字符.当时测试的链接如下:
--//http://blog.itpub.net/267265/viewspace-746524/  => [20160224]绑定变量的分配长度.txt
--//http://blog.itpub.net/267265/viewspace-1993495/ => [20121016]字符串长度与绑定变量的子光标.txt
--//里面提到1个情况，我当时没搞清楚，看完上面的链接一下明白过来,我通过重复测试来说明问题。

1.环境：
scott@test01p> @ ver1
port_string          version    banner                                                                       con_id
-------------------- ---------- ---------------------------------------------------------------------------- ------
ibmpc/win_nt64-9.1.0 12.2.0.1.0 oracle database 12c enterprise edition release 12.2.0.1.0 - 64bit production      0

scott@test01p> create table t (a varchar2(4000));
table created.
--//分析略.
--//建立脚本len.txt
declare
instring varchar2(&&1);
  begin
   for i in 1..1000 loop
     instring := rpad('x',i,'x');
     execute immediate 'select /*+ find_me &&1 */ count(*) from t where a=:instring' using instring ;
   end loop;
end;
/

2.测试一：
--//执行 @ len.txt 1000
--//执行完成后确定sql_id=4mv1hkjru31tp
scott@test01p> select sql_text,sql_id,child_number,executions,parse_calls,loads,invalidations from v$sql where sql_id='4mv1hkjru31tp';
sql_text                                                     sql_id        child_number executions parse_calls loads invalidations
------------------------------------------------------------ ------------- ------------ ---------- ----------- ----- -------------
select /*+ find_me 1000 */ count(*) from t where a=:instring 4mv1hkjru31tp            0       1000           1     1             0

scott@test01p> @ bind_cap 4mv1hkjru31tp ''
c200
------------------------------------------------------------
select /*+ find_me 1000 */ count(*) from t where a=:instring

sql_id        child_number was name                   position max_length last_captured       datatype_string value_string
------------- ------------ --- -------------------- ---------- ---------- ------------------- --------------- -------------
4mv1hkjru31tp            0 yes :instring                     1       2000 2018-12-30 20:18:38 varchar2(2000)  x

--//你可以发现查询v$sql_bind_capture视图,里面datatype_string记录的是varchar2(2000),而我定义的大小是varchar2(1000).
--//当然oracle按照定义不会分配2000空间,而是最大1000.
--//许多人都知道,如果字符串绑定变量长度变化会产生子光标.
--//通过测试可以知道字符串的长度变化是32，32+96=128,32+96+1872=2000.也就是分4个段 1-32,33-128,129-2000,2001-4000.
--//参考链接:http://blog.itpub.net/267265/viewspace-746524/
--//如果开始分配的字符串空间是按照实际使用大小来分配的,就会出现至少3个子光标的情况.而现在仅仅出现1个,说明oracle在开始执行就
--//分配1000个字符空间.

3.测试二:
scott@test01p> alter system flush shared_pool ;
system altered.

--//修改参数1001,执行 @ len.txt 1001

scott@test01p> @ len.txt 1001
pl/sql procedure successfully completed.
--//确定sql_id=as5nq40yutw9t

scott@test01p> select sql_text,sql_id,child_number,executions,parse_calls,loads,invalidations from v$sql where sql_id='as5nq40yutw9t';

sql_text                                                     sql_id        child_number executions parse_calls loads invalidations
------------------------------------------------------------ ------------- ------------ ---------- ----------- ----- -------------
select /*+ find_me 1001 */ count(*) from t where a=:instring as5nq40yutw9t            0         32           1     1             0
select /*+ find_me 1001 */ count(*) from t where a=:instring as5nq40yutw9t            1         96           0     1             0
select /*+ find_me 1001 */ count(*) from t where a=:instring as5nq40yutw9t            2        872           0     1             0

scott@test01p> @ bind_cap as5nq40yutw9t ''
c200
------------------------------------------------------------
select /*+ find_me 1001 */ count(*) from t where a=:instring

sql_id        child_number was name      position max_length last_captured       datatype_string value_string
------------- ------------ --- --------- -------- ---------- ------------------- --------------- --------------------------------------------------
as5nq40yutw9t            0 yes :instring        1         32 2018-12-30 20:30:47 varchar2(32)    x
                         1 yes :instring        1        128 2018-12-30 20:30:47 varchar2(128)   xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx
                         2 yes :instring        1       2000 2018-12-30 20:30:47 varchar2(2000)  xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx
                                                                                                 xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx
                                                                                                 xxxxxxxxxxxxxxxxxxxxxxxxxxxxx
--//我仅仅修改字符串的定义varchar2(1001),就导致相似的语句产生3个子光标。
--//说明一个问题当字符串长度大于1000时,oracle字符串的分配按需来分配,这样就会出现3个子光标的情况.
--//从执行次数上可以看出长度变化1-32, 33- 128(32+96),129-2000(2000可以从v$sql_bind_capture视图的datatype_string确定).

3.继续测试:
--//面前的测试在pl/sql进行的,在sqlplus测试看看.
scott@test01p> alter system flush shared_pool;
system altered.

variable instring varchar2(1000)
exec :instring := rpad('x',1);
select /*+ find_me */ count(*) from t where a=:instring;
--//确定sql_id=383pcxarzpwbg.

scott@test01p> @ bind_cap 383pcxarzpwbg ''
c200
-------------------------------------------------------
select /*+ find_me */ count(*) from t where a=:instring

sql_id        child_number was name                   position max_length last_captured       datatype_string value_string
------------- ------------ --- -------------------- ---------- ---------- ------------------- --------------- ------------------
383pcxarzpwbg            0 yes :instring                     1       2000 2018-12-30 20:42:17 varchar2(2000)  x


scott@test01p> alter system flush shared_pool;
system altered.

variable instring varchar2(1001)
exec :instring := rpad('x',1);
select /*+ find_me 1001x */ count(*) from t where a=:instring;
--//确定sql_id=fd4dr46guv82z
scott@test01p> @ bind_cap fd4dr46guv82z ''
c200
-------------------------------------------------------------
select /*+ find_me 1001x */ count(*) from t where a=:instring

sql_id        child_number was name                   position max_length last_captured       datatype_string value_string
------------- ------------ --- -------------------- ---------- ---------- ------------------- --------------- --------------------
fd4dr46guv82z            0 yes :instring                     1       2000 2018-12-30 20:44:41 varchar2(2000)  x

--//可以看出sqlplus就不是这样,按照定义分配.而仅仅pl/sql比较特殊.存在1个1001分界点.

4.上面文档是来自10g的官方文档.要找一个10g的版本重复测试看看,验证是否是2000.
--//等上班找个10g的环境来测试看看.

5.附上bind_cap.sql的脚本.
$ cat bind_cap.sql
set verify off
column value_string format a50
column datatype_string format a15
break on sql_id on child_number  skip 1
select  replace(sql_fulltext,chr(13),'') c200 from v$sql where sql_id='&1' and rownum<=1;

select sql_id,
       child_number,
       was_captured,
       name,
       position,
       max_length,
       last_captured,
       datatype_string,
       decode (
          datatype_string,
          'date', to_char (to_date (value_string, 'mm/dd/yy hh24:mi:ss'),
                           'yyyy/mm/dd hh24:mi:ss'),
          value_string)
          value_string
  from v$sql_bind_capture
 where sql_id = '&1' and was_captured='yes' and  dup_position is null and name=nvl('&&2',name)
 order by child_number,was_captured,position;
break on sql_id on child_number  skip 0