参考网址：he SQLite Bytecode Engine
前提准备
编译sqlite3：gcc -g -O0 -DSQLITE_DEBUG -DSQLITE_ENABLE_EXPLAIN_COMMENTS -DSQLITE_ENABLE_SELECTTRACE -DSQLITE_ENABLE_WHERETRACE -o sqlite3 shell.c sqlite3.c
这里单开DEBUG十分之有用，源码里面，还有一些输出调试控制。SQLITE_ENABLE_EXPLAIN_COMMENTS这个非常有用，可以explain某个sql语句的时候，查看具体要执行什么操作。
-
PRAGMA parser_trace=boolean：控制是否显示parser规约过程 -
PRAGMA vdbe_debug=boolean：显示sql对应的操作码，显示执行过程。
以下是官网的介绍：
- PRAGMA parser_trace
- PRAGMA vdbe_addoptrace
- PRAGMA vdbe_debug
- PRAGMA vdbe_listing
- PRAGMA vdbe_trace
具体例子
select 主键并且 without rowid
sqlite> select * from department where id = 33;
SQL: [select * from department where id = 33;]
VDBE Program Listing:
0 Init 0 11 0 00 Start at 11
1 OpenRead 1 2 0 k(1,) 02 root=2 iDb=0; sqlite_autoindex_department_1
2 Explain 2 0 0 SEARCH TABLE department USING PRIMARY KEY (id=?) 00
3 Integer 33 1 0 00 r[1]=33
4 SeekGE 1 10 1 1 00 key=r[1]
5 IdxGT 1 10 1 1 00 key=r[1]
6 Column 1 0 2 00 r[2]=department.id
7 Column 1 1 3 00 r[3]=department.dept
8 Column 1 2 4 00 r[4]=department.emp_id
9 ResultRow 2 3 0 00 output=r[2..4]
10 Halt 0 0 0 00
11 Transaction 0 0 2 0 01 usesStmtJournal=0
12 TableLock 0 2 0 department 00 iDb=0 root=2 write=0
13 Goto 0 1 0 00
VDBE Trace:
0 Init 0 11 0 00 Start at 11
SQL-trace: select * from department where id = 33;
11 Transaction 0 0 2 0 01 usesStmtJournal=0
12 TableLock 0 2 0 department 00 iDb=0 root=2 write=0
13 Goto 0 1 0 00
1 OpenRead 1 2 0 k(1,) 02 root=2 iDb=0; sqlite_autoindex_department_1
2 Explain 2 0 0 SEARCH TABLE department USING PRIMARY KEY (id=?) 00
3 Integer 33 1 0 00 r[1]=33
REG[1] = i:33
4 SeekGE 1 10 1 1 00 key=r[1]
REG[1] = i:33
6 Column 1 0 2 00 r[2]=department.id
REG[2] = i:33
7 Column 1 1 3 00 r[3]=department.dept
REG[3] = s6[test33](8)
8 Column 1 2 4 00 r[4]=department.emp_id
REG[4] = i:3
9 ResultRow 2 3 0 00 output=r[2..4]
REG[2] = i:33
REG[3] = s6[test33](8)
REG[4] = i:3
33|test33|3
10 Halt 0 0 0 00
熟悉汇编的，或者别的虚拟机字节码的对此应该不陌生。我们可以直接看最终的执行顺序，也就是VDBE Trace开始的字节码：
- 11 Transaction 这个格式有点多寄存器，Transaction p1 p2 p3 p4 p5
- p1 对应这打开的数据库序号，0 是主数据库文件， 1 是临时表用的文件，2及以上都是attach的数据库文件。
- p2 对应是读事务还是写事务，非0则是写事务，0为读事务。如果p2 >= 2 的话，则是排他性的事务
- p3 schema cookie，用来校验schema是否过期了
- p4 schema generation counter，尚不明确
- p5 控制是否检验schema cookie和schema generation counter。如果不一致就会返回SQLITE_SCHEMA ，此时sqlite3_step会reprepare statement
- 12 TableLock p1 p2 p3 p4:
- p1 表对应的数据库文件下标
- p2 该表的根页号
- p3 0 for 读锁，1 for 写锁
- 1 OpenRead p1 p2 p3 p4 p5：打开一个只读的游标
- p1 游标的句柄
- p2 表所在的根页
- p3 表所在数据库文件
- p4 依据是索引树还是普通b-tree。待查明
- p5 设置该游标的属性：0x02 代表它只用来等值搜索
- Integer p1 p2：32位整型的p1写入寄存器p2中，本例中是将r[1] = 33。
- SeekGE p1 p2 p3 p4：寻找关键值大于等于key的。游标p1，p2是失败的时候goto的地方。
- IdxGT p1 p2 p3 p4：判断找到的关键字是否大于key，是的话跳转到p2。
- Column p1 p2 p3：将游标p1看做一行数据，p2是第p2项数据，将其结果储存在寄存器p3中。
- ResultRow p1 p2 ：将从寄存器p1开始的p2个寄存器作为返回值，通知sqlite3_step() 返回SQLITEROW。
select 索引
sqlite> select * from department where dept == 'test33';
SQL: [select * from department where dept == 'test33';]
VDBE Program Listing:
0 Init 0 15 0 00 Start at 15
1 OpenRead 0 2 0 k(1,) 00 root=2 iDb=0; department
2 OpenRead 1 3 0 k(2,,) 02 root=3 iDb=0; depName_index
3 Explain 3 0 0 SEARCH TABLE department USING INDEX depName_index (dept=?) 00
4 String8 0 1 0 test33 00 r[1]='test33'
5 SeekGE 1 14 1 1 00 key=r[1]
6 IdxGT 1 14 1 1 00 key=r[1]
7 Column 1 1 2 00 r[2]=
8 NotFound 0 13 2 1 00 key=r[2]
9 Column 1 1 3 00 r[3]=department.id
10 Column 1 0 4 00 r[4]=department.dept
11 Column 0 2 5 00 r[5]=department.emp_id
12 ResultRow 3 3 0 00 output=r[3..5]
13 Next 1 6 1 00
14 Halt 0 0 0 00
15 Transaction 0 0 2 0 01 usesStmtJournal=0
16 TableLock 0 2 0 department 00 iDb=0 root=2 write=0
17 Goto 0 1 0 00
VDBE Trace:
0 Init 0 15 0 00 Start at 15
SQL-trace: select * from department where dept == 'test33';
15 Transaction 0 0 2 0 01 usesStmtJournal=0
16 TableLock 0 2 0 department 00 iDb=0 root=2 write=0
17 Goto 0 1 0 00
1 OpenRead 0 2 0 k(1,) 00 root=2 iDb=0; department
2 OpenRead 1 3 0 k(2,,) 02 root=3 iDb=0; depName_index
3 Explain 3 0 0 SEARCH TABLE department USING INDEX depName_index (dept=?) 00
4 String8 0 1 0 test33 00 r[1]='test33'
REG[1] = t6[test33](8)
5 SeekGE 1 14 1 1 00 key=r[1]
REG[1] = t6[test33](8)
7 Column 1 1 2 00 r[2]=
REG[2] = i:33
8 NotFound 0 13 2 1 00 key=r[2]
REG[2] = i:33
9 Column 1 1 3 00 r[3]=department.id
REG[3] = i:33
10 Column 1 0 4 00 r[4]=department.dept
REG[4] = s6[test33](8)
11 Column 0 2 5 00 r[5]=department.emp_id
REG[5] = i:3
12 ResultRow 3 3 0 00 output=r[3..5]
REG[3] = i:33
REG[4] = s6[test33](8)
REG[5] = i:3
33|test33|3
13 Next 1 6 1 00
6 IdxGT 1 14 1 1 00 key=r[1]
REG[1] = t6[test33](8)
14 Halt 0 0 0 00
sqlite>
这个过程就比较复杂，首先明确一定，dept这个索引，存储着的是dept与id的对应关系，sql遇到带索引的查询是先通过索引找与主键（一般情况下是rowid，但without rowid则是primary key）的映射关系。然后再通过主键搜索到对应的记录。
所以开始分析下这条sql语句
- 锁住数据库文件
- 打开两个游标cursor0，cursor1，分别指向主键id和dept索引的根页。
- 通过cursor1找到dept == 'test33'的映射关系。并且将主键id赋值到r[2]上
- 通过cursor0找到主键==r[2]的记录
- 这里比较有意思，主键id和dept在cursor1上有，故这两个直接由cursor1给出。emp_id只有cursor0有，故这个要从cursor0上获取。
非主键/索引 select
sqlite> select * from department where emp_id == 3;
SQL: [select * from department where emp_id == 3;]
VDBE Program Listing:
0 Init 0 12 0 00 Start at 12
1 OpenRead 1 2 0 k(1,) 00 root=2 iDb=0; sqlite_autoindex_department_1
2 Explain 2 0 0 SCAN TABLE department 00
3 Rewind 1 11 1 0 00
4 Column 1 2 1 00 r[1]=department.emp_id
5 Ne 2 10 1 (BINARY) 54 if r[1]!=r[2] goto 10
6 Column 1 0 3 00 r[3]=department.id
7 Column 1 1 4 00 r[4]=department.dept
8 Column 1 2 5 00 r[5]=department.emp_id
9 ResultRow 3 3 0 00 output=r[3..5]
10 Next 1 4 0 01
11 Halt 0 0 0 00
12 Transaction 0 0 2 0 01 usesStmtJournal=0
13 TableLock 0 2 0 department 00 iDb=0 root=2 write=0
14 Integer 3 2 0 00 r[2]=3
15 Goto 0 1 0 00
VDBE Trace:
0 Init 0 12 0 00 Start at 12
SQL-trace: select * from department where emp_id == 3;
12 Transaction 0 0 2 0 01 usesStmtJournal=0
13 TableLock 0 2 0 department 00 iDb=0 root=2 write=0
14 Integer 3 2 0 00 r[2]=3
REG[2] = i:3
15 Goto 0 1 0 00
1 OpenRead 1 2 0 k(1,) 00 root=2 iDb=0; sqlite_autoindex_department_1
2 Explain 2 0 0 SCAN TABLE department 00
3 Rewind 1 11 1 0 00
4 Column 1 2 1 00 r[1]=department.emp_id
REG[1] = i:33333
5 Ne 2 10 1 (BINARY) 54 if r[1]!=r[2] goto 10
REG[2] = i:3
REG[1] = i:33333
10 Next 1 4 0 01
4 Column 1 2 1 00 r[1]=department.emp_id
REG[1] = i:3
5 Ne 2 10 1 (BINARY) 54 if r[1]!=r[2] goto 10
REG[2] = i:3
REG[1] = i:3
6 Column 1 0 3 00 r[3]=department.id
REG[3] = i:33
7 Column 1 1 4 00 r[4]=department.dept
REG[4] = s6[test33](8)
8 Column 1 2 5 00 r[5]=department.emp_id
REG[5] = i:3
9 ResultRow 3 3 0 00 output=r[3..5]
REG[3] = i:33
REG[4] = s6[test33](8)
REG[5] = i:3
33|test33|3
10 Next 1 4 0 01
4 Column 1 2 1 00 r[1]=department.emp_id
REG[1] = i:4
5 Ne 2 10 1 (BINARY) 54 if r[1]!=r[2] goto 10
REG[2] = i:3
REG[1] = i:4
10 Next 1 4 0 01
4 Column 1 2 1 00 r[1]=department.emp_id
REG[1] = i:5
5 Ne 2 10 1 (BINARY) 54 if r[1]!=r[2] goto 10
REG[2] = i:3
REG[1] = i:5
10 Next 1 4 0 01
4 Column 1 2 1 00 r[1]=department.emp_id
REG[1] = i:6
5 Ne 2 10 1 (BINARY) 54 if r[1]!=r[2] goto 10
REG[2] = i:3
REG[1] = i:6
10 Next 1 4 0 01
4 Column 1 2 1 00 r[1]=department.emp_id
REG[1] = i:7
5 Ne 2 10 1 (BINARY) 54 if r[1]!=r[2] goto 10
REG[2] = i:3
REG[1] = i:7
10 Next 1 4 0 01
4 Column 1 2 1 00 r[1]=department.emp_id
REG[1] = i:4777
5 Ne 2 10 1 (BINARY) 54 if r[1]!=r[2] goto 10
REG[2] = i:3
REG[1] = i:4777
10 Next 1 4 0 01
4 Column 1 2 1 00 r[1]=department.emp_id
REG[1] = i:456520
5 Ne 2 10 1 (BINARY) 54 if r[1]!=r[2] goto 10
REG[2] = i:3
REG[1] = i:456520
10 Next 1 4 0 01
11 Halt 0 0 0 00
这里看起来就复杂很多了，因为没有索引，只能逐个遍历确认emp_id == 3;
- Rewind，重定位到该表第一项中。
- 将当前数据项的emp_id与r[2]作比较，同样则返回结果，否则获取下一项循环直至所有数据项被遍历完了。
总结一下：
- 如果是索引或者主键。rowid查找的时候，是直接由SeekXX系列的可以直接用的，并且支持多个字段的那种索引。例如SeekGE P1 P2 P3 P4，如果是index的话，就从p3开始的p4个寄存器都是索引的内容，依次序比较即可。
- 如果是非索引的查询，那么就是从第一项数据依次遍历一直往下找，直至结束。
insert 语句
最后让我们再来分析一个insert语句，其余的就不演示了。可以自行编译执行调试获知。
sqlite> insert into department values(100, 'test100', 90909);
SQL: [insert into department values(100, 'test100', 90909);]
VDBE Program Listing:
0 Init 0 24 0 00 Start at 24
1 OpenWrite 1 3 0 k(2,,) 00 root=3 iDb=0; depName_index
2 OpenWrite 2 2 0 k(1,) 00 root=2 iDb=0; sqlite_autoindex_department_1
3 Null 0 1 0 00 r[1]=NULL
4 Integer 100 2 0 00 r[2]=100
5 String8 0 3 0 test100 00 r[3]='test100'
6 Integer 90909 4 0 00 r[4]=90909
7 HaltIfNull 1299 2 2 department.id 01 if r[2]=null halt
8 Affinity 2 3 0 DBD 00 affinity(r[2..4])
9 Noop 0 0 0 00 uniqueness check for depName_index
10 SCopy 3 6 0 00 r[6]=r[3]; dept
11 SCopy 2 7 0 00 r[7]=r[2]; id
12 MakeRecord 6 2 5 00 r[5]=mkrec(r[6..7]); for depName_index
13 Noop 0 0 0 00 uniqueness check for sqlite_autoindex_department_1
14 SCopy 2 9 0 00 r[9]=r[2]; id
15 SCopy 3 10 0 00 r[10]=r[3]; dept
16 SCopy 4 11 0 00 r[11]=r[4]; emp_id
17 MakeRecord 9 3 8 00 r[8]=mkrec(r[9..11]); for sqlite_autoindex_department_1
18 Abortable 0 0 0 00
19 NoConflict 2 21 9 1 00 key=r[9]
20 Halt 1555 2 0 department.id 02
21 IdxInsert 1 5 6 2 10 key=r[5]
22 IdxInsert 2 8 9 1 11 key=r[8]
23 Halt 0 0 0 00
24 Transaction 0 1 2 0 01 usesStmtJournal=0
25 TableLock 0 2 1 department 00 iDb=0 root=2 write=1
26 Goto 0 1 0 00
简单分析一下啊，因为这里有两个索引，所以要分别打开两个游标。而且，因为创建表的时候用的是without rowid，所以r[1]=null。
- 先做约束性的校验，比如id != NULL。
- 再对两个索引打包，make record。depName_index 由 dept + id 组成，存放在r[5]中，id的由id + dept + emp_id组成，存放在r[8]中。
- 检测id是否重复，没重复就跳到插入那里去，否则退出，并且p1给出退出码。但是dept并没有检测，是因为index不一定要唯一性，只有创建的时候带上 unique 才会需要这玩意检测。
update 语句
sqlite> update department set emp_id=90910 where emp_id == 90909;
SQL: [update department set emp_id=90910 where emp_id == 90909;]
VDBE Program Listing:
0 Init 0 23 0 00 Start at 23
1 Null 0 10 10 00 r[10..10]=NULL
2 Noop 3 1 0 00
3 OpenWrite 2 2 0 k(1,) 00 root=2 iDb=0; sqlite_autoindex_department_1
4 Explain 4 0 0 SCAN TABLE department 00
5 Rewind 2 22 12 0 00
6 Column 2 2 12 00 r[12]=department.emp_id
7 Ne 13 21 12 (BINARY) 54 if r[12]!=r[13] goto 21
8 Column 2 0 10 00 r[10]=department.id
9 Once 0 10 0 00
10 IsNull 10 22 0 00 if r[10]==NULL goto 22
11 Column 2 0 7 00 r[7]=department.id
12 Column 2 1 8 00 r[8]=department.dept
13 Integer 90910 9 0 00 r[9]=90910
14 Affinity 7 3 0 DBD 00 affinity(r[7..9])
15 Noop 0 0 0 00 uniqueness check for sqlite_autoindex_department_1
16 SCopy 7 2 0 00 r[2]=r[7]; id
17 SCopy 8 3 0 00 r[3]=r[8]; dept
18 SCopy 9 4 0 00 r[4]=r[9]; emp_id
19 MakeRecord 2 3 1 00 r[1]=mkrec(r[2..4]); for sqlite_autoindex_department_1
20 IdxInsert 2 1 2 1 03 key=r[1]
21 Next 2 6 0 01
22 Halt 0 0 0 00
23 Transaction 0 1 2 0 01 usesStmtJournal=0
24 TableLock 0 2 1 department 00 iDb=0 root=2 write=1
25 Integer 90909 13 0 00 r[13]=90909
26 Goto 0 1 0 00
update思路感觉和select -> modify -> insert的思路差不多，先查询得到对应的数据项，然后修改对应列项，最后写回b树中。这里留下一个疑问，在key相同的情况下， 这个IdxInsert会做什么处理？感觉会是覆盖。
delete语句
sqlite> delete from department where emp_id == 90910;
SQL: [delete from department where emp_id == 90910;]
VDBE Program Listing:
0 Init 0 16 0 00 Start at 16
1 Noop 3 1 0 00
2 OpenWrite 2 2 0 k(1,) 00 root=2 iDb=0; sqlite_autoindex_department_1
3 Explain 3 0 0 SCAN TABLE department 00
4 Rewind 2 15 2 0 00
5 Column 2 2 2 00 r[2]=department.emp_id
6 Ne 3 14 2 (BINARY) 54 if r[2]!=r[3] goto 14
7 Column 2 0 1 00 r[1]=department.id
8 Once 0 10 0 00
9 OpenWrite 1 3 0 k(2,,) 08 root=3 iDb=0; depName_index
10 Column 2 1 4 00 r[4]=department.dept
11 Column 2 0 5 00 r[5]=department.id
12 IdxDelete 1 4 2 00 key=r[4..5]
13 Delete 2 1 0 department 02
14 Next 2 5 0 01
15 Halt 0 0 0 00
16 Transaction 0 1 2 0 01 usesStmtJournal=0
17 TableLock 0 2 1 department 00 iDb=0 root=2 write=1
18 Integer 90910 3 0 00 r[3]=90910
19 Goto 0 1 0 00
也是想查询到对应的数据项，然后删除索引，最后删表项。