相关
《Postgresql源码（77）plpgsql中参数传递和赋值（pl参数）》
《Postgresql源码（78）plpgsql中调用call proc()时的参数传递和赋值（pl参数）》
《Postgresql源码（79）plpgsql中多层调用时参数传递关键点分析（pl参数）》
《Postgresql源码（84）语义分析——函数调用结构CallStmt的构造与函数多态的实现（pl参数）》
场景：函数从tp88调入tp99，有两个入参和一个出参，出参由内层函数赋值。
下面提到“外层”是指调用tp88，“内层”是指调用tp99。
本篇回答下面问题
问题一：外层ExecuteCallStmt用什么构造参数列表fcinfo->args？
问题二：外层ExecuteCallStmt如何构造fcinfo->args？
问题三：内层ExecuteCallStmt用什么构造fcinfo->args？
问题四：内层ExecuteCallStmt如何构造fcinfo->args？
问题五：exec_stmt_block刚进入初始化哪些变量？
问题六：exec_stmt_block刚进入为什么要初始化哪些变量直接用不行吗？
问题七：内层函数执行完的变量是在哪赋给外层的？
CREATE or replace PROCEDURE tp99(a9 in int,b9 in int,c9 out int)
LANGUAGE plpgsql
AS $$
BEGIN
c9 := 88299;
raise notice '=====tp99=====';
raise notice 'a9: %', a9;
raise notice 'b9: %', b9;
raise notice 'c9: %', c9;
END;
$$;
CREATE or replace PROCEDURE tp88(a8 in integer, b8 out integer)
LANGUAGE plpgsql
AS $$
DECLARE
i8 int;
c8 int;
BEGIN
i8 := 100;
call tp99(i8,20,c8);
raise notice '=====tp88=====';
raise notice 'get from tp99 c8: %', c8;
END;
$$;
call tp88(1,2);
函数是从tp88调入tp99，有两个入参和一个出参，出参由内层函数赋值。
下面提到“外层”是指调用tp88，“内层”是指调用tp99。
standard_ProcessUtility <---- （外层）调用tp88
ExecuteCallStmt <---- （外层）按x构造参数列表fcinfo->args，x从哪来？
plpgsql_call_handler （外层）ExecuteCallStmt如何fcinfo->args？
plpgsql_exec_function <---- 按fcinfo->args初始化参数
exec_toplevel_block
exec_stmt_block <---- 按block->initvarnos初始化参数（哪来的？为什么初始化？）
exec_stmts
exec_stmt_call <---- 怎么构造新参数列表？
SPI_execute_plan_extended <---- SPI怎么使用？
_SPI_execute_plan
ProcessUtility
standard_ProcessUtility <---- （内层）调用tp99
ExecuteCallStmt <---- （内层）按x构造参数列表，x从哪来？
plpgsql_call_handler
plpgsql_exec_function
exec_toplevel_block
exec_stmt_block
exec_stmts
exec_stmt_raise
plantree_list -> [List]
{type = T_List, length = 1, max_length = 5}
\
ListCell -> [PlannedStmt]
{type = T_PlannedStmt, commandType = CMD_UTILITY,..., utilityStmt = 0x16600d8}
\
utilityStmt -> [CallStmt]
{type = T_CallStmt, funccall = 0x1660080, funcexpr = 0x1660958, outargs = 0x1660a78}
\
funccall -> [FuncCall] // from the parser
{type = T_FuncCall, funcname = 0x165ff60, args = 0x165fff0,.,funcformat = COERCE_EXPLICIT_CALL}
|
funcexpr -> [FuncExpr] // transformed call, with only input args
{xpr = {type = T_FuncExpr}, funcid = 24972, funcresulttype = 2249, funcretset = false,
funcvariadic = false, funcformat = COERCE_EXPLICIT_CALL, args = 0x1660a20, location = 5}
|
outargs -> [List] // transformed output-argument expressions
CallStmt的三个变量funccall来自parser，funcexpr和outargs都是transformed加工后的结构，参数列表在funcexpr中，注意只有入参的参数列表。
CallStmt->funccall：保存全部参数，用A_Const存，没有类型信息，只是初始状态的值。
funccall -> [FuncCall] // from the parser
{type = T_FuncCall, funcname = 0x165ff60, args = 0x165fff0,.,funcformat = COERCE_EXPLICIT_CALL}
funcname -> [List] -> [String] -> "tp88"
args -> [List] -> [A_Const] -> val = {type = T_Integer, val = {ival = 1, str = 0x1}}
-> [A_Const] -> val = {type = T_Integer, val = {ival = 2, str = 0x2}}
CallStmt->funcexpr：只保存IN参数：用Const存带类型
funcexpr -> [FuncExpr] // transformed call, with only input args
{xpr = {type = T_FuncExpr}, funcid = 24972, funcresulttype = 2249, funcretset = false,
funcvariadic = false, funcformat = COERCE_EXPLICIT_CALL, args = 0x1660a20, location = 5}
funcid -> 24972 ： 在系统表中找到函数添加到这里
funcresulttype -> 2249 ： 注意procedure返回的是record类型，用来给Out参数填值。虽然不能直接return：
但是在C函数中还是能返回的。SQL函数直接把结果按列吐给out类型参数。
args -> [List] -> [Const] -> {consttype = 23, consttypmod = -1, constvalue = 1}
-> 没了，只有入参，没有出参
CallStmt->outargs：只保存OUT参数：用Const存带类型
outargs -> [List] // transformed output-argument expressions
{consttype = 23, consttypmod = -1, constvalue = 2}
外层拿到上述优化器返回的CallStmt结构后，开始用FuncExpr的解析过得参数列表拼接fcinfo->args。
核心逻辑是表达式执行，拿结果。
ExecuteCallStmt
...
i = 0;
foreach(lc, fexpr->args)
{
...
exprstate = ExecPrepareExpr(lfirst(lc), estate); // 初始化表达式
val = ExecEvalExprSwitchContext(exprstate, econtext, &isnull); // 运行表达式拿值
fcinfo->args[i].value = val;
fcinfo->args[i].isnull = isnull;
i++;
}
输入：args -> [List] -> [Const] -> {consttype = 23, consttypmod = -1, constvalue = 1}
进入：ExecEvalExprSwitchContext
ExecEvalExprSwitchContext
ExecInterpExprStillValid
ExecJustConst： 直接赋值即可！
输入：args -> [List] -> [OpExpr] ->
{ xpr = {type = T_OpExpr},
opno = 551,
opfuncid = 177, -> int4pl
opresulttype = 23, -> int4
opretset = false,
opcollid = 0,
inputcollid = 0,
args = 0x1660478, -> [List] -> [Const] -> {consttype = 23, constvalue = 1}
-> [Const] -> {consttype = 23, constvalue = 1}
location = 11}
数据准备：ExecPrepareExpr
ExecPrepareExpr
expression_planner
eval_const_expressions_mutator
----> case T_OpExpr:
simplify_function
evaluate_function
evaluate_expr
ExecEvalExprSwitchContext
ExecInterpExprStillValid
ExecInterpExpr
int4pl
进入：ExecEvalExprSwitchContext
ExecEvalExprSwitchContext
ExecInterpExprStillValid
ExecJustConst
输入： args -> [List] -> [FuncExpr] ->
{ xpr = {type = T_FuncExpr},
funcid = 1397, -> abs
funcresulttype = 23, -> int4
funcretset = false,
funcvariadic = false,
funcformat = COERCE_EXPLICIT_CALL,
funccollid = 0,
inputcollid = 0,
args = 0x1660460, -> [Const] -> {consttype = 23, constvalue = 18446744073709551615}
location = 10}
数据准备：ExecPrepareExpr
ExecPrepareExpr
expression_planner
eval_const_expressions
eval_const_expressions_mutator
----> case T_FuncExpr:
simplify_function
evaluate_function
evaluate_expr
ExecEvalExprSwitchContext
ExecInterpExprStillValid
ExecInterpExpr
int4abs
进入：ExecEvalExprSwitchContext
ExecEvalExprSwitchContext
ExecInterpExprStillValid
ExecJustConst
进入内层堆栈
#0 ExecuteCallStmt (stmt=0x17afa68, params=0x176d370, atomic=false, dest=0xce9120 <spi_printtupDR>) at functioncmds.c:2203
#1 0x000000000097eea4 in standard_ProcessUtility (pstmt=0x17af9d0, queryString=0x1748ff8 "call tp99(10,20,c8)", readOnlyTree=true, context=PROCESS_UTILITY_QUERY_NONATOMIC, params=0x176d370, queryEnv=0x0, dest=0xce9120 <spi_printtupDR>, qc=0x7ffd124d0230) at utility.c:850
#2 0x000000000097e69b in ProcessUtility (pstmt=0x17b1f80, queryString=0x1748ff8 "call tp99(10,20,c8)", readOnlyTree=true, context=PROCESS_UTILITY_QUERY_NONATOMIC, params=0x176d370, queryEnv=0x0, dest=0xce9120 <spi_printtupDR>, qc=0x7ffd124d0230) at utility.c:527
#3 0x0000000000795198 in _SPI_execute_plan (plan=0x1747f80, options=0x7ffd124d0390, snapshot=0x0, crosscheck_snapshot=0x0, fire_triggers=true) at spi.c:2703
#4 0x000000000079192d in SPI_execute_plan_extended (plan=0x1747f80, options=0x7ffd124d0390) at spi.c:731
#5 0x00007fde4870b560 in exec_stmt_call (estate=0x7ffd124d0670, stmt=0x17124b0) at pl_exec.c:2197
#6 0x00007fde4870afd0 in exec_stmts (estate=0x7ffd124d0670, stmts=0x1782d90) at pl_exec.c:1995
#7 0x00007fde4870ad56 in exec_stmt_block (estate=0x7ffd124d0670, block=0x17830a8) at pl_exec.c:1910
#8 0x00007fde4870a563 in exec_toplevel_block (estate=0x7ffd124d0670, block=0x17830a8) at pl_exec.c:1608
#9 0x00007fde487085b8 in plpgsql_exec_function (func=0x169f070, fcinfo=0x7ffd124d0990, simple_eval_estate=0x0, simple_eval_resowner=0x0, procedure_resowner=0x169ff18, atomic=false) at pl_exec.c:611
#10 0x00007fde48722943 in plpgsql_call_handler (fcinfo=0x7ffd124d0990) at pl_handler.c:277
#11 0x00000000006ae193 in ExecuteCallStmt (stmt=0x16600d8, params=0x0, atomic=false, dest=0x1781f20) at functioncmds.c:2311
#12 0x000000000097eea4 in standard_ProcessUtility (pstmt=0x1660738, queryString=0x165f4e0 "call tp88(1,2);", readOnlyTree=false, context=PROCESS_UTILITY_TOPLEVEL, params=0x0, queryEnv=0x0, dest=0x1781f20, qc=0x7ffd124d1290) at utility.c:850
#13 0x000000000097e69b in ProcessUtility (pstmt=0x1660738, queryString=0x165f4e0 "call tp88(1,2);", readOnlyTree=false, context=PROCESS_UTILITY_TOPLEVEL, params=0x0, queryEnv=0x0, dest=0x1781f20, qc=0x7ffd124d1290) at utility.c:527
#14 0x000000000097d297 in PortalRunUtility (portal=0x16fa040, pstmt=0x1660738, isTopLevel=true, setHoldSnapshot=true, dest=0x1781f20, qc=0x7ffd124d1290) at pquery.c:1155
#15 0x000000000097d000 in FillPortalStore (portal=0x16fa040, isTopLevel=true) at pquery.c:1028
#16 0x000000000097c972 in PortalRun (portal=0x16fa040, count=9223372036854775807, isTopLevel=true, run_once=true, dest=0x1660828, altdest=0x1660828, qc=0x7ffd124d1460) at pquery.c:760
#17 0x000000000097663b in exec_simple_query (query_string=0x165f4e0 "call tp88(1,2);") at postgres.c:1213
#18 0x000000000097ab59 in PostgresMain (argc=1, argv=0x7ffd124d16f0, dbname=0x1688cb0 "postgres", username=0x1688c88 "mingjiegao") at postgres.c:4494
#19 0x00000000008b6d4e in BackendRun (port=0x1680a80) at postmaster.c:4530
#20 0x00000000008b66cd in BackendStartup (port=0x1680a80) at postmaster.c:4252
#21 0x00000000008b2b45 in ServerLoop () at postmaster.c:1745
#22 0x00000000008b2417 in PostmasterMain (argc=1, argv=0x16590d0) at postmaster.c:1417
#23 0x00000000007b4c93 in main (argc=1, argv=0x16590d0) at main.c:209
内层函数的调用：call tp99(i8,20,c8);
其中三个参数：
[CallStmt]
{type = T_CallStmt, funccall = 0x178ba70, funcexpr = 0x178bda8, outargs = 0x178bee8}
\
funccall -> [FuncCall] // from the parser
{type = T_FuncCall, funcname = 0x178bac8, args = 0x178bb68,.,funcformat = COERCE_EXPLICIT_CALL}
|
funcexpr -> [FuncExpr] // transformed call, with only input args
{xpr = {type = T_FuncExpr}, funcid = 24973, funcresulttype = 2249, funcretset = false,
funcvariadic = false, funcformat = COERCE_EXPLICIT_CALL, args = 0x178be00, location = 5}
|
outargs -> [List] // transformed output-argument expressions
funccall->args的值就很有意思了：对比外层调用的三个A_Const，这里有两个解析成了列引用类型。
funccall->args
-> [List]
-> [ColumnRef] -> {type = T_ColumnRef, fields = 0x178bbf8, location = 10}
-> [List] -> [Value] -> "i8"
-> [A_Const] -> {type = T_A_Const, val = {type = T_Integer, val = {ival = 20, str = 0x14}}
-> [ColumnRef] -> {type = T_ColumnRef, fields = 0x178bd08, location = 16}
-> [List] -> [Value] -> "c8"
在看看funcexpr->args：对比外层的一个Const，这里有两个入参，有一个需要取值的用Param表示。
funcexpr->args
-> [List]
-> [Param] -> {xpr = {type = T_Param}, paramkind = PARAM_EXTERN, paramid = 5,
paramtype = 23, paramtypmod = -1, paramcollid = 0, location = 10}
-> [Const] -> {consttype = 23, constvalue = 20}
答案：源头Plan是从exec_stmt_call里面拼出来的，经过_SPI_execute_plan送入优化器，生产出来CallStmt
_SPI_execute_plan: 入参plan
1. 从plan->plancache_list中拿到CachedPlanSource
2. 遍历plancache_list调用GetCachedPlan拿到cplan，需要的话重新生成PLAN
3. stmt_list = cplan->stmt_list
4. 遍历stmt_list，执行ProcessUtility传入stmt，stmt里面已经带CallStmt了
PlannedStmt {utilityStmt = 0x1761478 (CallStmt) }
_SPI_execute_plan的入参？哪来的？
SPIPlanPtr 0x177fbe0
{ magic = 569278163,
saved = true,
oneshot = false,
plancache_list = 0x177fc38,
plancxt = 0x177fac0,
parse_mode = RAW_PARSE_DEFAULT,
cursor_options = 0,
nargs = 0,
argtypes = 0x0,
parserSetup = 0x7fde48705264 ,
parserSetupArg = 0x1783758}
exec_stmt_call
SPI_execute_plan_extended(expr->plan, &options)
expr->plan 就是入参PLAN
后面看exec_stmt_call的时候看看PLAN是怎么生成的。
内层拿到的funcexpr->args有一个Param，我们看看这个参数的值是怎么拿到的
funcexpr->args
-> [List]
-> [Param] -> {xpr = {type = T_Param}, paramkind = PARAM_EXTERN, paramid = 5,
paramtype = 23, paramtypmod = -1, paramcollid = 0, location = 10}
-> [Const] -> {consttype = 23, constvalue = 20}
代码还是不变：
ExecuteCallStmt
...
i = 0;
foreach(lc, fexpr->args)
{
...
exprstate = ExecPrepareExpr(lfirst(lc), estate); // 初始化表达式
val = ExecEvalExprSwitchContext(exprstate, econtext, &isnull); // 运行表达式拿值
fcinfo->args[i].value = val;
fcinfo->args[i].isnull = isnull;
i++;
}
输入
[Param] -> {xpr = {type = T_Param}, paramkind = PARAM_EXTERN, paramid = 5,
paramtype = 23, paramtypmod = -1, paramcollid = 0, location = 10}
ExecPrepareExpr：没做什么具体工作
ExecPrepareExpr
expression_planner
eval_const_expressions_mutator
----> case T_Param:
return (Node *) copyObject(param)
ExecInitExpr
ExecInitExprSlots ------ 执行表达式后面统一整理一篇
ExecInitExprRec
ExprEvalPushStep
ExecReadyExpr
ExecEvalExprSwitchContext
ExecEvalExprSwitchContext
ExecInterpExprStillValid
ExecInterpExpr ----- 表达式执行核心函数
----> EEO_CASE(EEOP_PARAM_EXTERN)
ExecEvalParamExtern
1. 拿到ParamListInfo：{paramFetch = 0x7fde48712d96 ,
paramFetchArg = 0x7ffd124d0670,
paramCompile = 0x7fde4871305c ,
paramCompileArg = 0x0,
parserSetup = 0x7fde48705264 ,
parserSetupArg = 0x1783758,
paramValuesStr = 0x0,
numParams = 6,
params = 0x1743b90}
2. 发现拿到ParamListInfo的拿数据钩子存在：paramFetch = 0x7fde48712d96
3. 进入PL逻辑plpgsql_param_fetch拿值存入ParamExternData
结束
typedef struct ParamListInfoData
{
ParamFetchHook paramFetch; /* parameter fetch hook */
void *paramFetchArg;
ParamCompileHook paramCompile; /* parameter compile hook */
void *paramCompileArg;
ParserSetupHook parserSetup; /* parser setup hook */
void *parserSetupArg;
char *paramValuesStr; /* params as a single string for errors */
int numParams; /* nominal/maximum # of Params represented */
/*
* params[] may be of length zero if paramFetch is supplied; otherwise it
* must be of length numParams.
*/
ParamExternData params[FLEXIBLE_ARRAY_MEMBER];
} ParamListInfoData;
当前值
{ paramFetch = 0x7fde48712d96 ,
paramFetchArg = 0x7ffd124d0670,
paramCompile = 0x7fde4871305c ,
paramCompileArg = 0x0,
parserSetup = 0x7fde48705264 ,
parserSetupArg = 0x1783758,
paramValuesStr = 0x0,
numParams = 6,
params = 0x1743b90}
看下外层函数
CREATE or replace PROCEDURE tp88(a8 in integer, b8 out integer)
LANGUAGE plpgsql
AS $$
DECLARE
i8 int;
c8 int;
BEGIN
i8 := 100;
call tp99(i8,20,c8);
raise notice '=====tp88=====';
raise notice 'get from tp99 c8: %', c8;
END;
$$;
call tp88(1,2);
在进入exec_stmt_block时：会有一些变量清空/赋默认值 的逻辑，那么哪些应该初始化？如何决定的？
static int
exec_stmt_block(PLpgSQL_execstate *estate, PLpgSQL_stmt_block *block)
{
volatile int rc = -1;
int i;
/*
* First initialize all variables declared in this block
*/
estate->err_text = gettext_noop("during statement block local variable initialization");
for (i = 0; i < block->n_initvars; i++)
{
int n = block->initvarnos[i];
PLpgSQL_datum *datum = estate->datums[n];
/*
* The set of dtypes handled here must match plpgsql_add_initdatums().
*
* Note that we currently don't support promise datums within blocks,
* only at a function's outermost scope, so we needn't handle those
* here.
*/
switch (datum->dtype)
{
case PLPGSQL_DTYPE_VAR:
{
PLpgSQL_var *var = (PLpgSQL_var *) datum;
/*
* Free any old value, in case re-entering block, and
* initialize to NULL
*/
assign_simple_var(estate, var, (Datum) 0, true, false);
if (var->default_val == NULL)
{
/*
* If needed, give the datatype a chance to reject
* NULLs, by assigning a NULL to the variable. We
* claim the value is of type UNKNOWN, not the var's
* datatype, else coercion will be skipped.
*/
if (var->datatype->typtype == TYPTYPE_DOMAIN)
exec_assign_value(estate,
(PLpgSQL_datum *) var,
(Datum) 0,
true,
UNKNOWNOID,
-1);
/* parser should have rejected NOT NULL */
Assert(!var->notnull);
}
else
{
exec_assign_expr(estate, (PLpgSQL_datum *) var,
var->default_val);
}
}
break;
case PLPGSQL_DTYPE_REC:
{
PLpgSQL_rec *rec = (PLpgSQL_rec *) datum;
/*
* Deletion of any existing object will be handled during
* the assignments below, and in some cases it's more
* efficient for us not to get rid of it beforehand.
*/
if (rec->default_val == NULL)
{
/*
* If needed, give the datatype a chance to reject
* NULLs, by assigning a NULL to the variable.
*/
exec_move_row(estate, (PLpgSQL_variable *) rec,
NULL, NULL);
/* parser should have rejected NOT NULL */
Assert(!rec->notnull);
}
else
{
exec_assign_expr(estate, (PLpgSQL_datum *) rec,
rec->default_val);
}
}
break;
default:
elog(ERROR, "unrecognized dtype: %d", datum->dtype);
}
}
...
...
关注block->n_initvars，发现这个值只在pl_gram.y中有过赋值：那么就是在语法解析时就决定了执行时，应该对哪些值做清零。
pl_block是上层的执行语法块，注意到n_initvars是从声明区域拿到的decl_sect：
pl_block : decl_sect K_BEGIN proc_sect exception_sect K_END opt_label
{
PLpgSQL_stmt_block *new;
new = palloc0(sizeof(PLpgSQL_stmt_block));
new->cmd_type = PLPGSQL_STMT_BLOCK;
new->lineno = plpgsql_location_to_lineno(@2);
new->stmtid = ++plpgsql_curr_compile->nstatements;
new->label = $1.label;
new->n_initvars = $1.n_initvars;
new->initvarnos = $1.initvarnos;
new->body = $3;
new->exceptions = $4;
check_labels($1.label, $6, @6);
plpgsql_ns_pop();
$$ = (PLpgSQL_stmt *)new;
}
;
当前语法块的声明区域：
只有带变量声明区域的规则有可能返回initvarnos，因为decl_sect的声明结构是declhdr{char *label;int n_initvars;int *initvarnos;}所以这里可以直接赋值。
decl_sect : opt_block_label
{
/* done with decls, so resume identifier lookup */
plpgsql_IdentifierLookup = IDENTIFIER_LOOKUP_NORMAL;
$$.label = $1;
$$.n_initvars = 0;
$$.initvarnos = NULL;
}
| opt_block_label decl_start
{
plpgsql_IdentifierLookup = IDENTIFIER_LOOKUP_NORMAL;
$$.label = $1;
$$.n_initvars = 0;
$$.initvarnos = NULL;
}
| opt_block_label decl_start decl_stmts
{
plpgsql_IdentifierLookup = IDENTIFIER_LOOKUP_NORMAL;
$$.label = $1;
/* Remember variables declared in decl_stmts */
$$.n_initvars = plpgsql_add_initdatums(&($$.initvarnos));
}
;
继续看plpgsql_add_initdatums函数：总结下逻辑，就是把所有PLPGSQL_DTYPE_VAR、PLPGSQL_DTYPE_REC类型的记录下来。
也就是所有VAR、REC类型都需要初始化，清零 或 赋默认值。
int
plpgsql_add_initdatums(int **varnos)
{
int i;
int n = 0;
for (i = datums_last; i < plpgsql_nDatums; i++)
{
switch (plpgsql_Datums[i]->dtype)
{
case PLPGSQL_DTYPE_VAR:
case PLPGSQL_DTYPE_REC:
n++;
break;
default:
break;
}
}
if (varnos != NULL)
{
if (n > 0)
{
*varnos = (int *) palloc(sizeof(int) * n);
n = 0;
for (i = datums_last; i < plpgsql_nDatums; i++)
{
switch (plpgsql_Datums[i]->dtype)
{
case PLPGSQL_DTYPE_VAR:
case PLPGSQL_DTYPE_REC:
(*varnos)[n++] = plpgsql_Datums[i]->dno;
default:
break;
}
}
}
else
*varnos = NULL;
}
datums_last = plpgsql_nDatums;
return n;
}
结论：为了赋默认值，默认值是不在执行块中的，如果不在进入时赋值，那么就永远拿不到默认值了。
一开始我以为是因为内外层共享变量的原因：结果测试发现PG的内外层block是完全隔离的，嵌套块执行也不需要重新初始化变量。
CREATE or replace PROCEDURE tp1()
LANGUAGE plpgsql
AS $$
DECLARE
i int;
BEGIN
i := 1;
raise notice 'outter i: %', i;
DECLARE
i int;
BEGIN
i := 2;
raise notice 'inner i: %', i;
END;
raise notice 'outter i: %', i;
END;
$$;
call tp1();
postgres=# call tp1();
NOTICE: outter i: 1
NOTICE: inner i: 2
NOTICE: outter i: 1
CALL
看第一次进入exec_stmt_block的变量就知道了，这里有两个i，完全没关系。
(gdb) p *(PLpgSQL_var*)estate->datums[0]
$6 = { dtype = PLPGSQL_DTYPE_VAR, dno = 0, refname = 0x17b22b0 "found", lineno = 0,
isconst = false, notnull = false, default_val = 0x0, datatype = 0x17b2238,
cursor_explicit_expr = 0x0, cursor_explicit_argrow = 0, cursor_options = 0,
value = 0, isnull = false, freeval = false, promise = PLPGSQL_PROMISE_NONE}
(gdb) p *(PLpgSQL_var*)estate->datums[1]
$7 = { dtype = PLPGSQL_DTYPE_VAR, dno = 1, refname = 0x17b2e08 "i", lineno = 3,
isconst = false, notnull = false, default_val = 0x0, datatype = 0x17b2cf8,
cursor_explicit_expr = 0x0, cursor_explicit_argrow = 0, cursor_options = 0,
value = 0, isnull = true, freeval = false, promise = PLPGSQL_PROMISE_NONE}
(gdb) p *(PLpgSQL_var*)estate->datums[2]
$8 = { dtype = PLPGSQL_DTYPE_VAR, dno = 2, refname = 0x176cce8 "i", lineno = 8,
isconst = false, notnull = false, default_val = 0x0, datatype = 0x176cbd8,
cursor_explicit_expr = 0x0, cursor_explicit_argrow = 0, cursor_options = 0,
value = 0, isnull = true, freeval = false, promise = PLPGSQL_PROMISE_NONE}
在exec_stmt_call最后给外层赋值的：
(gdb) bt
#0 assign_simple_var (estate=0x7ffd124d0670, var=0x17b3490, newvalue=88299, isnull=false, freeable=false) at pl_exec.c:8446
#1 0x00007fde4871120e in exec_assign_value (estate=0x7ffd124d0670, target=0x17b3490, value=88299, isNull=false, valtype=23, valtypmod=-1) at pl_exec.c:5084
#2 0x00007fde487146bc in exec_move_row_from_fields (estate=0x7ffd124d0670, target=0x1783c20, newerh=0x0, values=0x7ffd124d0140, nulls=0x7ffd124d0100, tupdesc=0x17a5740) at pl_exec.c:7178
#3 0x00007fde48713c9a in exec_move_row (estate=0x7ffd124d0670, target=0x1783c20, tup=0x17a5858, tupdesc=0x17a5740) at pl_exec.c:6791
#4 0x00007fde4870b69f in exec_stmt_call (estate=0x7ffd124d0670, stmt=0x17836c0) at pl_exec.c:2227
#5 0x00007fde4870afd0 in exec_stmts (estate=0x7ffd124d0670, stmts=0x1783628) at pl_exec.c:1995
#6 0x00007fde4870ad56 in exec_stmt_block (estate=0x7ffd124d0670, block=0x1783ae8) at pl_exec.c:1910
#7 0x00007fde4870a563 in exec_toplevel_block (estate=0x7ffd124d0670, block=0x1783ae8) at pl_exec.c:1608
#8 0x00007fde487085b8 in plpgsql_exec_function (func=0x169f070, fcinfo=0x7ffd124d0990, simple_eval_estate=0x0, simple_eval_resowner=0x0, procedure_resowner=0x169ff18, atomic=false) at pl_exec.c:611
#9 0x00007fde48722943 in plpgsql_call_handler (fcinfo=0x7ffd124d0990) at pl_handler.c:277
#10 0x00000000006ae193 in ExecuteCallStmt (stmt=0x16600d8, params=0x0, atomic=false, dest=0x1781f20) at functioncmds.c:2311
#11 0x000000000097eea4 in standard_ProcessUtility (pstmt=0x1660738, queryString=0x165f4e0 "call tp88(1,2);", readOnlyTree=false, context=PROCESS_UTILITY_TOPLEVEL, params=0x0, queryEnv=0x0, dest=0x1781f20, qc=0x7ffd124d1290) at utility.c:850
#12 0x000000000097e69b in ProcessUtility (pstmt=0x1660738, queryString=0x165f4e0 "call tp88(1,2);", readOnlyTree=false, context=PROCESS_UTILITY_TOPLEVEL, params=0x0, queryEnv=0x0, dest=0x1781f20, qc=0x7ffd124d1290) at utility.c:527
#13 0x000000000097d297 in PortalRunUtility (portal=0x16fa040, pstmt=0x1660738, isTopLevel=true, setHoldSnapshot=true, dest=0x1781f20, qc=0x7ffd124d1290) at pquery.c:1155
#14 0x000000000097d000 in FillPortalStore (portal=0x16fa040, isTopLevel=true) at pquery.c:1028
#15 0x000000000097c972 in PortalRun (portal=0x16fa040, count=9223372036854775807, isTopLevel=true, run_once=true, dest=0x1660828, altdest=0x1660828, qc=0x7ffd124d1460) at pquery.c:760
#16 0x000000000097663b in exec_simple_query (query_string=0x165f4e0 "call tp88(1,2);") at postgres.c:1213
#17 0x000000000097ab59 in PostgresMain (argc=1, argv=0x7ffd124d16f0, dbname=0x1688cb0 "postgres", username=0x1688c88 "mingjiegao") at postgres.c:4494
#18 0x00000000008b6d4e in BackendRun (port=0x1680a80) at postmaster.c:4530
#19 0x00000000008b66cd in BackendStartup (port=0x1680a80) at postmaster.c:4252
#20 0x00000000008b2b45 in ServerLoop () at postmaster.c:1745
#21 0x00000000008b2417 in PostmasterMain (argc=1, argv=0x16590d0) at postmaster.c:1417
#22 0x00000000007b4c93 in main (argc=1, argv=0x16590d0) at main.c:209
过程：
exec_stmt_call
第一步：从上次执行结果中拿值记录到SPI内部。
SPI_execute_plan_extended
_SPI_execute_plan
SPITupleTable *my_tuptable = NULL;
...
ProcessUtility
ExecuteCallStmt
plpgsql_call_handler
plpgsql_exec_function
exec_toplevel_block
exec_stmt_block
begin_tup_output_tupdesc // CALL完了开始构造输出tuple
spi_dest_startup
_SPI_current->tuptable = tuptable = (SPITupleTable *) palloc0(sizeof(SPITupleTable));
ExecStoreHeapTuple
tstate->dest->receiveSlot // 把结果tuple塞到_SPI_current->tuptable里面
my_tuptable = _SPI_current->tuptable;
...
SPI_tuptable = my_tuptable;
第二步：拿出来向变量列表中塞值。
SPITupleTable *tuptab = SPI_tuptable;
exec_move_row(estate, stmt->target, tuptab->vals[0], tuptab->tupdesc);
ExecuteCallStmt
...
// 调用函数
retval = FunctionCallInvoke(fcinfo);
/* Handle the procedure's outputs */
if (fexpr->funcresulttype == VOIDOID)
{
/* do nothing */
}
如果结果是Record类型，需要把结果整理到SPI记录结果的全局变量中。
else if (fexpr->funcresulttype == RECORDOID)
{
/* send tuple to client */
HeapTupleHeader td;
Oid tupType;
int32 tupTypmod;
TupleDesc retdesc;
HeapTupleData rettupdata;
TupOutputState *tstate;
TupleTableSlot *slot;
if (fcinfo->isnull)
elog(ERROR, "procedure returned null record");
EnsurePortalSnapshotExists();
拼接一个物理元组HeapTupleData
td = DatumGetHeapTupleHeader(retval);
tupType = HeapTupleHeaderGetTypeId(td);
tupTypmod = HeapTupleHeaderGetTypMod(td);
retdesc = lookup_rowtype_tupdesc(tupType, tupTypmod);
tstate = begin_tup_output_tupdesc(dest, retdesc,
&TTSOpsHeapTuple);
rettupdata.t_len = HeapTupleHeaderGetDatumLength(td);
ItemPointerSetInvalid(&(rettupdata.t_self));
rettupdata.t_tableOid = InvalidOid;
rettupdata.t_data = td;
用物理元组HeapTupleData组装一个tts，因为DestReceiver只能传入tts。
slot = ExecStoreHeapTuple(&rettupdata, tstate->slot, false);
执行SPI的DestReceiver，把tts的信息解析出来装入SPI结果中。
tstate->dest->receiveSlot(slot, tstate->dest);
end_tup_output(tstate);
ReleaseTupleDesc(retdesc);
}
else
elog(ERROR, "unexpected result type for procedure: %u",
fexpr->funcresulttype);
FreeExecutorState(estate);
}
exec_stmt_call的最后一步把SPI记录的结果元组，赋值给out类型的变量
exec_stmt_call
...
...
/*
* Check result rowcount; if there's one row, assign procedure's output
* values back to the appropriate variables.
*/
if (SPI_processed == 1)
{
SPITupleTable *tuptab = SPI_tuptable;
if (!stmt->is_call)
elog(ERROR, "DO statement returned a row");
exec_move_row(estate, stmt->target, tuptab->vals[0], tuptab->tupdesc);
}