Hello,
Jorgen Loland a écrit, Le 19.05.2011 13:10:
> Guilhem,
>
> You have my blessing on the security patch. The security checks look
> good, but I'm trusting Dmitri's judgment on the correctness. A few notes:
>
> === modified file 'sql/sql_prepare.cc'
> --- sql/sql_prepare.cc 2011-03-21 17:55:41 +0000
> +++ sql/sql_prepare.cc 2011-05-05 11:28:31 +0000
> @@ -2021,8 +2018,7 @@ static bool check_prepared_statement(Pre
> if (res == 2)
> {
> /* Statement and field info has already been sent */
> - res= FALSE;
> - goto end;
> + DBUG_RETURN(FALSE);
>
> jl: false
In fact, I had changed this code in the past in this WL, but then
noticed I didn't need the change, so reverted back to how it is in
trunk. Which is why there is this old-looking FALSE.
This way, the diff between wl4800 and trunk is only:
[T35 14:28 /m/bzrrepos_new/mysql-next-mr-opt-backporting-wl4800 $]
bzr diff -r3182.15.136 sql/sql_prepare.cc
=== modified file 'sql/sql_prepare.cc'
--- sql/sql_prepare.cc 2011-04-15 12:14:35 +0000
+++ sql/sql_prepare.cc 2011-05-06 12:05:20 +0000
@@ -113,6 +113,7 @@
#include <mysql_com.h>
#endif
#include "lock.h" //
MYSQL_OPEN_FORCE_SHARED_MDL
+#include "opt_trace.h" // Opt_trace_object
/**
A result class used to send cursor rows using the binary protocol.
@@ -1964,6 +1965,17 @@
if (tables)
thd->warning_info->opt_clear_warning_info(thd->query_id);
+ /*
+ For the optimizer trace, this is the symmetric, for statement
preparation,
+ of what is done at statement execution (in mysql_execute_command()).
+ */
+ Opt_trace_start ots(thd, tables, sql_command,
+ thd->query(), thd->query_length(), NULL,
+ thd->variables.character_set_client);
+
+ Opt_trace_object trace_command(&thd->opt_trace);
+ Opt_trace_array trace_command_steps(&thd->opt_trace, "steps");
+
switch (sql_command) {
case SQLCOM_REPLACE:
case SQLCOM_INSERT:
which is really what wl4800 needs to work.
Can I leave it as it is? I'd prefer the wl4800 final patch to not have
"style changes just for pleasure".
> @@ -2107,19 +2102,15 @@ static bool check_prepared_statement(Pre
> + DBUG_RETURN(stmt->is_sql_prepare() ?
> + FALSE : (send_prep_stmt(stmt, 0) ||
> thd->protocol->flush()));
> +error:
> + DBUG_RETURN(TRUE);
>
> jl: false, true
same here
> === modified file 'sql/opt_trace.h'
> @param support_dbug_or_support_missing_priv 'true' if this statement
> should have its trace in the dbug log (--debug),
> or if missing_privilege() may be called on this
> trace
>
> jl: support_dbug_or_support_missing_priv is an awfully long name. Can
> you make it shorter, at least by removing the second "support"?
Now using support_dbug_or_missing_priv.
Pushed.
