doStartTableScan() result not checked

[Stewart Smith]

In many places the error from doStartTableScan() is not checked. This can be spotted by adding warn_unused_result gcc attribute to it and seeing things fail.

It can also be spotted by returning an error from doStartTableScan() and instead seeing drizzled core instead of be sensible.

[Stewart Smith (stewart)]

for MySQL/MariaDB this is rnd_init and this small patch produces about 10 warnings about actual bugs in mysql 5.1.46:

--- ../mysql-5.1.46/sql/handler.h 2010-04-07 00:06:06.000000000 +1000
+++ sql/handler.h 2010-06-01 17:37:48.230727373 +1000
@@ -1188,7 +1188,7 @@
     inited=NONE;
     DBUG_RETURN(index_end());
   }
- int ha_rnd_init(bool scan)
+ int ha_rnd_init(bool scan) __attribute__ ((warn_unused_result))
   {
     int result;
     DBUG_ENTER("ha_rnd_init");

[Monty Taylor (mordred)]

On 06/01/2010 06:09 AM, Stewart Smith wrote:
> Public bug reported:
>
> In many places the error from doStartTableScan() is not checked. This
> can be spotted by adding warn_unused_result gcc attribute to it and
> seeing things fail.
>
> It can also be spotted by returning an error from doStartTableScan() and
> instead seeing drizzled core instead of be sensible.

We should really start adding in warn_unused_result attributes to things.

Should I perhaps make us a pandora-build test/define for it?

[Michael Widenius (monty)]

Fix committed to 5.2 branch

[Stewart Smith (stewart)]

On Fri, 16 Jul 2010 16:25:02 -0000, Monty Taylor <email address hidden> wrote:
> We should really start adding in warn_unused_result attributes to
> things.
>
> Should I perhaps make us a pandora-build test/define for it?

yes. something that'll "just work" on non-gcc (apart from replacing them
with gcc)

--
Stewart Smith