maria-developers team mailing list archive

[Kristian Nielsen <knielsen@xxxxxxxxxxxxxxx>]

Oleksandr Byelkin <sanja@xxxxxxxxxxxxxxxx> writes:

>>> Making something more I think do not worth it because:
>>> 1) in 10.2 C/C will replace the library and it has no the problem
>> But Connector-C should have exactly the same problem. If it frees the
>> options on failed connect, it will break async API. If it does not, code
>> that skips mysql_close() will leak. How does C/C solve this problem?
>> If we are making any changes in 10.0, we should preferably make things work
>> the same as they will work in 10.2.

> Don't it has rewritten async operation? (at least I got such impression).

My understanding is that Georg implemented all the async API also in
connector-c. It should be using much the same code (mysql_async.c and
such).

But this is not about the implementation. This is about the semantics of how
libmysqlclient works. My understanding is that connector-c is supposed to be
binary compatible with libmysqlclient.so?

So either connector-c removes any mysql_options() set when
mysql_real_connect() fails without CLIENT_REMEMBER_OPTIONS. Or it does
not. In the former case, there is the problem with MYSQL_OPT_NONBLOCK. In
the latter case, the problem with leak when mysql_close() is not called.

I really think you need to understand the problem fully, including how this
is going to work in 10.2, before you make any changes to how the client
library works...

 - Kristian.