Fix a SEGV when using mariadb connector

[atoomic]

When using mariadb connector 3.0.3 with DBD::mysql
we can notice a segfault on reconnection due
to the fact that the mysql extensions are not reset.

Doing the extra init as performed by DBD::mariadb
with some extra save and restore fixes the issue.

[atoomic]

note: that appveyor CI issue seems not related to the issue

Building Time-HiRes-1.9755 ... ! Installing Time::HiRes failed. See C:\Users\appveyor\.cpanm\work\1521052705.1676\build.log for details. Retry with --force to force install it.
166FAIL

[atoomic]

error discussed upstream here: https://jira.mariadb.org/browse/CONC-316

[pali]

This patch is of course incorrect/incomplete and not for production usage. I created it for DBD::MariaDB just as quick fix until I fix it properly.

[atoomic]

@pali thanks for your input, I understand that this code is not optimized and might be incomplete, when trying to restore the context on a freshly created connection, but I think in most cases disconnect/reconnect should not be frequent, and clearly prefer a full reconnection rather than a segfault at this point.

What other issues do you have in mind? Any known problems with this approach outside performance?

Would you not consider using mariadb connector 3.0.3 with DBD::mysql? Would you prefer using DBD::MariaDB instead?

thanks

[pali]

I will try to fix this problem properly without touching internal MYSQL structures... basically it is needed for proper support when compiled against new version of MariaDB Connector/C.

For production code I suggest to stick with old client libraries MySQL 5.6/5.7 or MariaDB 10.0, both are still supported by upstream.

For testing you can use DBD::MariaDB with Connector/C 3.x or MariaDB 10.3, but be aware that there can be changes in future, prior uploading version to CPAN.

[pali]

As wrote this copied code in this pull request is wrong and should not be used.

Proper fix for this problem is now in DBD::MariaDB master branch from this pull request:
perl5-dbi/DBD-MariaDB#63

[dveeden]

Closing this as this is now handled in #291 ( dc32f84 ).