Avoid using uninitialised struct

[mikhainin]

In some scenarios, there is a possibility that PACKET_FREE will be called with an uninitialised fields_eof.
Added some sanity to avoid negative impact of such an event

[kamil-tekiela]

What are these scenarios you mentioned?

[mikhainin]

For example, in
mysqlnd_query_read_result_set_header, I can see two places where we break from the inner do..while (one, two) and fall through straight to PACKET_FREE() while the fields_eof is not yet initialised.

[kamil-tekiela]

Thanks for the PR. I reviewed it now. What you are proposing is not the correct solution. We do not initialize structs with = {0} but with a dedicated constructor method. Thus it is guaranteed that the "object" will call memset and will have its methods assigned.

What you found is indeed a bug. We are calling PACKET_FREE in a path where the init_* method wasn't yet called. The solution would be to move PACKET_FREE(&fields_eof); one line above to be inside the while block.

[mikhainin]

Hi @kamil-tekiela, thank you for the comment.

I updated the PR as you were suggesting.

[kamil-tekiela]

This is not needed. It is already guaranteed.

[mikhainin]

Removed

[nielsdos]

Looks right to me too, but hasn't been committed yet.
I'll commit this to 8.1+ so it doesn't get lost.