Обсуждение: Re: [COMMITTERS] pgsql: Properly unregister OpenSSL callbacks when libpq is done with

[ still catching up on back email ]

mha@postgresql.org (Magnus Hagander) writes:
> Properly unregister OpenSSL callbacks when libpq is done with
> it's connection. This is required for applications that unload
> the libpq library (such as PHP) in which case we'd otherwise
> have pointers to these functions when they no longer exist.

I find it fairly disturbing that this patch was committed with a bug
that ensured it wouldn't even compile for the case of SSL and
THREAD_SAFETY both enabled.  Which one would think would have been
the primary case to test.  Would anyone like to reassure us why this
patch shouldn't just be reverted in toto until it's actually been
tested a bit?
        regards, tom lane

Tom Lane wrote:
> [ still catching up on back email ]
> 
> mha@postgresql.org (Magnus Hagander) writes:
>> Properly unregister OpenSSL callbacks when libpq is done with
>> it's connection. This is required for applications that unload
>> the libpq library (such as PHP) in which case we'd otherwise
>> have pointers to these functions when they no longer exist.
> 
> I find it fairly disturbing that this patch was committed with a bug
> that ensured it wouldn't even compile for the case of SSL and
> THREAD_SAFETY both enabled.  Which one would think would have been
> the primary case to test.  Would anyone like to reassure us why this
> patch shouldn't just be reverted in toto until it's actually been
> tested a bit?

The patch itself was tested. I applied the wrong version of the patch to
the main tree when I moved it from my testing tree to the one where I
apply it from :-( (which was configured without thread-safety)

That part was also included in the part of the patch that was tested by
the PHP guy who originally reported the problem.

//Magnus