Thrown errors are uncatchable

[Julie Iaccarino]

It is not possible to catch errors thrown by libzdb during operation.

Code example below:

a arglist.v.list[1].v.str of “butts” will result in !url being true, and a successful handling of the incorrect input.

However a postgresql://user:localhost/moo url, or other ‘semi valid’ url, with incorrect or missing logging details causes a full SIG11 fault.

        url = URL_new(arglist.v.list[1].v.str);
        if (url)
        {
            try {
                TRY
                {
                    pool = ConnectionPool_new(url);    
                    if (!pool) {
                        ret->type = TYPE_ERR;
                        ret->v.err = E_PERM;    
                    }
                }
                ELSE {
                    ret->type = TYPE_ERR;
                    ret->v.err = E_PERM;
                }
                END_TRY;
            } catch (...) 
            {
                ret->type = TYPE_ERR;
                ret->v.err = E_PERM;
            }
        } else {
            ret->type = TYPE_ERR;
            ret->v.err = E_INVARG;
        }

Since these values can be provided during runtime of the application, this is a critical bug as it would be possible to crash the entire server.

Other SQLException errors are similarly uncatchable, even when using the macro patterns in Exception.h.

[Tildeslash]

You cannot send a C++ string to URL_new which expect a c_str. From C++, you should import zdbpp.h instead and use this API

[Julie Iaccarino]

I’m sorry, I actually am using zdbpp.h.

[Julie Iaccarino]

Just to clarify are you saying there isn’t documentation for zdbpp.h, so using the idiomatic syntax present on the webpage examples will result in the c interface being used, causing anomalous behavior like what I’m seeing, and I should instead just puzzle out how to use the zdbpp.h interface from the header file itself?

[Julie Iaccarino]

If that’s correct, updating the documentation would be fantastic, as that is not clear.

[Julie Iaccarino]

I assumed I interpreted you correctly and I went ahead and made the change and received the same error.

‌

        try {
            TRY
            {
                ConnectionPool pool = ConnectionPool(arglist.v.list[1].v.str);    
                if (!pool) {
                    ret->type = TYPE_ERR;
                    ret->v.err = E_PERM;    
                }
            }
            ELSE {
                ret->type = TYPE_ERR;
                ret->v.err = E_PERM;
            }
            END_TRY;
        } catch (...) 
        {
            ret->type = TYPE_ERR;
            ret->v.err = E_PERM;
        }

SIG11 and program failure.

‌

‌

[Julie Iaccarino]

Some more information I’ve found: This happens only if I try to run this code off the main thread, if I run it on the main thread I actually do catch the exception.

[Tildeslash]

http://tildeslash.com/libzdb/#api demonstrates the main use cases for the C++ API. (scroll down to C++ examples). There is also this example code https://bitbucket.org/tildeslash/libzdb/src/master/test/zdbpp.cpp

Best practice is to create and start the Connection Pool on the main thread. The Connection Pool is thread-safe, but initialisation of databases like MySQL, should be done on the main thread.

Ps. Don’t use the C Exception macros in C++. Just use regular C++ try/catch

[Julie Iaccarino]

Thanks Tildeslash, have a great holiday