CQN notification objects are not freed correctly, causing segmentation faults

[K900]

The code responsible for freeing CQN message objects contains a typo that causes occasional segfaults, use-after-frees, memory leaks and other related weirdness:

dpiSubscr.cpp, line 279:

for (j = 0; j < query->numTables; j++) {
    if (query->tables[i].numRows > 0)
        dpiUtils__freeMemory(query->tables[i].rows);
}

The code erroneously uses i as a counter for freeing tables, where it should be j.

[K900]

I've also pushed a fixed branch to my fork of the repo, but the policy here prevents me from submitting a pull request.

[ddevienne]

Such a bug would not happen with C++11 and range-for :)
At one point, ODPI should take the plunge on modern C++, for a future version.
Any plans in that direction, to use the 8 years old standard now?

[anthony-tuininga]

Good catch! It isn't commonly discovered due to the simple fact that most of the time only a single table is being reported. :-)

[K900]

The way we encountered the issue was with multiple queries and a single table, making this an out-of-bounds free.

[K900]

Also, any chance of backporting this to the stable branch?

[anthony-tuininga]

You mean 3.1.x?

[K900]

Yes. I've built a fixed version locally, but it would be nice to have an upstream release.

[anthony-tuininga]

Can you verify that the changes I just made match the ones you made yourself?

[K900]

The diffs seem identical: mine, yours.

[anthony-tuininga]

Excellent. Thanks.

[anthony-tuininga]

Pushed to the v3.1.x branch as well.

[K900]

Thanks! Waiting for the next release for this and cx_Oracle then :)