Hi everyone,
Recently I decided to take up issue #34772 because, of all things, I was offered a bounty to fix it! I successfully fixed the bug, but another developer, seeing that I had assigned the issue to mysef, posted their own patch which is very different to what I did, but succeeds in what it was advertised to do. The two patches can't really be applied together (at least I think I can't), and I want to know what the policy is when it comes to two different solutions on offer.
The developer, Sergei Gorelkin, gave me his patch to help save me some work, but was fine as it was, so I was surprised he didn't upload it earlier, given that this issue has been known since last December. Granted, I had already fixed the bug but had not let uploaded it due to double-checking things with the bounty and then because of the server maintenance.
Not to blow my own trumpet, but I want to say my patch is slightly better because it's clearer as to what the code is doing to mitigate the issue, and in one particular case (an implicit finally block by itself), the compiler produces slightly more efficient code with "Exit" - Sergei's patch puts the jump destination before a "nop" instruction, while my code (and the trunk) places it after said "nop".
Still, either way, the patch fixes the problem. I also wrote an extensive test, "tw34772.pp" that checks all different kinds of procedures with implicit finally blocks, safecall and explicit finally and except blocks to see if everything is fixed, along with overridden GetMem and FreeMem routines that track how many times they are called (increments for each GetMem and similar calls, decrements for each FreeMem call, so a pass is given if ReferenceCount = 0).
Gareth