Re: Strange account activation problem

Hi, Daniel O'Connor wrote:

I had a brief look at pearweb code, there are some obvious problems. First, confirmation of account on bug page is checked via a flag in bugdb table, while it is checked via absence of record in bug_account_request table on resend-request page.

Next, the biggest problem is related to pearweb using The Most Popular Open Source Database: My$QL doesn't allow having fulltext index on a transactional table, so bugdb table has no constraints and / or foreign keys. Therefore confirmRequest() method is extremely fragile, with a large amount of updates outside a transaction block and without any constraint checking. Also there is that strange PEAR_Bug_Accountrequest::cleanOldRequests() method that probably does not work as expected, since the unconfirmed bug linked above is still here.

Oh god my eyes. It does work in the constructor.

Yeah, it does *work* but probably doesn't work *as expected*... I have a feeling that a record is deleted from bug_account_request somehow, but all the other stuff is left intact, and thus a person can never confirm it afterwards. BTW, can someone having access to an actual DB test this hypothesis?

*shivver* I'm a bit scared of charging around and refactoring that because its huge, and there is no test coverage.

Yep, touching it without test coverage is a no-no.

Alexey, I don't suppose you'd mind generating a bunch of small (obvious) test cases which break it / adding bug reports for them/throwing them into svn?

OK, seems I'll finally have to defile my computer by installing a dolphin-shaped abomination that is MySQL... I'll see what I can do once I understand how the current test suite for pearweb is organized.