tigefa4u projects Trac

Make the connection pool robust when the server uses a thread pool.

In a multi-threaded server, when a request is finished and `shutdown(tid)` is called, there might still exist some !PooledConnection objects for this `tid` which are not yet collected at this point.

In the event the thread gets reused (*) and acquires again a connection from the pool, a new connection will be created and stored in the `_active[tid]` entry. The old !PooledConnection objects will therefore have a different cnx, and this is why the `assert` used to fail.

With this change, we simply acknowledge the fact that this can happen. Also, when this happens, there's no need to take special action as the "old" connection has already been closed during `shutdown(tid)`.

Related to #3504.

(*) even without a thread pool this could happen if the OS reuses the thread `tid` for a new thread. But in this case, the likelihood for having uncollected !PooledConnection with this `tid` still around seems to be low...

Ignore duplicate deletions. Fixes #3778 and adds a specific test case for it.

Reverting r3794:3795, will solve #3778 differently.

Quick follow-up to r3794, should really fix #3778 now.

Fix caching of deletion changes, when there are multiple delete events for the same base file. Closes #3778.

Fix iCalendar compliance in roadmap, based on patch by hvr. Closes #3767.

make parsing of password files more tolerant of incorrectly formatted lines (fixes #3786)

Follow-up to r3763, which was not 100% correct. Hopefully the final word on #3504...

Fix CGI startup on Windows. Thanks to pkou for the pach! Closes #3781.

For Great Justice!