If you have 6.12.26 Manager problems, where when you close BOINC Manager you find that tasks continue to run and you have to restart BOINC Manager and then exit it again to get the exit dialog, then yes, this is a bug in 6.12.26. It has been fixed already.

Please instead try 6.12.33, available from http://boinc.berkeley.edu/download_all.php. It is available for Windows, Linux and Macintosh and has multiple other bug fixes:

Change Log 6.12.26 => 6.12.27:

• client: Detect processor counts when the number of processors exceed 64 on Windows.
  
• client: change the handling of account manager replies:
  
  --> Old:
  If the AM sends us a project we're already attached to, and the authenticator is different, print an error message and don't change anything.
  
  --> Problem:
  If the AM is using weak authenticators, and the user has changed their password, the weak authenticator changes. In this case the AM will send the new weak auth, the client will ignore it, and all subsequent scheduler RPCs will fail until the user removes/adds the project.
  
  --> Solution:
  If the AM sends us a new auth for a project, use it.
  
  Note: From the time the password is changed on the project to the next AM RPC, the client will have a bad weak auth and scheduler RPCs will fail. That's OK.
  
• client: update to previous commit: only change the authenticator if both old and new authenticators are weak (strong authenticators don't change).
  
• client: add <http_transfer_timeout_bps> config option. Defines the transfer rate below which the connection is considered idle, and timeout starts.
  
• client: implement "report immediately" at the level of individual jobs rather than globally. To use this, projects must add <report_immediately/> to the <result> elements in job templates.
  
• client: don't check memory usage immediately after hibernation.
  
• client: show HTTP errors in text form as well as number.
  
• client: fix problem with recursion in dir_size(), this time for real.
  
• client/manager: don't show resource usage for non-CPU-intensive jobs.
  
• client: XML-escape the contents of stderr files; otherwise non-ASCII characters in client_state.xml make it invalid XML.
  
• client: fix (I think) to scheduling logic. A job is preemptable if it's finished its time slice and
  --> Old: has checkpointed in last 10 sec.
  --> New: has checkpointed since the end of the time slice.
  
• client: Account for new registry location that the new VirtualBox installer stashes the version number. Check the new location first, if not found go back to the original location.
  
• client: add log messages showing steps in exiting all tasks, enabled by <task_debug/>
  
• client: remove redundant task-start messages.
  
• client: on hibernate, don't wait for apps to exit (NCI apps never will).
  
• client: if project is set to "don't request more work", report completed tasks immediately.
  
• Mac SCR: Fix Mac crash bug when user clicks SS Test button in system preferences
  

Change Log 6.12.27 => 6.12.28:

• MGR: Fix shutting down of Client on Linux.
  
• client: escape only non-ASCII chars in stderr out, not control chars.
  

Change Log 6.12.28 => 6.12.30:

• client: equate notices if their text is the same after removing digits, so that "need 25 GB disk" and "need 24 GB disk" don't result in 2 notices.
  
• client: if <ncpus> is specified in config file, set host_info.p_ncpus to that value, so that scheduler requests report that number of CPUs.
  
• client: when killing a nonresponsive task, kill its descendant processes too.
  
• client: Rom pointed out that on Win in secure mode, the client won't be able to convert descendant PIDs to handles, and therefore won't be able to terminate them. Sigh.
  So terminate the main process using its handle (we'll still kill descendants in the non-secure case).
  
• client: my checkin of 8 Apr 2011 didn't work in some cases, e.g.:
  client finishes a job before upload starts, work fetch runs and decides to fetch work from that project.
  
  Solution:
  set PROJECT::last_upload_start when job finishes, and clear it when a job is uploaded. Defer scheduler RPC for a limited time if last_upload_start is set, even if a transfer is not active.
  
• MGR: Fix bug introduced on 6 Aug 2010: if currently connected host name is empty string, we must treat it as localhost. Implement different logic to skip exit confirmation dialog and avoid shutting down client if second instance of Manager is launched but Select Computer dialog is canceled.
  
• MGR: Cleanup compiler warnings. Fixes [trac]#1051[/trac] (From: Steffen Möller).
  
• client: condition "Got server request to delete file" messages on <file_xfer_debug>.
  
• MGR: change simple view notices window title from Messages to Notices.
  
• client: fix boinc_make_dirs() (from Josh Highley).
  
• client: allow "non_cpu_intensive" to be specified independently for different apps in a project. This is intended to support projects that use the Attic file distribution system, which needs to have a daemon running.
  
• Mac: Add new procinfo.cpp to XCode Project.
  
• MGR: Use same exit confirmation dialog wording for Linux as for Windows.
  
• MGR: tweak strings related to task FLOP size.
  
• MGR: Fix About dialog to show (PowerPC) instead of (x86) when appropriate.
  
• MGR: Rework exit dialog, client shutdown logic for second Manager instance.
  
• Shut down local Client even If not connected to it.
  

Change Log 6.12.30 => 6.12.31

• MGR: Disable list view rules for now.
  

Change Log 6.12.31 => 6.12.32:

• MGR: Restore list view rules again.
  
• MGR: Revert wording in Exit and Properties dialogs so existing translations will work.
  
• MGR: Event Log menu item restores Event Log when Minimized.
  

Change Log 6.12.32 => 6.12.33:

• MGR: Event Log menu item restores Event Log when Minimized; fixed on Mac and Windows XP, hopefully on other versions of Windows OS.
  
• client: fix bug in logic that defers work fetch if upload active.
  
• client: initialize some PROJECT fields (from Ian Hay). May fix bug in defer work fetch while uploading logic.