5.7.4

Howdy Folks,


This build has an updated messages dialog in the Simple GUI. The regular buttons will be replaced with graphical buttons as soon as I get the new bitmaps. Until then, imagine them looking like the close button.

This build should also remove the "compute_work_requests(): start" message.



----- Rom



Here are the list of changes:

- core client: put if() around "compute_work_request(): start" msg

- MGR: Add "Anytime" to the list of start times for the simple preferences dialog.

- MGR: Fix instructions when a connection failure happens.

- MGR: Let the core client manage the snooze timer. Changes for both the advanced gui and simple gui.

- MGR: Handle the clear preferences case better by setting the mask flags to false instead of just calling set preferences override with an empty string.

- MGR: Make sure we grab the correct localized dialog name during construction.

- MGR: Revamp the messages dialog so that it is resizable.

- MGR: Remember the size of the various message columns.

- MGR: Remember the size of the dialog.

- MGR: Add clipboard functionality.

- MGR: Cleanup code in various other places.

- MGR: Change points to credits.

---

Kathryn :o)

5.8.1

Howdy Folks,


We are looking for show stoppers, if none are found this will be the build for public release. Showstoppers are crashes, data corruption issues, and things that cause people to think the software just doesn't work. We'll have to follow up with a release or two for localization changes, and anything we missed in testing.



Changes for this release:

1. Help system updated so that hitting F1, Help, or the ? on dialogs launches a browser and points them to some online help. The Simple GUI page is blank at the moment but that'll be fixed before launch.

2. Simplified snooze to mean suspend only the processor. Suspending the processor from the advanced gui and then switching over to the simple gui should display the resume button instead of the pause button.

3. Fixed remembering the messages dialog column widths as well as the dialog width and height.

4. Fixed screensaver bug where the screensaver didn't shutdown when the user just did a single mouse click.

5. Added the ability to migrate CPDNBBC data to the BOINC directory. This will only work on Windows 2000 or better and only if the BOINC directory does not already exist.

6. Added the ability to save and restore options the user selects during setup. This will only work from this version of BOINC and going forward. Two fields that are not stored is the launch program checkbox at the end, and the service password. Saving and restoring setup values only works if the Setup GUI is displayed.

7. On Windows XP or better, setup no longer requires that the service account has a password.

---

Kathryn :o)

5.8.2

Howdy Folks,

Here is another build, please log results to http://boinc.berkeley.edu/alpha/


Thanks in advance.



----- Rom

Change List:



- MGR: Fix segfault in simple gui when message button is clicked. SetBackgroundStyle() needs to be called after the window is created.

- core client: if swap space is zero, it must not have been measured on this host; don't use it in process scheduling

- DIAG: Clean up some warnings and add a new exception type to the list of known exceptions.

- Win SETUP: Make Windows XP or better require passwords for the service accounts again.

- Win SETUP: Fix a spelling mistake that JM7 found.

- core client: fixed crashing bug that happens when attaching to a project with wrong URL (e.g. setiathome.ssl.berkeley.edu). The fix allows the project to attach but prints a warning. It would be better to not allow the attach; I'll get back to this later

- core client: Change function that calculates RAM and SWAP used by a science project to return the max used by a process instead of totaling the process/child processes. Fixes the problem where the same RAM and SWAP used gets counted multiple times for some science apps.

- Manager: get rid of "CPU throttled" messages

- Manager: enhancements to Statistics tab (can click on project name in multi-project view) from Mifistor_x

- core: if we're preempting an app and leave it in memory because no checkpoint yet, say so in message

- MGR: Updates to disk usage pie charts from Frank Weiler.

- core client: fix problems when user aborts result (mark it as CPU_SCHED_PREEMPTED so scheduler doesn't try to preempt it; that confuses things)

- changed format of Deferring Comm messages

- core client: backwards > in PROJECT::some_download_stalled(). This caused the client to keep fetching work from a project with stalled downloads.

- core client: sort daily stats records in case they get out of order somehow

- manager: use user-friendly app names

- SCR: Disable the diagnostic support for release builds of the Windows screensaver. A smarter thing to do in the future is to write the log files to the 'temp' directory.

---

Kathryn :o)

5.8.3

Howdy Folks,

Here is another build for testing. Please report any issues to http://boinc.berkeley.edu/alpha.


----- Rom



Change List:

- manager: use user-friendly app names (fix to work w/ 5.4.11)

- MGR: Possble fix for the Simple GUI messages dialog?

- manager: don't prepend "(Activities suspended by user)" to result status. There is no longer a notion of "activities"

-MGR: Copy missing source file from HEAD branch for Frank Weiler's pie charts updates.

- MGR: Streamline the message dialog sync operation a little bit.

- core client: lots of debugging messages for cpu_sched (may remove later)

- Manager: check for ERR_DB_NOT_UNIQUE, not ERR_NONUNIQUE_EMAIL

- Update libCurl to 7.16.0 and use sync name resolves instead of async name resolves.

- core client: fix a bug that could cause computation to stop.
The problem: if we suspended a task by sending it a <quit/> message, we were changing its scheduler state to CPU_SCHED_PREEMPTED but not changing its task state (PROCESS_EXECUTING). If, for some reason, the process didn't exit right away, the scheduler would treat it as still running, since it builds its "currently running" list based on task_state.

Solution: the scheduler shouldn't be looking at task_state. Use scheduler_state instead. It's still a mystery why the app wasn't quitting.

---

Kathryn :o)

5.8.4

Howdy Folks,

An Item of note, the manager will now attempt to restart the CC if the CC has crashed and the manager started the CC in the first place. If you have been experiencing crashes with 5.8.3 please check your log files regularly for:

“garbage_collect(); still have active task for acked result; state”



This might help with getting our head around the actual problem that is causing the crash.



----- Rom

Change Log:



- MGR: Add the help button to the list of images to update when the user selects a different skin.

- Mac: Fix build script for Curl 7.15.5 (should have been done 3 months ago.)

- manager: show correct app name when connected to 5.4.11 client

- core client: reset 5-minute timer for network activity is user explicitly changes network mode

- don't start benchmarks if they're already running

- core client: if starting a file xfer, and file is already as large or larger than it should be, start transfer from the beginning. This fixes some pathological cases found by Rattledagger.

- core client: don't garbage-collect a RESULT if an ACTIVE_TASK for it still exists. In principle this shouldn't happen, but maybe it is.

- limit duration correction factor to the range 0.01 .. 100

- mem_usage.C Bug fix: procfs access cannot use 64 bit files under solaris

- Added solaris application working set calculation

- Fixed problem on solaris when _FILE_OFFSET_BITS was set before testing for procfs.h

- Removed assumption of 4k pages in procinfo_unix.C

- MGR: Make sure the result records have been updated before displaying the taskbar balloon.

- lib: change retry logic for file system ops so that it does does at least 5 retries AND takes at least 5 seconds.

- core client: if resuming a persistent file xfer, don't print error messages if file has wrong size/checksum/signature. That's a normal condition is a file xfer was interrupted.

- MGR: If the manager detects that the CC has crashed, and it was the one who started the CC to begin with, restart the CC after 30 seconds.

- MGR: Cleanup some of the taskbar interactions where both the tooltip and balloon were being displayed on Windows. Only one should be shown.

- MGR: Allow arguments to be passed to the CC from the manager's parameter list. /b "..." or /boincargs "..." should work.

- DIAG: Fix an uninitialized string issue that cropped up after removing the forground window information from the diagnostic output.

- MGR: take Close() function call out of #if/#endif block so the manager exits when the 'exit' menu item is selected.

---

Kathryn :o)

5.8.5

Howdy Folks,

Per David's instructions this release was built straight out of the HEAD branch for the purposes of tracking down what was formally a crashing bug, but is now just improper behavior in the CC.

Please create a cc_config.xml file with the following settings:

       <cc_config>
               <log_flags>
                       <task_debug>1</task_debug>
               </log_flags>
       </cc_config>

The message we are looking for is:
2007-01-23 21:08:03 [---] garbage_collect(); still have active task for acked result; state 0

When you see it could you stop BOINC and zip up your state files, log files, sched_request, and sched_reply files and email them to us?

Log any bugs you find to http://boinc.berkeley.edu/alpha/.

Thanks in advance.

----- Rom

---

Kathryn :o)

5.8.6

Howdy Folks,

It appears that the rash of crashes, -161s, state file errors, and new upload errors has subsided. If you still have upload errors with that read socket error message, abort them. We are still looking for any CC and manager issues that would block a public release.

We do not need any new bug reports on the grid based tabs, we'll be working over the weekend to clean-up the issues you all have raised.

The game plan as of right now is to beat the hell out of this build, come Monday we'll do another build which will contain the grid based tabs if we can get them into a polished state in time, if not the new release will contain the old tabs. We'll test the new build for a couple of days, if no showstoppers are found we'll release to the public.

So please give the client some 24 hours of testing and running looking for CC issues, log results to http://boinc.berkeley.edu/alpha.

I would like to reach 90% combined coverage for builds 5.8.6 and 5.8.7.

----- Rom

Change Log from last build:

- MGR: Disable advanced preferences dialog box.

- MGR: Do not refresh the Simple GUI when the manager is hidden.

- Manager: use a scientifically chosen set of colors for disk pie charts

- core client: there was logic that wouldn't start a result for a project if it had > 2 active uploads (intended to prevent an unbounded buildup of files for a project whose upload server is broken). It is was never triggered since the default is 2 active transfers per project. Maybe I meant persistent file transfers. In any case, this is a bad policy because there may be a project with lots of output files per result, possibly large, and there's no reason to keep it from computing. So instead, I changed it to not start a result from a project if its number of results in FILES_UPLOADING state exceeds 2*ncpus.

- core client and manager: reorganize message priorities. There are now three priorities:
MSG_USER_ERROR
Conditions that require user intervention. These messages are written for non-tech users, and tell the user exactly what to do.
GUI: red letters
MSG_INTERNAL_ERROR
Conditions that may indicate a problem or bug in BOINC itself, or with a BOINC project or account manager. The message is prepended with [error]".
GUI: black letters
MSG_INFO
All other messages.
GUI: black letters

Note: all messages that must be enabled by cc_config.xml
use MSG_INFO priority

- MGR: Flip the sorting icons for the Grid Control.

- MGR: Remove legacy tabs.

- MGR: Fix the OnFrameRender function in the Simple GUI, some of the logic in it blows out of the function with a return which was leaving the static variable that was guarding against repeated executions to true. This in turn kept the GUI from updating.

- MAC: Change MSG_ERROR to MSG_INTERNAL_ERROR in Mac-specific source file.

- MGR: Add a new menu item that'll shutdown the core client that the manager is connected too. It'll prompt for a new computer afterwards.

- core client: another attempt at fixing the problem where a project is issuing work but not uploading files.
New solution: don't fetch work from a project with more than 2*ncpus results in FILES_UPLOADING state.
Note: this doesn't address John McLeod's point that we shouldn't increase LTD for projects in this state.

---

Kathryn :o)

5.8.7

Howdy Folks,
Here is a release candidate. If no show stoppers are found with this release, this will be the first 5.8 release we release as a stable client.

The grid controls have been turned back off for this release.

Thanks for all the hard work on the 5.8.6 release, the level of feedback we got from the test results was great.

----- Rom

Change Log:

- core client: if account manager reply doesn't specify resource share for a project, revert to the share from project server

- MGR: fixed bug in dialog options where it doesn't set the slider size automatically. This is a bug in wxGTK so for Linux set the size explicitly.

- MGR: Update default BOINC skin.

- core client: fix account-manager resource share logic

- manager: Statistics tab fixes (from Mifistor)

- core client: make sure that RAM limits are not zero

- core client: on Unix, raise app stack size limit to max possible value before exec'ing the program.

- Manager: fix crash if select a project that doesn't have a name yet. There were two separate problems:
1) CMainDocument::project() was searching by name but not by master URL (BTW: all that at(), out-of-range exception stuff isn't needed, or the cut-and-pasted comment).
2) CViewProjects::UpdateSelection() (and its grid counterpart) were checking for project == NULL some of the time, but not all the time

- MGR: Give the user a change to abort a core client shutdown when they select 'shutdown current client'.

- MGR: Special case the authorization error dialog when it is attempting to use the default password, so that it also informs the user they need to run the manager from the same directory as the CC.

- MGR: Revert back to original tabs, grd controls are not going to make it for this release.

---

Kathryn :o)

5.8.8

Howdy Folks,

Pretty much the same as the last release without the advanced preferences dialog. This is the new release candidate.

----- Rom

BOINC 5.8 has now ready for public release.

Some of the highlights for this release include:
* Simple GUI (Complements of World Community Grid)
* Basic CPU identification
* CPU throttling
* New CPU scheduler
* New work fetch policy
* Improved memory management

We have one outstanding issue with the Linux client at this time. As soon as we have resolved it we'll release it as well.

We would like to thank everyone for their hard work in making this a solid release.

----- Rom

---

Kathryn :o)

5.8.9

Howdy Folks,

Here is a new build for the 5.8 line. It contains a few bug fixes. If all goes well this will become the new stable build in a few days.

----- Rom

Change Log:

- core client: computation of duration_correction_factor was messed up. E.g. when the correct value is .5, it converges to sqrt(.5)

- core client: tweak DCF update computation

- core client (Unix): tweak stack-size logic, and don't bump stack size limit beyond 64MB

- core client: fix bug where client thinks an app is running but it isn't.
Basic problem: there are two layers
1) process control: start/stop/suspend/resume, handle exits.
files: app_*.C
It owns ACTIVE_TASK::task_state
2) CPU scheduling: decide what should run
files: cpu_sched.C
It owns ACTIVE_TASK::scheduler_state.
These layers weren't cleanly separated; e.g. the CPU sched layer relied on the process control layer to set scheduler_state in some cases, and this wasn't always being done. I fixed this; the layers are now separated.
Also: added new task_state value, PROCESS_QUIT_PENDING

- core client: small tweak to DCF calculation

- core client: comment out stack-size stuff

- core client: print error messages if gethostbyname() fails for entries in remote_hosts.cfg

- Manager
changed "defect" to "stop using"
changed 2006 to 2007

- core client (Unix): removed code for SIGTSTP and SIGCONT. never got it to work, don't need anyway; handle SIGPWR.

- core client: print message if task finishes and output file is missin

---

Kathryn :o)

5.8.10

Howdy Folks,

This release is primarily geared towards tracking down Linux benchmark issues.

This core client was compiled with the '-O3 -fomit-frame-pointer -fforce-addr -ffast-math' compiler options using GCC 3.0.x.

DL link:
http://boinc.berkeley.edu/dl/boinc_5.8.10_i686-pc-linux-gnu.sh

Please report benchmark results to the alpha email list. Please remember to turn on the <benchmark_debug> flag before executing the core client.

----- Rom

---

Kathryn :o)

5.8.11

Howdy Folks,

Another build which should fix the CPU scheduling problem, and at least one issue with regards to running the screensaver while throttling your CPU.

----- Rom

Changelog:

- core client: when new version runs for the first time, do a network access to reference site. This will trigger personal firewall alerts, that otherwise might happen later when user is not there.

- core client: add messages to debug CPU benchmarks

- changed name of debug flag from <measurement_debug> to <benchmark_debug>

- SS: Remove reference to the Climate Change Experiment when the client is being built for WCG.

- *nix: Attempt to handle the various optimation flags better by default.

- core client: fix bug where screensaver would turn on an off if CPU throttling being used

- core client: fix bug that causes tasks to hang (not running, but BOINC thinks it is) on Windows

- MGR: Add simple validation for the authenticator returned by InternetGetCookie.

- core client: when a scheduler reply has a FILE_INFO for a file for which we already have a FILE_INFO record, overwrite the file_signature field only if the new record actually has one. This fixes a bug where a project used the same file in both an APP_VERSION and a WORKUNIT. The scheduler reply contained two <file_info> elements for the file; the first had a signature, the second didn't. The second one overwrote the first, and the app version was rejected because one its file wasn't signed.

---

Kathryn :o)

5.8.12

Howdy Folks,

This build has some bug fixes for the Windows Installer. It'll now forcefully terminate boinc.exe, boincmgr.exe, and boinc.scr if it detects they are running on the system. It should help reduce the number of setup related issues.

----- Rom

Change Log:

- WININSTALL: Force the termination of BOINC and BOINCMGR if we could not gracefully shut them down.

- MAC: If relaunching AppStats due to an error, close any open file descriptors (we communicate with AppStats via a bidirectional pipe.)

- WININSTALL: Force the termination of BOINC.SCR if we could not gracefully shut it down.

- Linux: Prevent crash by initializing the Taskbar pointer, even though the Linux Boincmgr currently doesn't support that feature.

---

Kathryn :o)

5.8.13

Howdy Folks,

This release should fix the problems the Win9X guys were having with installing the latest builds.

Please log the results of your tests to http://boinc.berkeley.edu/alpha.

Thanks in advance.

----- Rom


Change Log:

- core client: don't print misleading error messages when reference site access succeeds on initial startup.

- core client: a few more messages for <benchmark_debug>

- core client: fix bug where if a task is aborted (e.g. because it exceeds CPU limit) it's restarted on the next enforce_schedule().
The problem: we're deleting the ACTIVE_TASK, but the result is still in the ordered_scheduled_results list.
The solution: call request_schedule_cpus() in handle_finished_apps() when an ACTIVE_TASK is deleted.

- WININSTALL: Use the Unicode version of ShutdownBOINCManager on NT based platforms and use the ANSI version on 9X based platforms.

---

Kathryn :o)

5.8.14

Howdy Folks,

I think I got it this time.

----- Rom

Change log:

- WININSTALL: Forgot to make the ShutdownBOINC/ShutdownBOINCScreensaver custom actions NT only.

---

Kathryn :o)

5.8.15

Howdy Folks,

Here is another build to test, if all goes well this will become another build we release to public.

----- Rom

Change Log:

- core client: don't print "finish file not found" message when app quits because of preemption

- MAC: Eliminate use of AppStats helper application, which needed to be run setuid root. We weren't using the extra items it returns, so revert to older method of using popen(ps ...).

- MAC Installer: Add debugging statements to postinstall, postupgrade scripts

---

Kathryn :o)

5.8.16

Howdy Folks,

Here is a new build to test, hopefully the stuck upload/download problems are reduced. Please stress the upload/download functionality. See the change list for the other changes for the release.

As always, please log results to the alpha site.

----- Rom

Change List:

- core client, Unix: add '[Family x Model x Stepping x]' in p_model string. This will enable better homogeneous redundancy resolution

- core client: if benchmark time is in the future (due to user tweak) always run benchmarks

- core client: when resetting or detaching a project, if a file transfer was active for a project file, the file delete would fail because Curl still has the
file open. This would cause the manager to freeze for N*5 seconds.
Fix: in CLIENT_STATE::reset_project(), delete FILE_XFER and PERS_FILE_XFER objects.

- core client: the limits on numbers of simultaneous file transfers (total and per-project) now apply to uploads and downloads separately. We don't want a situation where an upload is blocked because there are N downloads.

- core client: if benchmark time is in the future (due to user tweak) always run benchmarks

- add <checkpoint_debug> message flag; shows when apps checkpoint.

- core client, Win: set processor affinity mask for benchmark threads

- core client: disable connection caching in Curl. This may reduce cases where file uploads hang because a server process holds file lock.

- manager: fix status text for file transfers

- core client: add bool detach_when_done to PROJECT. If set, detach from project when it has not results (i.e. when all work is finished and acknowledged). Save/restore in client state file.

- core client: add optional <detach_when_done> and <dont_request_more_work> flags to account element in account manager reply messages. See acct_mgt.php

- boinc_cmd: change --lookup_account and --create_account commands so that they poll until the operation is done, and then show a meaningful result (e.g. an account key). Remove --lookup_account_poll and --create_account_poll commands

- core client: when detach from a project, delete its sched_request, sched_reply, and master files

Win: Fix check mark on Snooze taskbar menu item on Windows XP

Mac: Fix OS 10.3.x permissions problem when running debug build; this problem caused shared memory communications to fail between client and project applications.

Mac: Fix serious bugs in Mac_SA_Insecure.sh and Mac_SA_Secure.sh scripts when run under OS 10.3.x. They sometimes created bad users and groups with IDs that were duplicates of other users and groups. They ran correctly under OS 10.4.x. Users who ran an older version of either script under OS 10.3.x should first run the current version of Mac_SA_Insecure.sh to delete the bad entries and then run Mac_SA_Secure.sh to create new good entries.

Mac: if we started core client, close it with rpc->quit() instead of kill().

---

Kathryn :o)

5.8.17

Version 5.8.17 of BOINC for Linux is now available for download.

The core client was compiled with GCC 4.1 which should improve the benchmarks.

Walt

I have posted BOINC version 5.8.17 for the Macintosh to the alpha
download page.

This has only a few minor changes from 5.8.16:
- Correctly report Physical RAM sizes > 2GB
- Improved backtrace for debugging support
- Fix a typo which caused the installer to work incorrectly on some
Macs, resulting in a -1010 error.

Please test this as soon as possible. If there no problems are
reported, we expect to release this to the public in a day or two.

Cheers,
--Charlie

---

Kathryn :o)

5.9.3

Howdy Gang,

We are releasing a new test version of the BOINC software.

Here are just a few of the changes that made it into the 5.10 release cycles:
1. Sort-able views in the Advanced GUI.
2. Progress bars in the Advanced GUI.
3. Advanced preferences dialog.
4. Improved Vista compatibility.

I still have to go through the check-in notes to compile a complete list of changes.

----- Rom

---

Kathryn :o)

The latest changes to the 5.9 version can be found at http://boinc.berkeley.edu/trac/roadmap, also showing which bugs (tickets) are still open and which have been closed.

---