Contenu

9   Historique des versions
2.3.1—May 16, 2005
• Works with Apple Mail on Mac OS X 10.4.1. SpamSieve should automatically install a new copy of its Apple Mail plug-in the first time it’s launched, but if it doesn’t you can do so by choosing Install Apple Mail Plug-In from the SpamSieve menu.
• Various accuracy improvements.
• Renamed the Apple Mail
training commands to Train as Good and Train as Spam.
• Train as Spam is much faster in Apple Mail on 10.4 when there are multiple messages selected.
• If
SpamSieve’s Dock icon is hidden, a new SpamSieve - Open Window command will appear in Apple Mail’s Message menu. This lets you quit SpamSieve or access its windows and settings while its menu bar is hidden.
• Worked around Apple Mail bug so that Train as Good now moves spam messages out of Mail’s Spam folder on 10.4, provided that
SpamSieve had put them there.
• Reduced the maximum size of the Growl
notification bubbles.
• Better at parsing malformed messages.
• The HTML parser is much faster with certain pathological spam messages that could previously drag parsing out for a minute or more.
• Worked around
Entourage bug that could lead to the creation of multiple Uncertain Junk categories.
SpamSieve no longer complains about the permissions on Apple Mail’s Bundles folder if you’re using a different mail program.
• Removed the Prune Corpus command. With auto-training being much smarter than in earlier versions, it’s almost never advisable to prune, and improper pruning severely reduces accuracy.
• The demo reminder window updates the number of days remaining if you leave
SpamSieve running for days or weeks at a time.
• Fixed bug that could cause a crash when using Instant
Purchase.
• Worked around crash caused by bug in 10.2.
• Updated localizations.
2.3—April 25, 2005
• General
• Works with Mac OS X 10.4 (Tiger).
• Made lots of improvements to
SpamSieve’s parsers and tokenizer for better accuracy. To fully take advantage of this, you will need to reset SpamSieve’s corpus and re-train it (e.g. with 300 recent good messages and 600 recent spams). However, this is certainly not required, and I expect that most people will opt for the simpler upgrade of just installing the new SpamSieve application.
• Auto-training works better.
SpamSieve is smarter about selecting which incoming messages to train itself with, so that it reaches a high level of accuracy sooner, and it is better at adapting to new types of mail that you receive.
• Improved the
training tips and the training section of the manual so that it’s clearer what the best practices are.
• When
training the whitelist, SpamSieve will now create rules that match the addresses of the other recipients. This should reduce false positives from people who haven’t previously sent you mail, but who correspond with the same people you do.
• Added Update Address Book “Me” Card command.
• Scriptability
• Added selection AppleScript property, which can be used to get the selected token infos in the corpus or the selected rules in the whitelist or blocklist window. One use of this is demonstrated in a script that adds the addresses of the selected whitelist rules to the Address Book.
• Added
AppleScript properties and commands to access SpamSieve’s windows and log. One use of this is demonstrated in a script that lets you open SpamSieve’s windows if its Dock icon is hidden (and thus the menubar is inaccessible).
Notification
• Added option to control whether Growl notifications contain exerpts of the message body.
• The new message count in
SpamSieve’s Dock icon is now auto-positioned.
• Fixed regression where dragging the slider to adjust the size of the number in the Dock icon didn’t temporarily show the number so that you could preview the results.
• Apple Mail Integration
• Improved accuracy due to better decoding of Apple Mail messages.
• The commands for
training SpamSieve from Apple Mail are now in Mail’s Message menu, rather than in the Scripts menu. This lets you access those commands in the same way on 10.2 through 10.4, and you don’t have to worry about changing the scripts around if you boot into a different version of the OS.
• The Apple Mail plug-in is faster at processing messages.
• If the Apple Mail plug-in is installed but disabled for some reason (e.g. a location switch or OS re-install),
SpamSieve will automatically re-enable it (takes effect when you relaunch Mail).
• Worked around bug in Mail that could prevent Train Good or Train Spam from working properly when
training multiple messages at once that were already in their final destination.
• Eudora Integration
• The installer no longer asks you to find Eudora except when absolutely necessary.
• The Eudora plug-in is now installed in the Application Support folder rather than inside the Eudora application package. Thus, it’s no longer necessary to re-install the plug-in when updating Eudora, although you still may need to disable the
SpamWatch and SpamHeaders plug-ins if updating Eudora re-enables them.
• Added a Cancel button to the alert that you get if Eudora is already running when you ask
SpamSieve to install the Eudora plug-in, in case you don’t want to quit Eudora at that time.
• Fixes and Tweaks
• The Apple Help is now multiple linked pages, rather than a single page, so you can now use Help Viewer’s search box.
• Simplified the setup instructions in the manual, and added new sections on customization.
• Improved duplicate message detection.
• Fixed bug where e-mail addresses were not always correctly parsed out of mail headers.
• Tweaked the format of Trained entries in the log.
• Worked around OS bug that could cause
SpamSieve to freeze when installing AppleScripts.
SpamSieve is better at following aliases and more tolerant of incorrect permissions when looking for its support files.
• Updated to eSellerate Engine 3.6.1.
• Improved serial number name entry.
• Various localization fixes.
• The ? buttons in the
Preferences window now open the specific help sections on Jaguar.
• Fixed longstanding unreported bug where you could launch multiple instances of
SpamSieve if you were using Jaguar.
2.2.4—February 10, 2005
• Increased the speed of SpamSieve’s message processing.
• Processing messages with
Entourage is much faster. To realize the speed increase, re-install SpamSieve’s Entourage scripts and let it replace the existing ones.
• Improved accuracy through better HTML and header processing. (It is not necessary to reset the corpus.)
• Colors to indicate spamminess in Apple Mail are now enabled by
default.
• Can now play sounds (and bounce the Dock icon) when mail arrives, even if
SpamSieve or a mail program is frontmost.
• Added option to
Entourage’s Change Settings to make all spam messages marked with category Junk instead of marking some of them with Uncertain Junk.
• Added more standard blocklist rules for non-Latin character sets.
AppleScript errors complaining about not being able to find the SpamSieve application should be much less common now. If you do see such an error, asking SpamSieve to re-install the scripts for Apple Mail or Entourage (and letting it replace the existing scripts) should cure things.
• Improved name and serial number entry.
• The Edit Log command is no longer an alternate menu item.
• Can now load
Entourage addresses when Launch Services can’t find Entourage.
• The
default classifier reports better scores.
• The behavior of the
default classifier (if no rules or other classifiers match, and the Bayesian classifier is disabled) may now be changed by setting the DefaultIsGood default. Normally this is YES, but you can set it to NO to make SpamSieve treat all non-whitelisted messages as spam.
• Pruning now is recorded in the log.
• Updated the Russian localization.
• Added Swedish and Vietnamese localizations.
• Expanded and improved the manual.
• The
Entourage scripts no longer mark messages with multiple copies of the same category.
• Fixed bug where
SpamSieve would sometimes incorrectly think that it had made a mistake (if auto-training was on and the user had asked it to reclassify a message).
• Fixed bug in the Change Settings scripts.
• No longer reports an error when trying to reset an empty initial corpus.
• Fixed crash when quitting after resetting an empty corpus.
• Fixed bug where
SpamSieve would crash when reading a damaged corpus file.
• Worked around OS bug that could cause crashes when creating the Uncertain Junk category in
Entourage.
Notification dismissal now works on Tiger.
• Apple Mail’s scripts are now installed in the correct location when running on Tiger.
2.2.3—November 15, 2004
• Fixed bug where SpamSieve would complain of a permissions error at startup.
2.2.2—November 14, 2004
• Can Notify with Growl when good messages arrive, and to help spot false positives without looking through the entire contents of the spam folder.
• Added Change Settings commands for Apple Mail and
Entourage, which let you configure script options (spam folder name, behavior when using Train Good and Train Spam, etc.) without having to edit the scripts. The settings are stored in SpamSieve’s preferences file, so they will persist if you update or re-install the scripts.
• Fixed bug where
SpamSieve would report a syntax error when processing certain messages (typically in non-ASCII character sets).
• Better accuracy due to improved header and body analysis.
• Better accuracy due to smarter auto-creation of From (name) rules.
• Auto-training now takes scores into account; the corpus is updated using messages that were harder to classify, to forestall errors.
• Better handling of spam messages that lie about their encoding.
Entourage now uses two separate categories for spam messages, so that you can see which ones are more (or less) spammy.
• Improved accuracy when processing messages in German.
• Any Address rules now match
SendTo addresses.
• Better handling of messages with “From ” lines.
• Worked around Mail performance problem where Add Good would hang if there were a lot of recipients.
• The Dock (and
DragThing) good message counter no longer increases for messages received while the mail program is active.
• Added Russian localization.
• Improved the French localization.
• Better recovery from damaged Rules and History.db files.
• Added Italian-localized installation instructions.
• Improved launch time.
• Better parsing of PGP messages.
• Worked around 10.2 bug that could cause crashes when the
Training Tip window was updated.
• Fixed bad interaction between auto-training and duplicates when the corpus was small.
• More tolerant of incorrect file and folder permissions.
• If Launch Services cannot find the Eudora application, the installer will now try some heuristics and prompt the user to locate Eudora, rather than giving up.
• The Apple Mail and Eudora plug-ins are better at finding and launching the
SpamSieve application, and they will notify the user if they are unable to do this.
• Updated to eSellerate 3.5.9, which provides support for registration names using non-ASCII characters.
• When you type in your serial number,
SpamSieve normalizes its spacing and case.
• The Add Good and Add Spam
AppleScripts have been renamed Train Good and Train Spam. When installing the new scripts, SpamSieve will automatically move the old ones to the trash.
• The
Entourage spam folder is now called Junk E-mail instead of Spam, since Entourage 2004 already has a so-named folder.
2.2.1—September 20, 2004
• Shows count of new good messages in DragThing 5.3 and later, if you install an integration script. (This script will be built into future versions of DragThing.)
• Fixed accuracy regressions from 2.1.4 and made some accuracy improvements.
• Added Italian localization.
• Auto-training will no longer add duplicate messages to the corpus, because they interfere with undo.
• You can now start a new paragraph in the crash reporter by pressing Return rather than
Command-Return.
• Tries to add itself to the Launch Services database at launch, to make it easier for the scripts and plug-ins to find the
SpamSieve application.
• Fixed bug where the help buttons in the
Preferences window didn’t work unless the main help had previously been opened.
• Improved the French and Japanese localizations.
• The Edit Log command will now open the log using an editor, rather than Console, if BBEdit is unavailable.
• Improved accuracy statistics when using Apple Mail IMAP accounts.
• The mailbox parser is faster and more tolerant of malformed mbox files.
• No longer tries to roll over the log more than once per day.
• Fixed regression where adding to the whitelist or blocklist didn’t work if a disabled rule matched the message.
• Fixed crash that could happen when showing progress bar at launch.
• Fixed crash when scanning System 7–format sounds.
• Fixed bug where slightly spammy Apple Mail messages would be colored and recorded as spam, but not moved to the Spam mailbox.
2.2—August 24, 2004
• Accuracy Improvements
SpamSieve is smarter about what text to tokenize, it extracts more information from messages, it’s better at detecting invisible text, and it can undo more spammer obfuscations.
• The Bayesian classifier can leverage the results of
SpamAssassin’s heuristic tests. It also assigns better word probabilities and is better at deciding which parts of the message are important.
• Messages containing Habeas headers are now checked against the Habeas Whitelist. This protects against spam messages that include Habeas headers in order to get through
filters.
• Removed the Import Seed Spam command, as it would reduce accuracy with recent versions of
SpamSieve.
• General Improvements
• Auto-training is more automatic. There is now a single checkbox to enable auto-training and, if it’s on, SpamSieve will try to do the right thing when it processes new messages. That is, it will train itself using messages of the types that it needs to see more of, and it will prevent the corpus from growing unbalanced or overly large. Also, it will update the whitelist using every incoming good message, so that you can have a complete whitelist without bloating the corpus.
• The rules and corpus windows now have toolbars, and they support iTunes-style filter-searching.
• The corpus and rules are now scriptable, and
SpamSieve now supports AppleScriptKit terminology.
• Assorted performance enhancements make
SpamSieve faster and improve the responsiveness of its user interface.
• Added Portuguese localization.
• Blocklist and Whitelist
• Rules support more match fields: From (name), Any Recipient, Any Address, Any Character Set, and Any Attachment Name.
• The new Import Addresses… command lets you import blocklist and whitelist rules from text files (e.g. address book exports or mbox files).
• Added
default blocklist rules for .pif and .scr attachments and a default whitelist rule for lists.c-command.com.
• Rules can now match the empty string, e.g. when there is a subject header, but the subject is blank. They can also match absent headers by matching against, e.g. <
SpamSieve-Unknown-Subject>.
• The Text to Match field now abbreviates using an ellipsis if the text won’t all fit.
Notification
• Can make a Griffin PowerMate flash when new good messages arrive.
• Now supports System 7–format
notification sounds as well as Entourage sound sets. In addition to the Library/Sounds folders, SpamSieve will now look for sounds installed with Apple Mail, Entourage, Mailsmith, and PowerMail.
• Doesn’t play the
notification sound if you are using fast user switching and SpamSieve is running in one of the background sessions.
• Notifications that require dismissal (continuous bouncing,
PowerMate flashing) are no longer triggered when SpamSieve or the mail program is already frontmost. Thus, you don’t have to deactivate the mail program and then activate it again to dismiss the notification.
• Apple Mail Integration
• Sets the colors of spam messages to reflect how spammy they are, making it easier to skim the Spam mailbox for false positives.
• Protects against Web bugs by telling Mail not to load the images from messages that are classified as spam.
• The Add Good script is better at moving messages from the Spam folder back to the correct inbox, and it will mark them as unread when it does so.
• The Add Good and Add Spam now work with Mac OS X 10.3.5.
• The plug-in will look for
SpamSieve in the Applications folder, even if Launch Services is confused and says it can’t find it.
Entourage Integration
• The Exclude my addresses switch now also applies to the Entourage address book, to reduce the chances of a forged spam getting through.
• Added new scripting verb add
entourage addresses. This will allow you to keep the Entourage addresses in sync by using an Entourage shutdown schedule.
• Can automatically add the recipients of the messages you send to the whitelist.
• Sanity-checks addresses that are loaded from
Entourage. For instance, if both the name and the address were entered in Entourage’s address field, SpamSieve will now extract just the address, rather than taking Entourage’s word for it.
• The Add Good script now works with the Possible Spam folder and marks false positives as unread.
• Messages can have more than one category, so
SpamSieve now adds and removes the Junk category from the message, rather than replacing the category. This means that marking a message as junk doesn’t wipe out other categories on the message.
• Shows a progress bar while importing
Entourage addresses.
• Eudora Integration
• Removed the option to use the full Eudora junk score range; this setting is now always in effect.
Mailsmith Integration
• Can automatically add the recipients of the messages you send to the whitelist.
PowerMail Integration
• Can automatically add the recipients of the messages you send to the whitelist.
• Statistics and Log
• The Statistics window shows the number of blocklist and whitelist rules and the number of spam messages received per day. The date sheet has a Now button that enters the current date and time.
• The log is now stored in ~/Library/Logs/
SpamSieve. The Open Log command now opens the log in Console, so that it’s easier to monitor as it changes. To get the old behavior of opening the log in a text editor, hold down Option and choose Edit Log.
• If the log file grows larger than 5 MB,
SpamSieve compresses it, archives it by date, and starts a new log.
• When
SpamSieve makes a mistake, the log records which classifier made the error and what the message’s score was. The log also now records parse errors and the number of messages in the corpus.
• Fixes and Tweaks
SpamSieve can optionally show an alternate, more photo-realistic, icon in the Dock.
• There is a new mailing list for discussion of
SpamSieve.
• Each tab of the
Preferences window now contains a help button that will open the corresponding section of the Apple help.
• Updated to PCRE 4.5 and SQLite 2.8.15.
• Fixed bug where
SpamSieve could crash when generating a messages’s identifier if a system routine failed.
• Fixed bug in EDMessage that could cause crashes when decoding certain quoted-printable messages.
• Better handling of 8-bit subjects with no specified encoding.
• Worked around Panther bug that could cause crashes or drawing artifacts on the desktop by no longer trying to draw on the Dock icon when it’s hidden.
• Worked around OS bug that prevented certain
notification sounds from playing.
• Fixed bug where the selection was not always preserved when editing blocklist and whitelist rules.
• The Show Corpus, Show Statistics, and Preferences… commands in the Dock menu now bring
SpamSieve to the front.
• The progress window when exporting the corpus sometimes lagged a bit before closing automatically.
• Fixed bug where
SpamSieve would get confused if you entered the empty string as a date.
2.1.4—April 15, 2004
• Fixed bug where SpamSieve could crash when installing the Eudora plug-in if you were using Eudora 6.0.x.
• Fixed regression where
SpamSieve couldn’t process certain messages, resulting in a script error and reduced accuracy.
• Added French localization.
• Updated AOL instructions, since AOL now supports IMAP.
2.1.3—April 13, 2004
• Added menu commands for installing the Apple Mail plug-in and scripts, the Eudora plug-in, and the Entourage scripts. These items are now stored inside the SpamSieve application package.
• Scripts for the other applications are now stored inside the application bundle, not at the root of the disk image. The Show Other Scripts menu command will reveal them in the Finder.
• Apple Mail and Eudora users should update their plug-ins, using the commands in the
SpamSieve menu.
• Added the score script command, which returns an integer between 0 and 100 indicating how spammy the message is. 50 and higher mean spam.
• The Predicted lines in
SpamSieve’s log now show the scores of the messages.
• Can now use Eudora’s full 0-to-100 junk score range if you check the appropriate box in the
Advanced preferences.
• Improved parsing of messages with 8-bit transfer data.
• Faster at processing messages.
• Added support for
Outlook Express 5.
• Worked around OS bug that could cause
SpamSieve to come to the front each time a message was processed in Apple Mail or Eudora (usually if an X11 application was frontmost).
• Fixed bug where errors encountered while processing messages were not reported in the log.
• Worked around Cocoa problem where certain
notification sounds wouldn’t play.
• Made the
Purchase window easier to understand, and added a button for looking up lost serial numbers.
• Trims the text in the serial number field so people don’t accidentally paste the number in twice.
• Software updater is better at checking whether the computer can connect to the Internet.
SpamSieve now tries to parse Eudora messages according to RFC822, even though this will sometimes fail, as many Eudora messages are not RFC822-compliant.
• Adjusted the list of headers that
SpamSieve ignores.
• Added keyboard shortcuts for Apple Mail scripts.
• Improved the
training tips.
• Updated to SQLite 2.8.13.
• Updated to eSellerate SDK 3.5.5.
• The Send Report button in the crash reporter is no longer a
default button, so there’s no longer confusion about entering returns in the comment field.
• No longer prints fragments of spam messages to the console when it gets confused.
• Replaced the copy of the manual outside the app with a read-me.
2.1.2—January 26, 2004
SpamSieve can now move Apple Mail POP messages to the Spam folder. Thus, it now fully supports Apple Mail on Jaguar and Panther.
• Honor Habeas headers is now off by
default.
• Fixed regression where blocklist and whitelist rules got deselected after editing their text.
• When loading addresses from
Entourage, SpamSieve now picks up addresses that are not associated with any contact (that is, they appear only in a group).
• The
default date shown in the Statistics window is now the date that SpamSieve was first launched, rather than September 2002.
• The Apple Mail Add Good script is better at finding the proper inbox when moving false positives out of the Spam folder.
• The Apple Mail Mark If Spam script can mark the spam messages as read.
• The
Purchase window now makes it more clear when a serial number has been accepted.
• In the Statistics window, Set… is now Set Date… and Copy is now Copy Stats.
• Fixed crash that could happen when processing messages in Japanese encodings.
• Added Japanese localization.
2.1.1—January 8, 2004
• Much faster at processing messages when there are many blocklist and whitelist rules. Also improved the speed of loading, deleting, and sorting rules.
• Improved accuracy tracking with the Panther version of Apple Mail; previously,
SpamSieve couldn’t always tell when it was being corrected.
• Catches more spam because it knows about more spammer obfuscation tricks and also which headers it should ignore.
• Fixed bug (introduced in 2.0) where the Bayesian engine didn’t work if Mac OS X’s
default language was set to Japanese.
• The
SpamSieve Eudora Plug-In is better at launching the SpamSieve application if it is not already running.
• Loading
Entourage addresses now adds to the addresses that were previously loaded, rather than replacing them. This makes it possible for Entourage users who have more than one Entourage identity to give SpamSieve the addresses from all their address books (by loading once for each identity). Hold down Option when clicking Load to get the old behavior of replacing the previously loaded addresses.
• The sound pop-up menu in the
Preferences window now immediately notices when new sounds are installed; previously, it would only check when updating the rest of the preferences window.
• You can now add a rule without a the Blocklist or Whitelist window being frontmost.
SpamSieve will ask which type of rule to add.
• Regex rules can now start with an options modifier such as (?-i).
• Copying rules to the clipboard now just copies the text to match (typically an e-mail address), not all the columns. To get all the columns, you can print to PDF.
• The
Entourage Add Good script now finds localized inboxes, rather than creating a folder called Inbox.
• The
Entourage Add Spam script can now remove spam messages from the server.
• The Statistics window now shows percentages instead of ratios.
SpamSieve will now quit at launch if another copy of the application is already running.
• Re-targeted broken Habeas URL.
• Added the following menu commands: Close All Windows, Minimize All Windows, and Zoom.
2.1—December 9, 2003
• Added support for Apple Mail POP accounts. POP messages can be marked as junk and colored, but (due to limitations in the present version of Apple Mail) they cannot be moved to another mailbox.
• Added a
Training Tip window that gives advice on how to improve SpamSieve’s accuracy, based on the current state of the corpus and preferences.
• Rules in the whitelist and blocklist are no longer limited to just matching sender addresses. They can now match a variety of message fields (To, CC, Subject, etc.), as well as the message body. In addition to exact matches, rules now support the following match styles: contains, starts with, ends with (useful for matching domains), and Perl-compatible regular expressions. You can now edit rules and add new rules manually (as opposed to automatically, as a result of
training SpamSieve with a message).
• When trained with a good message from a mailing list,
SpamSieve will automatically create a whitelist rule based on a mailing list header, if present.
SpamSieve can now read in the Entourage address book and use it as a whitelist. Thus, the Entourage rule can now give SpamSieve all the messages, not just the ones that were from unknown senders. This means that SpamSieve can now accurately notify the user when non-spam messages are received. Also, the statistics it keeps will be more complete.
• Improved the accuracy of the Bayesian classifier when the corpus is unbalanced.
• Made a variety of low-level changes to improve
SpamSieve’s accuracy, for instance: adjusted the list of headers that are analyzed and how words are tokenized.
• The Apple Mail Add Spam script now has an option to control whether the messages are moved to the Spam folder.
• The
Entourage Add Good script now moves messages to the inbox if they’re located in the Spam folder.
Mailsmith users can now auto-train using only spam or good messages by turning off training in Mailsmith and turning on one of the auto-train checkboxes in SpamSieve.
• Improved the importing of mbox files that do not have blank lines between the messages, such as some Eudora mailboxes. Fixed a bug where the the mbox parser could crash if a message had length zero. Also,
SpamSieve now shows a progress bar while counting the number of messages that will be imported.
• Improved the corpus and rule list displays. You can now enter and leave editing mode by typing Return. Type-ahead works better; for instance, if you type “g” and there are no rows that start with “g,” it will look for one that starts with “f.” When you delete a word or rule, you can cancel out of the confirmation sheet by typing Escape. To avoid the confirmation sheet entirely, you can delete using
Command-Delete instead of Delete. When a word or rule is deleted, SpamSieve selects a nearby rule so that you don’t lose your place. When deleting many words at once, SpamSieve no longer shows a progress window for deletions that will not take very long.
• Entering the name and serial number to personalize
SpamSieve is now more foolproof: SpamSieve strips leading and trailing whitespace, and it detects when you enter a coupon code in the serial number field. Fixed regression where SpamSieve rejected names containing non-ASCII characters. In addition, there’s a new button for quickly redeeming coupons.
• Updated to the latest eSellerate SDK so that
purchasing SpamSieve from within the application is faster.
SpamSieve now requires Mac OS X 10.2.6 or later.
• Fixed bug where dates entered in the Statistics window were sometimes parsed in GMT instead of the local time zone, thus causing the date to be off by a few hours.
• Improved the reliability and user interface of the crash reporter.
• No longer crashes when parsing certain non-RFC822-compliant Eudora messages.
2.0.2—October 1, 2003
• Now works with Apple Mail (IMAP and .Mac only, not POP).
• The message count in the Dock icon now resets when an e-mail client becomes active (rather than just when
SpamSieve became active). You can also control the size and position of the number in the Dock icon.
• Much faster at deleting lots of rules at once.
• Replaced the message store database with custom code that’s faster and more reliable.
• Improved accuracy for HTML messages containing links.
• Importing mbox files is faster.
• Fixed bug where you couldn’t use Web registration after the demo period had expired.
• The log records which addresses matched the whitelist or blocklist.
• The log records corpus imports.
• Auto-training is faster.
• The
Entourage Add Spam script can close the frontmost window if it’s spam.
• Fixed bug where the date in the Statistics window could get cut off if you changed it to use a more verbose format.
• Worked around OS bug that caused dates like “01.09.2003” to be interpreted as January 9 in German-style locales.
• Added Copy button to the Statistics window.
• Fixed problem updating certain history databases from 1.3.1.
• More resilient to minor corpus file corruption.
• Fixed crash that could happen with improperly formed multi-part messages.
• Shows the number of blocklist or whitelist rules in the title bar.
• The whitelist now contains some c-command.com addresses by
default.
• Fixed crash when opening the Statistics window while adding messages.
• The Statistics window shows ratios, where applicable.
• Assorted minor performance improvements.
• The modification dates of the
AppleScripts are now the actual modification dates, not the date the distribution was built.
2.0.1—September 17, 2003
• Replaced the database engine that was being used to store the corpus with some custom code. This should be much faster and more reliable.
• Loading and saving the rules is faster, due to a better file format.
• The rules and corpus message counts are now saved to disk during idle time rather than when quitting. This should prevent data loss in the event that
SpamSieve doesn’t quit normally.
• The whitelist and blocklist are more memory-efficient.
• Plugged memory leak in EDMessage.
• Fixed crash involving certain really long header lines.
• Fixed bug where the Whitelist and Blocklist windows weren’t always up to date.
• Table views are smarter about not scrolling unnecessarily to maintain their selections.
• The Whitelist and Blocklist windows now secondary sort by sender.
• Changes to the
preferences are saved to disk immediately.
• Fixed bug where tables saved their sorted columns but didn’t restore them.
• When
SpamSieve gets a fatal error, it now quits like it says it will.
2.0—September 10, 2003
SpamSieve now extracts a lot more information from each message. This makes it much more accurate and also makes it learn faster.
• Now integrates with
Eudora 6 (Sponsored or Paid) via a plug-in. It can now process every incoming Eudora message and can be trained using the Junk and Not Junk commands in Eudora’s Message menu.
SpamSieve now has a blocklist and a whitelist. These are automatically maintained based on the senders of messages that SpamSieve is trained with. The blocklist makes sure that all messages from known spammers are caught and speeds processing for these messages. The whitelist lets you be sure that certain messages will never be marked as spam; this was possible before, but now you don’t have to clutter your address book with addresses from online retailers, etc.
• You can now control how conservative or aggressive
SpamSieve is at catching spam.
SpamSieve can now play a sound or bounce its Dock icon after a batch of non-spam messages has arrived. This is meant to replace your e-mail client’s new mail notification, which you don’t want going off if all the new messages are spam.
• Shows the number of new good messages in the Dock icon.
• Now parses HTML so that it can better extract relevant information from HTML messages, and also handle various HTML-based tricks that spammers use to fool
filters.
• New method of calculating word probabilities makes
SpamSieve better at discerning which words in the message are important.
• Includes a corpus of seed spam, to jump-start spam recognition for users who do not have many saved spam messages.
• The corpus is now stored in databases rather than in a property list. This makes it launch faster and use much less memory, as the corpus doesn’t have to be all in RAM at the same time.
• The statistics file format (for History.db) has changed in order to enable performance improvements and more statistical displays in future versions.
• Handles more types of plain text obfuscations, and is much faster at undoing them.
• Added option for the address book whitelist to only use other people’s addresses, so that spam messages from your own address don’t match the whitelist.
• Can mark all messages with Habeas headers as good.
• Can mark all messages with some variant of “ADV” at the start of the subject as spam.
• Can mark all base64-encoded HTML messages as spam.
• New probability combiner increases accuracy.
• Uses stop words to speed processing and reduce false negatives.
• When filtering a message, considers the number of occurrences of the words, not just which words are present.
• Can import messages from mbox files.
• Can import the corpus from and export it to an XML property list (the same format used by 1.x).
SpamSieve can now check for updated versions of itself.
• Added crash reporter.
• Added Dock menu containing frequently used commands.
• The entries in the log are more detailed.
• The corpus now stores the date at which each word was last accessed.
• Fixed bug where storing statistics would fail on systems that didn’t know about GMT.
• Fixed bug where
SpamSieve could throw away long runs of HTML thinking they were attachments.
• Added button for opening the Mac OS X Address Book from inside
SpamSieve.
• The Statistics window now has a contextual menu item for copying the displayed information.
SpamSieve no longer wastes cycles updating the Statistics window after it’s been closed.
• The Statistics window is smarter about updating only the portions that could have changed.
• No longer shows Good Words and Spam Words stats.
• Logging has less overhead.
• Updates the history asynchronously, resulting in faster message processing.
• Checks for mistakes in a background thread.
• False negatives are now written to disk in a background thread.
• Re-arranged the Corpus window.
• Pruning the corpus now works by access date rather than by word counts. Of course, you can manually prune the old way by sorting the Corpus window by Total.
• Updated to SQLite 2.8.6 and tuned it for speed.
• Updated to PCRE 4.3.
• Updated to eSellerate 3.5, which should fix crashes some people saw after registering on 10.2.6.
• Now looks at headers of subparts of messages from
Mailsmith.
• Time-consuming operations now either have a progress bar or a progress spinner.
• Better at extracting malformed e-mail addresses from headers.
• Copying rows from the Corpus window to the clipboard now uses the order of the columns in the window rather than the
default column order.
• Fixed regression where the
Entourage scripts no longer created the Spam folder if it didn’t exist.
• Fixed potential crash with regex replacements at the end of a string.
• History.db and the corpus can now be aliases.
• Automatically trims carriage returns and other illegal characters when you paste in your name and serial number.
• Now saves the name and serial number to disk as soon as they’re entered.
• The Spam folder in
Entourage no longer has to be top-level.
Entourage can mark good messages as unread.
• Type-selecting in table views is quicker.
• No longer nags constantly when unregistered.
• Fixed bug where it could look as though
SpamSieve had hung if it started up in the background with an empty corpus.
1.3.1—June 18, 2003
• Added direct integration with Mailsmith 2.0 and later. Enabling SpamSieve is as easy as clicking a checkbox. You can train SpamSieve directly from Mailsmith’s Message menu. Bare Bones Software has seamlessly integrated it with Mailsmith’s powerful filtering system, and Mailsmith knows not to bounce its Dock icon after receiving a batch of messages that are all spam.
• Fixed crashing bug triggered by incorrectly encoded headers.
• Regex substitutions are faster and much more memory efficient.
• When adding spam messages to the corpus, the
default is now for SpamSieve to move them to the Spam folder.
• The
PowerMail Move If Spam script now changes the color of spam messages.
• The
Emailer scripts now pass text and HTML attachments on to SpamSieve for analysis.
• Added instructions for using the
Entourage and PowerMail address books as whitelists.
• Compacted the ED frameworks to reduce application size and memory use.
• Disabled SQLite’s file locking so that
SpamSieve’s data folder can now be located on an AppleShare volume.
• Caches the Address Book to speed whitelist lookups 100 fold.
• The statistics database is faster due to an updated version of SQLite.
• Fixed bad pointer in header extractor that caused unnecessary fallbacks to the secondary parser (reducing accuracy).
• Added experimental support for moving
Entourage messages on IMAP accounts to the (local) Spam folder. This can be enabled by editing the scripts with Script Editor.
• For clarity, the names of
AppleScripts that ask SpamSieve to predict the category of a message now contain the word “if.”
• Added lots of minor clarifications to the documentation.
1.3—February 11, 2003
• More resilient to spammers’ tricks for obfuscating words.
• Can use e-mail addresses in the system Address Book as a whitelist. Messages sent from those addresses will never be marked as spam.
• Greatly reduced overall memory usage as well as launch and quit times.
• Can save false negatives to disk for later reporting to
SpamSieve’s developer.
• You can edit the spam and good counts associated with a word, remove selected words from the corpus, and reset the corpus entirely.
• Type-ahead navigation in the Corpus window. Type the first few letters of a word or number to select it (and scroll to it).
• You can hide statistics from before a set date, to better see the current accuracy and spam reception rate.
• Improvements to the Corpus window: Shows all words rather than only those considered statistically significant. Re-sorting by numeric columns is twice as fast. You can copy the selected rows to the clipboard or drag them to another application. The selection is preserved when you change the sort column, you can sort in descending order, and the sorted column is remembered between launches. The Home and End keys work.
• The Prune Corpus command now tells you how many words it would remove and asks for confirmation.
• The statistics tracking is smarter about handling duplicate messages.
• The statistics have tooltips explaining what they mean, and you can copy all of the statistics to the clipboard at once.
• Improved accuracy tracking of
PowerMail and Emailer messages.
• Eudora
Integration: Can mark spam messages as read and/or mark them for removal from the server.
• Expanded the
AppleScript dictionary, to enable better integration with mail and news clients.
Entourage Integration: Creates Junk category if there isn’t one, and can mark spam messages as read.
Mailsmith Integration: The adding scripts now set the appropriate message properties.
• Better parsing of messages with illegal characters in the headers.
SpamSieve’s Info.plist file contains an LSUIElement entry. Change the 0 to a 1 to hide the application’s Dock icon. (You’ll need to change it back to access the preferences.)
• The message count display has moved from the Corpus window to the Statistics window.
• Better error message when the corpus couldn’t be saved.
• Added tooltips to
preferences.
• The registration window gives better feedback when you personalize.
• Better recovery from errors in the corpus file.
• The secondary parser is better at handling DOS linebreaks.
1.2.2—November 20, 2002
• Fixed bug in the PowerMail Add Good script.
• Added uninstaller for Eudora users.
• Better handling of errors while adding messages to the corpus.
• Removed bloat from the
Entourage Mark Spam script.
• The application icon now has an alpha channel, so it doesn’t appear with a white halo when viewed on a colored background.
• Minor changes to the manual.
1.2.1—November 18, 2002
• Modified Info.plist to work around a bug in Mac OS X 10.1 that could cause the Finder to crash when launching SpamSieve.
1.2—November 18, 2002
• Added support for Emailer 2.0v3 and Eudora (5.2 and later).
• Decodes base64 and quoted-printable text parts, thus finding words that spammers try to hide from anti-spam software.
• Decodes subjects that use different character sets (e.g. big5).
• Adds special tokens for MIME entities such as part boundaries and uninterpretable message parts.
• Keeps track of the messages added to the corpus, and can optionally prevent you from adding the same message more than once (biasing the counts). Thus, you no longer have to remember which messages you’ve already added.
• You can now “undo” adds to the corpus, e.g. if you added a message as good when you meant to add it as spam.
• Can now add messages to the corpus as they are filtered, so after the initial
training you only have to add messages when SpamSieve makes a mistake.
• When filtering a message,
SpamSieve can optionally check whether the message is in the corpus. If it is, SpamSieve looks up the answer rather than trying to predict. One use of this feature is that if SpamSieve makes a mistake, you can Add Spam and then Label/Move If Spam and be sure that the message will be labeled/moved.
• Keeps a log of additions to the corpus, filtering results, and errors.
Mailsmith: If SpamSieve thinks a message is spam, it sets the deleted property of the message to true; otherwise it sets the flagged property of the message to true. Therefore, if SpamSieve has classified the message then exactly one of the properties will be true, and if it hasn’t they’ll both be false. (Normally, neither of these message properties is used by Mailsmith itself.)
Entourage and PowerMail: If you tell SpamSieve to move spam messages to a spam folder and the spam folder doesn’t exist, the script will create the spam folder for you.
• When you add spam messages to the corpus, can optionally move them to a Spam folder.
• Added status indicators in the Dock icon (like Norton
DiskLight).
• The spam probability of unknown words is now 0.4 instead of 0.2.
• The Corpus window uses less memory and sorts much faster.
• Accuracy tracking is faster and uses less memory and disk space.
• Fixed bug where accuracy tracking didn’t work for some
Mailsmith messages with multiple parts.
• Improved the manual’s instructions for e-mail client
integration.
• Compiled with GCC 3 for greater speed.
• Uses the latest version of the eSellerate SDK, which eliminates a crash at startup under certain circumstances.
• No longer shows the “Upgrading From 1.0” message when starting with a blank corpus.
1.1—September 19, 2002
• E-Mail Client Integration
• Added support for PowerMail.
• Added instructions and an
AppleScript for making Mailsmith download and filter mail faster.
• Added an
AppleScript for Entourage that moves spam into a Junk folder.
• Performance
• Launches about 60% faster than 1.0.
• You can now prune the corpus to remove words that are taking up memory without contributing to spam recognition. This can also dramatically decrease
SpamSieve’s launch time.
• Recalculating spam probabilities is about 10% faster and uses less memory.
• Quitting is faster because
SpamSieve now writes corpus changes to disk during idle time.
• Saving the corpus is slightly faster.
• Displays statistics about the number of messages filtered, SpamSieve’s accuracy, and the types of words in the corpus.
SpamAssassin’s X-Spam-Status headers are now treated as single words. This means that if SpamAssassin is running on your mail server, SpamSieve will learn to respect (or ignore) its judgment.
• Does a better job of ignoring e-mail attachments, thus reducing corpus bloat.
• Installs the eSellerate Engine if it’s not present, thus enabling “Instant Registration” for more users.
• Asking
SpamSieve to categorize a message now forces an update of all the word probabilities. Previously, the update only happened during idle time.
• Highlights the sorted column in the Corpus window. The columns themselves have shorter names. There’s a new “Total” column. Auto-resizing of the columns works better. You can now manually resize any column, and manual resizings and reorderings are saved between launches.
• Shows fatal errors as alert panels rather than just printing them on the console.
• The Corpus.plist data file is now sorted by word. This makes it easier to examine the corpus manually, and to compare it to other users’ corpora.
1.0—September 10, 2002
• First public release.

Contenu