On Friday 12 June 2009 20:02:06 Matthew Toseland wrote: > A series of increasingly desperate pre-release builds: > 1219: > - Various translation and wording improvements. > - Much less disk access for inserts when a segment has been completed, and a > fix for a NullPointerException that was causing some site inserts to stall at > 100%. > - Better handling of plugins load failures, sort the dropdown list in > alphabetical order, workaround a Freemail bug so it doesn't get shown on that > list if it is already loaded, class loader fix. > - Reduce the severity of the peers-say-key-is-blown warning. > - Make clean-dropdown the default theme, this has a horizontal menu with > pulldown submenus, saves lots of space, and doesn't jump around when you > mouseover. > - Simple vs advanced flag is now applied globally immediately. > - Tell the user when a file is compressing, and when it is waiting to be > compressed. > - Remove the "je dump" button from the stats page. > - Improve appearance of the confirmation pages for high friends seclevel / > high network seclevel in the wizard. > - Fix a rare bug causing a transfer to stall for over an hour. > - Change some strings to say 0.7.5 not 0.7. Add a new "release version" since > we can't actually change nodeVersion safely. > > 1220: > - Don't tell the user a file is being compressed if it has the dont compress > flag on. > - Change the default fproxy max-size to 2.1MB, workaround bugs in site > insertion code, several sites are like this atm. > - Show the filename on the MIME type warning page. > - Make the peers-say-key-is-blown page even less alarming, basically it is > saying why we can't update at the moment. > - German update, mention incognito mode and Chrome on the separate-browser > warning page at the beginning of the wizard. > > 1221: > - Various changes to the auto-updater to prevent the current deliberate > malicious attack which is causing lots of nodes to say that their peers have > found the revocation key. > > 1222: > - Don't show the progress page if ?forcedownload is set, this causes us to > send a new copy of the data every 2 seconds! > - Fix a NullPointerException preventing startup on some nodes introduced in > 1221. Sorry folks! >
FYI, detailed changelog for 1221: 1221: Auto-updater: handling of transfer of revocation certs from peers: - Fix a bug where we would not remove from nodesSayKeyRevoked when the URI isn't parsable. - Only count connected nodes when deciding whether we should suspend update (mightBeRevoked(). - For each peer, track how many revocation failed transfers we've had from a peer since the last disconnection. - Only send UOM announcement talking about revocation certs if we have actually revoked, NOT if we just have a local problem equivalent to revocation. - Separate messages for the auto-update key being blown versus the auto-update key might have been blown or we might have a serious local problem preventing us from knowing whether it has been. In both cases we turn off updating. - Move the revocation key size limits to NodeUpdateManager. Enforce them when decoding a revocation cert. If it is TOO_BIG then it will result in a fatal error and we will treat it as if we had got the revocation cert. Reduce the revocation cert blob max size to 8KB from 4MB. - Don't add to the failed transfers list unless it is an actual failed transfer that might not be the peer's fault. If it's likely to be the peer's fault (e.g. wrong URI), don't worry about it. - Remove from nodesSayKeyRevoked and add to the transferring list just before transfer starts. Don't add to the transferring list at the start of the function; if we asked, will already be in nodesSayKeyRevoked, this is enough; if we don't, it's probably bad anyway. - If a node announces it has the revocation cert and then instantly disconnects before we can ask them, log an error and ignore it. - Bugfix: Remove from the transferring list *before* calling maybeNotRevoked(), when we have a real revocation cert transfer failure. - Tell UpdateOverMandatoryManager when a node is disconnected (and via onRemove, when it is removed too). Fixes memory leaks and necessary for the next change. - Don't start a transfer in UOMAnnounce handler if there is already one running. - Retry a failed transfer for 3 attempts. - For purposes of mightBeRevoked(), ignore any peer which has tried to send the revocation cert and failed transfer more than 3 times. - Don't care about failed transfers in mightBeRevoked, only consider what is waiting for a transfer and what is transferring. Bulk transfer: - Reduce the inter-block timeout on bulk transfers to 30 seconds from 5 minutes. This is still very generous, and it avoids attacks. Dev stuff: - Logging. sdiz toad -------------- next part -------------- A non-text attachment was scrubbed... Name: signature.asc Type: application/pgp-signature Size: 835 bytes Desc: This is a digitally signed message part. URL: <https://emu.freenetproject.org/pipermail/devl/attachments/20090619/f145a4cf/attachment.pgp>
