November 12th, 2012
· This is an important release correcting severe bugs affecting the overall wealth of the system: the PARQ upload queue was not properly restored, the DHT was not always sending correct values back (resulting in failed lookups), and SDBM (the hash-on-disk database) had two well-hidden bugs that could cause crashes at startup or that compromised regular runtime operations.
· This release uses a reliable UDP layer to transfer out-of-band query hits, ensuring that these important messages are properly sent and received and never lost.
March 24th, 2011
· UPnP and NAT-PMP support was added.
· More status icons were added for DHT, port mapping, and upload pre-stalling information.
· 12 minor bugs were fixed and 25 "under the hood" improvements were made.
March 29th, 2009Bug Fixes:
· #1932909 "columns in search window do not scale properly".
· Fixed improper HTTP chunking when outputing chunks larger than 64 KiB.
· #2628899 "tries to download from my old self after IP changes".
· Was not always resuming the interrupted SHA-1 computation of a finished file when servent was closed in the middle of processing.
· Fixed bad range checking test in the download part.
· Was not paying attention to the "give_server_hostname" property and instead sent "HNAME" GGEP extensions in query hits provided a non-empty name was defined as the "server_hostname".
· Fixed improper upload slot counting which introduced race conditions, sometimes leading to having more slots used than the configured maximum (in addition to the "quick" upload slots).
· Prevent occasional crashes at shutdown time, during subsystem cleanups.
· Fixed corner cases where we could lose a server hostname and downright bug where we discarded the hostname after a DNS lookup if the subsequent connection attempt did not succeed.
· Refuse to share certain directories: /, $HOME, $GTK_GNUTELLA_DIR, and the incomplete and corrupted directories.
· Downloads GUI: added checkboxes to invert the select/filter patterns.
· Downloads GUI: added "Modified" column and detail showing the timestamp of the last file modification.
· Optional beautification operations on filenames: strip leading "-", remove consecutive "_" and prevent "_" surrounding punctuation characters.
· Fixed significant performance issue in the Gtk+ 1.2 front-end: update of upload statistics was O(n) but is now O(1).
· Background TTH file verification now lights an icon in the GUI.
· Show TTH tree depth in the download "Details" pane.
· Update the bitzi metadata info box when a query finishes, so that it is not required to click on the result again after requesting bitzi data.
· Added MIME type column for search results.
· Added "Last Request" and "Last Upload" columns to the Upload History.
· Almost complete DHT support (missing only publishing of data), but code needs to be explicitly enabled by setting "enable_dht" to TRUE. The DHT knows how to look for alt-locs for files and push-proxies for firewalled hosts, and will store local data for others to query.
· New shell commands "stats", "download", "echo".
· It is now possible to rename files being downloaded (through GTK2 or shell, no support in GTK1).
· Added more filename extensions to the built-in search filters.
· Switch resources after getting an HTTP error if the connection is kept alive and the error permits follow-ups (non-busy conditions).
· Improved ability to switch resources after file completion by ignoring trailing swarming data, to keep the connection alive.
· Bitzi tickets are refreshed when explicitely requested, avoiding reuse of stale metadata
· Completed files are displayed during hash verification.
· Only propagate good alt-locs for files we are downloading, i.e. addresses of hosts which we know to be serving that file because we contacted them.
· Many new statistics counters.
· Allow scheduling of special downloads without considering per-host limits, as these are of a much shorter duration (THEX and browse host requests).
· Always seed completed files, regardless of their size (no PFSP minimum as with partial files), when their SHA-1 has been verified.
· Incorrectly encoded Bitzi metadata is translated to Unicode.
· Browse-host responses are sorted alphanumerically on relative pathnames to allow users to relate files together if they don't understand GGEP "PATH".
· Improved routing of PUSH messages by also looking whether targetted hosts happen to be one of our immediate neighbours.
· Updated Norwegian Bokmal and Spanish translations.
Under the hood:
· Parse legacy Shareaza THEX URIs "urn:tree:tiger/:" (extra "/" here).
· Handle non-standard X-Thex-URI headers if peer provides a valid X-Content-URN header (Shareaza 188.8.131.52).
· Also attempt to give priority to the last "pfsp_last_chunk" bytes of downloaded files, since that can contain important indexing information to enable pre-view of partial video files.
· More MIME types known (derived from file extensions).
· Can now download THEX data at a tree depth of 11 for files larger than 1 GiB.
· Can download sub-optimal THEX data, as non-optimal data is better than none, but keeps looking for THEX data at the optimal tree depth nonetheless.
· Reimplemented scanning of shared directories as a background task so that the process (not just the GUI) is fully responsive during the whole scan.
· Let the background task scheduler compute the proper amount of ticks to use for each task.
· Upload statistics now record the file's SHA-1, so that we can consolidate partial file sharing with library sharing statistics, with the filename being forced to the library's filename.
· Avoid starting moving a completed file if there is not enough space on the targetted file system.
· Recognize and propagate the "DHT" GGEP extension in pongs.
· Added support for "DHTIPP" GGEP extension in pings, used for DHT bootstrap.
· Added more entropy to the random number seed.
· Plugged memory leaks
· No longer send SHA-1 queries on the Gnutella network. They are now performed through the DHT, which has the suitable topology to do so efficiently.
· Fixed UDP bandwidth computation, and removed any limit on UDP incoming.
· Updated list of hostile IP adress ranges, Geo-IP data, spam patterns and spam samples.
· Property loading from files was in O(n*n), made it O(n).
· Now detect property files being blindly copied over from another gtk-gnutella's config directory and reset properties whose value needs to be unique.
· The GUID and the KUID (Kademlia ID, the DHT node ID) were made sticky.
· Optimization of QRP lookups in tables ranging from 16k to 2M slots.
· Limit amount of hits to 10 per query hit packet, as LimeWire chokes otherwise.
· Limit amount of alt-locs in hits to 10 or LimeWire chokes.
· Keep connection alive when reporting an HTTP error to allow remote resource switching if necessary.
· Emit and parse new X-Downloaded header, apparently introduced by LimeWire.
· Added parsing of X-FW-Node-Info headers to grab push-proxies from firewalled hosts, as well as gather the proper GUID and host address/port.
· Emit X-FW-Node-Info if node is firewalled but continue to list our push-proxies in the standard X-Push-Proxies header.
· Avoid endless aggressive chunk splitting during swarming.
· Proper parsing of comma-separated values in Accept and other Accept-Encoding HTTP headers.
· Reject UDP packets from unspecified addresses.
· Drain more data out of the UDP receive buffer to prevent it from filling up.
· Strip at most 5 trailing characters from words when building the QRP table, thereby inserting less radices and making the table more sparse.
· Removed workarounds for dealing with broken 0.96.2 nodes and earlier 0.95.x.
· Extended syntax for push:// exact sources in magnets to be able to supply all the known push-proxies, even if that is an empty set. Since downloads are persisted as magnets, this also allows keeping all the previously known push-proxies accross sessions.
· Upgraded PARQ to version 1.1 (mostly specification cleanups).
· Added additional anti-hammering for cases where many files on the same server are queued and gtk-gnutella is configured to launch more than 1 connection per IP. Connection attempts to the same server are now limited to 5 per minute.
· DHT code is still experimental, although functional: configuration of parameters can only be done through the shell, and there is no support for publishing yet.