wiki

d distributions which build on Debian to make `hyphanet-fred` available where it is most important. Once this is done, tools which build on Hyphanet — like FMS, but also jSite and tools from pyFreenet — can be packaged to work out of the box, using Hyphanet as an ordinary background service. That’s a step towards Hyphanet as decentralized, privacy-preserving communication backend for other applications. Another step towards this is accepting the Schema hypha[net] to simplify writing browser extensions that forward hypha:-links to Hyphanet. The networking layer was optimized significantly. Searching packet types is often stopped early and common or cheaper checks are done before less common or time-consuming checks. This gives significant reductions of CPU load, especially for very fast nodes. Juiceman fixed a bug limiting MTU to 1280 where not needed. And recently failed and data not found cooldown times were reduced to 5 minutes and 3 minutes, reducing one of the big annoyances when accessing a site quickly after upload. On the data transfer layer, healing was optimized. After 1495 strongly increased the amount of healing to keep large files available for longer, 1498 specializes healing to keys close to the node location. This reduces healing per file, but improves privacy, because healing inserts are then more similar to forwarding — they mostly send data close to the nodes location — and it reduces the network load of healing, because the specialized healing inserts need fewer hops to reach the optimal storage location in the network. In addition to these changes deep down, there are a number of directly visible improvements. The plugins KeepAlive and Sharesite are updated (the latter now uses the new Night Zen Garden style). The UPnP2 plugin is now visible in simple mode. It can replace UPnP and should work better. On the flipside the Library plugin is moved to advanced plugins, because it does not work reliably enough. The plugin list is easier to navigate by removing the defunct option to download plugins from the clearnet and by adding better styling. Downloading from the clearnet was an unnecessary privacy risk since we’ve been bundling essential plugins with the installer for a few years now. The noderef for friend-to-friend connections is shown in simple mode again, because it is robust enough with the changes in recent years. This should remove a barrier to adding direct connections and enabling fully confidential messages between friends. There are new configuration options to allow connecting via local services. That’s a step towards making it easy to add a second layer of security, for example confining connections to a local network. Thanks goes to s7r for these changes! When bandwidth detection fails, the upload bandwidth now defaults to 160KiB/s. Also the NLM config is now disabled statically. This was a testing setup which could still be active in old nodes, but it would break connectivity nowadays. The default bookmarks include the Opennet SeedNodes statistics, the generate media site to create decentralized streaming sites, and the high-impact-tasks. The bookmarks are also re-ordered to be a better match for newcomers. Starting category: first steps, clean spider, Index of Indexes. For the software category ordered by ease of use from fproxy. For website authors, more CSS elements, selectors and combinators (`:checked`, `word-wrap: anywhere`, `focus-within`, `^=`, `$=`, `=`, `>`, `+`, `~`) and additional HTML elements (`summary`, `details`, `<meta name="Viewport"...>`) are available. This strongly expands the possibilities of websites authors in Hyphanet, because Javascript or webassembly are no viable options in an environment where a privacy breach could put people at risk. We’ve seen with Java applets, that untrusted code will always break out of its containment. The CSS improvements in contrast provide a safe way to enable limited interactivity. Streaming support via m3u lists was improved to allow accessing segments of up to 200MiB. And using `-1` as version in a USK now properly finds version `0`, if this is the only existing version. There were a number of Java 21 fixes, including all our tests (thanks to Bombe!), and improvement to the github actions (thanks to AHOHNMYC). In addition to that there was a lot of polish. Bert Massop and Veniamin Fernandes replaced our homegrown CurrentTimeUTC with modern Java options. Alex fixed the pronoun used in strings. Bombe added getters for all direct field access in the node. Hiina reduced logging level of store warnings so no unneeded backtraces are created for node with large stores and Juiceman updated code to use more modern structures. Time-dependence of compressor selection was removed. This caused non-determinism for inserts and could cause keys to be non-reproducible on systems with faster or slower network. And finally the new [exe signing workflow][] we built to fulfill the requirements of SignPath, our new windows installer signing provider for the upcoming releases, runs the [verify-build script][] on every release to ensure that the jar we release has actually been built from the sources. This provides a second safety net, in addition to anonymous users running the script and posting the results (thanks to all who did this — please keep it up, otherwise people have to fully trust github). The release is not yet byte-by-byte reproducible, because the jar MANIFEST defines among other info the exact java version used to compile it, and the java version available differs by distribution and time, so it would get harder over time to verify the build. A special thanks goes to Bombe for many careful reviews! [high-impact-task]: https://github.com/hyphanet/wiki/wiki/High-Impact-tasks [Roadmap]: https://github.com/hyphanet/wiki/wiki/Roadmap [exe signing workflow]: https://github.com/hyphanet/sign-windows-installer [verify-build script]: https://github.com/hyphanet/scripts/blob/master/verify-build

ub.com/freenet/wiki/wiki/Projects Developing dynamic stuff is taking so long because it is a complex endeavor: On the regular Internet, censorship happens by "look up who owns the IP, go to their address, remove the computer." Since this is not possible on Freenet as everyone is anonymous, censorship will happen by denial of service: For example forum systems would be spammed to death to get rid of unwanted content. Thus the architecture of censorship-resistant systems has to be reinvented from scratch, you can't just take a regular forum system and stick Freenet on top of it. It has to be decentralized to be resistant against DoS - there must not be Tor-alike central servers ("hidden sites" / .onion sites). Instead messages are stored across the whole network and replicated automatically if they are downloaded more often and thus need more bandwidth (the added redundance also makes them more censorship-resistant). And spam filtering need to be a first-class application, I have worked for years only on that. So the different architecture is the primary pitfall which many projects which decided "Freenet is too old, we're gonna build this from scratch with nice Javascript etc." fell into IMHO: First it's "we'll develop a regular app, we can bolt Tor onto it later", then they realize that the threat-model is so different that this is just not possible and the projects never become anonymous/censorship-resistant. So privacy needs to be built in from the start. Luckily, Freenet did that right (even though it was the first anti-censorship + privacy network!), and I don't mind that it's taking decades to develop because of the extended threat model: That's still better than being one of wheel-reinventing post-Freenet projects which then abandon the privacy idea in the end anyway, or postpone it forever.