On Sun, Oct 20, 2002 at 03:04:51PM +0200, Oskar Sandberg wrote: > On Sat, Oct 19, 2002 at 11:15:38PM -0400, Michael Wiktowy wrote: > > I certainly didn't interpret a week-countdown reset each time a bug is found > > in Oskar's bug release pseudo-code. > > You interpreted it wrong then, because that is what I meant (or rather, > I meant we should set the time back every time we _fixed_ a bug, but > that we should not fix any non-critical bugs in the release candidates). > I don't see what the problem is, we don't have any financial deadlines, > and we have gone more than three years without producing anything even > moderately useable - I doubt anything is going to blow up in the next > couple of weeks if we take a strategy that a final release can only be a > candidate that gets upgraded after we are sure it contains no > critical bugs. > > What we need to ask ourselves is why we are making a even numbered > release at all, it's not like we are running out of numbers. Ian says it > is because he wants this code to supersede the ancient 0.3 stuff as > something stable - which would be credible if he was just going to > change a snapshot and some text on the website. But that's not the case: > he's already made it clear that he wants to use it as a PR oppertunity, > contacting journalists, getting on slashdot, and whatnot. > > We have led people on too many times. For the last three years, freenet > has been shoved on people again and again and again, but we have never > been able to produce anything that is actually of any value to anyone. A Heh. > lot of people will already ignore any news of a new release because of > this, and for every time we do it, that number increases. I think this > may well be our last chance to get anyone to care what we do. Oskar's general thrust here is correct. > > I'm not against making releases for PR reasons - as far as I'm concerned > that is the only reason for doing a release. But good PR starts with > quality control - if we get a chance in the spotlight, what is wrong > with being damn sure that we have something working to offer to people > who are lured by it? > > That means that packages and installers should work well, in both > windows and linux. That means documentation for sticky issues (like NATs > and setting the nodes address etc) need to directly available to anyone > who downlaod. That means we shouldn't be ignoring serious bugs because > they appear to be JVM issues (we hadn't worked around any JVM issues in > fred, we would have had a beautiful piece of code right now that > wouldn't run anywhere). That means we can't ignore the startup time > issue, which when we tracked it down comes down to a horribly slow size > trim of the routing table - something that is done once every maybe 5 > minutes when the node is running and will prevent any routing during the > 1.5 minutes it seems to take on a large node. That means we can't > release until we have properly made sure that the node does NOT stuff 54 > nodes from a seed file into the routing table every time it restarts, > pretty much setting it back to square one. And it means we don't release We need to fix that one before releasing for the good of the network. This means that there will be a new prerelease on monday - I'm not holding _THAT_ fix back. > until we are sure that any there are no regressions from previous > bugfixes. I'm tending to agree with this position now. pre6 on monday/tuesday, final on the following monday/tuesday. Meantime, development and testing goes on but only vital fixes are committed and the rest is kept locally. And if we find something _critical_ (eg stuffing all the node refs into the routing table on startup), we reset. > > Time is not critical here. Nobody who cares about a Freenet release > today will care less if it comes in a week or two. What is important is > the experience of those would be induced to give us another chance by > this release - if we burn them again we might as well start packing. This is the key. We want to make a much better impression than last time, then we might actually keep some of the thousands of users that will probably try it out.
The other technical issue is seedNodes. Does hawk provide enough seednodes? Will it and all the seednodes be immediately swamped? Do we need to institute collection of refs and a rotator or something? Or at least up hawk's rtMaxNodes to 1000 or so, and have a script grab 10 random working ones? -- Matthew Toseland [EMAIL PROTECTED] [EMAIL PROTECTED] Freenet/Coldstore open source hacker. Employed full time by Freenet Project Inc. from 11/9/02 to 11/11/02. http://freenetproject.org/
msg04558/pgp00000.pgp
Description: PGP signature
