We released Miro 2.5 a couple of weeks ago and it was a messy release which resulted in a Miro 2.5.1 a day or two later and a Miro 2.5.2 a few days after that. It sucked. I wish it had been better.
Before a release has happened, a messy release looks like any other release: development has wound down, lots of bugs have been fixed, a bunch of new features have been implemented, testing is going well, people who have looked at the nightlies and release candidates like the new features and haven't encountered any issues, ... Then we decide to pull the trigger and push it out.
It's only when a release is out that the messy release rears its ugly head and the obviousness of its messiness is revealed. Usually this is made abundantly clear by abusive emails from people who encounter problems with the release, become frustrated, and express themselves with abusive vitriol aimed at me and all the other people who worked hard on the release.
It's at this point, I bite the bullet, prepare for the world of pain I'm about to get involved in, and try to reach out to as many people as I can, gather information to identify the issues so that we can fix them, and provide support to people who are encountering problems to enable them to get through them.
So what the hell happened with Miro 2.5?
Miro 2.5 started out with a big flaw: there was no upgrade dialog showing when Miro was transforming the database from the old style to the new style. If you have a small database like mine (8 mb), you'd never notice. People who had huge databases would launch Miro 2.5 for the first time and ... seemingly nothing would happen for minutes. The person would assume Miro isn't starting, would kill the process, and then try again. At this point, Miro would see a wedged database and wouldn't start. We think there were on the order of 30,000 downloads at this point and we had maybe 50 people who had problems.
We recognized the problem and we kicked into emergency mode. Janet and I worked the GetSatisfaction forums, blog, and identi.ca. Ben poured over the incoming data, made a few fixes and put in a progress dialog. We pushed out Miro 2.5.1. Janet and I went through and talked to each user who had problems to help them unwedge their database and get back to a working Miro.
A day or two later, we found another couple of problems. I made some fixes and pushed out Miro 2.5.2. Then I wrote a script people could run to re-download data that was lost. Dataloss totally sucks. We did everything we could to help alleviate the problem after we found out it was there.
There are still a bunch of people for whom Miro doesn't work. I'm talking with a few of them. Others have given up. It's always been the case that there are people for whom Miro doesn't work and we do what we can to figure out the causes.
And that's where we are now.
Going forward, we're going to figure out how to better test Miro release candidates. We thought that 600 people would have fleshed out most of the issues, but it seems that there are classes of people whose usage patterns aren't reflected in our testers and testing. If you can help us with testing, we'd love to have you. You will have a direct impact in making Miro better.
Going forward, I'm going to work on making it easier for Miro to help people identify problems and let us know. I'd also love to get some help fixing Miro so that it doesn't lose data when things go wrong. Code contributions would help a ton here.
Going forward, I'm going to ask for help more often. There are only 4 PCF developers and we're spread very thin between Miro, Miro Guide, Miro Community, and the other things we're working on. We need your help. This isn't a new situation. What is new is that I'm going to do a better job making it explicit what I need help with right now.
I love working on Miro. I love what Miro is doing for people and for communities. I love our mission. I need your help to help me help you. Miro is your project as much as it is mine.
I hope the initiatives we're working on will eliminate messy releases in the future.