Slashdot Mirror
How To Track the Bug-Trackers?
schneecrash writes "Submitting bug reports — and waiting for responses etc. — seems to be SOP for developers and users alike, these days. Every project has some sort of bug-tracker — bugzilla, trac, mailing list, etc. E.g., we currently track 200+ external bugs across ~40 OSS projects. Half the bugs depend on something else getting fixed, first. Every bug has its own email thread, etc. Management asks 'How we doin' overall?,' and suddenly everyone involved gets to work removing dried gum from the bottom of their shoe. What do Slashdotters use/recommend for centrally keeping track of all the bugs you track across all those different bugtrackers? In particular, managing communications and dependencies across bugs? So far, the best method I've managed to use is bunches of PostIt-notes stuck to the screen of an out-of-commission 32" TV (glossy, non-matte screen, of course!)."
I have a nice wire-frame basket designated for bug reports, electric bills, fast food wrappers, and soda cans.
We track bugs internally using Bugzilla. When we raise a bug against a project we depend on, we also raise a bug in our internal Bugzilla that links to the other bug, then we can use Bugzilla's depends-on and blocks to track the external bug.
The only downside is that someone needs to go and check periodically on the state of that external bug. It would be nice if Bugzilla let you mark a bug as "depends-on" a bug in someone else's Bugzilla.
That's just the start.
Dog is my co-pilot.
That sounds great, until a week in you get to a bug that you can't fix because something else is blocking it. Now you start working on that second one for about a week before you realize that it, too, is being blocked by another bug. And so, and so on. Pretty soon, you're six months in, haven't accomplished anything, and are no closer to solving those dependencies. And without mapping them out as you go, you might not know which ones to go back and fix in which order. Not to mention, not all bugs are created equal. Some you can ignore for now, some cause the entire project to stop dead in its tracks.
No, the best solution is always to stop what you're doing, back up, and make sure everything is mapped out. Better to spend two weeks doing that and get all your ducks in a row rather than running off and making things worse than they already are.
God invented whiskey so the Irish would not rule the world.
Must not be a widely-used application then. While this approach is laudable, it's hardly sustainable for any project of substance.
and if you use LDAP instead of Crowd for user management, then in the darkness it can in fact bind them.
"Once bugzilla implements the features in the next version and/or fixes bug #233412 and #455354, and Trac patches the rss feed problem, and...well, I just don't know how I'm going to keep track of all these bugfixes so I can get my bugfix tracking program working properly!"
This doesn't scale.
Consider large-scale projects, something like the linux kernel, glibc, or X, where it is unreasonable to try and fix all bugs.
There are various reasons for this...some bugs can't be reliably reproduced so you add instrumentation to gather information when they do recur, others require months of uptime to reproduce, or special hardware/software that you don't have so it takes a long period of back-and-forth with someone that does have it.
Consider cases where you're too close to release to fix the bug "properly", so you paper over it temporarily but you want to leave a bug report open to remind you to fix it in the next release.
Seriously, I've found them to be the best method of issue tracking.
Your .NET program just crashed with a stack trace of a size that is only rivaled by Java. Please visit your postal clerk in a few days to pick up the extremely large package I sent (at considerable personal cost) containing 12,345 post-it notes.
There's no place like
a.k.a. "Look at my dick, it's enormous."
Judging by what you didn't discuss, you don't seem to distinguish fixing surface manifestations from root causes. I find that when I focus on addressing root cause, I don't run into this regression. Code is correct when it performs as specified in the design and API documents. If that means the entire application bombs out because something on the other side is broken, that's a separate consideration.
Put in the correct code (as if the other side is also correct) and then mark the bug as fixed, pending final validation when the other side is also fixed.
If addition of correct code makes the surface manifestation unbearable (application constantly bombs), then you can restore the broken code as a temporary work-around, to be removed as soon as the bug on the other side of the interface is properly addressed.
If one side of the API knows which cases are failing on the other side, you might be able to detect the suspect case and write a record to the log file "overlapping FOO handled by dubious heuristic BAR". Then if the work-around code causes a downstream cascade of weirdness, people don't get too quick to decide that another 100 bugs need to be reported.
I've also found that conservative design and coding conventions mitigates these interdependency cascades.
It can be a good idea wading into an unstable code base to pare down to a completely stable build. This is a build where you compile out functionality until what remains is believed to be rock solid. If your rock solid build is hugely divergent from your production build, then you likely have a serious cultural or management problem.
It's almost always faster in the long run to merge solid code back into the rock solid build than jello diving on an unstable production build.
On a big project, it's great to have a couple of world class jello divers to actually get the product out the door.
It should never be half of your release team. An ecology of jello dancers tends to support only one big fish, as the Microsoft story illustrates. When half of your release team is 100 FTEs, it's amazing what you can pass off the buying public long enough to rinse and repeat.
I can also say as an architect that I'm wounded to the core if my code base degenerates into a cascade of jello dancing. More than once I've seen the exact moment where a solid code base takes a bad turn. One time I took a several weeks off due to a death in the family at a critical design juncture. Everyone worked really hard to fill in the gap, and some new features were completed in record time by bastardizing one facility to do what it hadn't been intended to do. The project lived to see another day. The software was never as solid again. It's brutally hard to back these things out.
I've also sat down to discuss implementation strategies with other developers only to discover the person was navigating around the core difficulty in the belief that he ended up with would handle 99% of the cases required by the next shipment point. The person might look at this as the difference between one day to whip it out (cue infantile jokes), instead of three days to nail the specification, at a point in the deliver schedule when days (and nights) are precious resources.
It's a false calculus once you consider the downstream investment in validating the code, shipping the code and obtaining marketplace confidence, bug reporting cycles, and the persistent niggling fear that every subtle bug "could be one of five or ten different corner cases we shaved off in the last iteration".
No large project is ever shipped without cutting corners, but you have to choose your battles carefully. If you end up with a cascade of interdependent bugs, someone along the way made some unwise choices.
By the time you are proposing to build a meta-bug tracker because these interaction chains exceed what can be mentally managed, it sounds like a invitation to a bring in an MBA student to write a case study of a business whose business model had degenerated into managing their own dysfunction. I would think long and hard about this before I mastered my queasy feeling.