In any project that's making active use of its bug tracker, there is always a danger of the tracker turning into a discussion forum itself, even though the mailing lists would really be better. Usually it starts off innocently enough: someone annotates an issue with, say, a proposed solution, or a partial patch. Someone else sees this, realizes there are problems with the solution, and attaches another annotation pointing out the problems. The first person responds, again by appending to the issue...and so it goes.
The problem with this is, first, that the bug tracker is a pretty cumbersome place to have a discussion, and second, that other people may not be paying attention—after all, they expect development discussion to happen on the development mailing list, so that's where they look for it. They may not be subscribed to the issue changes list at all, and even if they are, they may not follow it very closely.
But exactly where in the process did something go wrong? Was it when the original person attached her solution to the issue—should she have posted it to the list instead? Or was it when the second person responded in the issue, instead of on the list?
There isn't one right answer, but there is a general principle: if you're just adding data to an issue, then do it in the tracker, but if you're starting a conversation, then do it on the mailing list. You may not always be able to tell which is the case, but just use your best judgement. For example, when attaching a patch that contains a potentially controversial solution, you might be able to anticipate that people are going to have questions about it. So even though you would normally attach the patch to the issue (assuming you don't want to or can't commit the change directly), in this case you might choose to post it to a mailing list instead. At any rate, there eventually will come a point in the exchange where one party or the other can tell that it is about to go from mere appending of data to an actual conversation—in the example that started this section, that would be the second respondent, who on realizing that there were problems with the patch, could predict that a real conversation is about to ensue, and therefore that it should be held in the appropriate medium.
To use a mathematical analogy, if the information looks like it will be quickly convergent, then put it directly in the bug tracker; if it looks like it will be divergent, then a mailing list or IRC channel would be a better place.
This doesn't mean there should never be any exchanges in the bug tracker. Asking for more details of the reproduction recipe from the original reporter tends to be a highly convergent process, for instance. The person's response is unlikely to raise new issues; it's simply going to flesh out information already filed. There's no need to distract the mailing list with that process; by all means, take care of it with a series of comments in the tracker. Likewise, if you're fairly sure that the bug has been misreported (i.e., is not a bug), then you can simply say so right in the issue. Even pointing out a minor problem with a proposed solution is fine, assuming the problem is not a showstopper for the entire solution.
On the other hand, if you're raising philosophical issues about the bug's scope or the software's proper behavior, you can be pretty sure other developers will want to be involved. The discussion is likely to diverge for a while before it converges, so do it on the mailing list.
Always link to the mailing list thread from the issue, when you choose to post to the mailing list. It's still important for someone following the issue to be able to reach the discussion, even if the issue itself isn't the forum of discussion. The person who starts the thread may find this laborious, but open source is fundamentally a writer-responsible culture: it's much more important to make things easy for the tens or hundreds of people who may read the bug than for the three or five people writing about it.
It's fine to take important conclusions or summaries from the list discussion and paste them into the issue, if that will make things convenient for readers. A common idiom is to start a list discussion, put a link to the thread in the issue, and then when the discussion finishes, paste the final summary into the issue (along with a link to the message containing that summary), so someone browsing the issue can easily see what conclusion was reached without having to click to somewhere else. Note that the usual "two masters" data duplication problem does not exist here, because both archives and issue comments are usually static, unchangeable data anyway.