I have spam filtering active. When I connect to my IMAP (over TLS) account that has several hundred messages to check, I can't read any messages until the spam filter has finished running spam check agains the new messages in the mail store. When I click on a new message (while spam check is running), it opens a new window that is completely blank. Mozilla/5.0 (Windows; U; Windows NT 5.1; en-US; rv:1.2b) Gecko/20021108

Taking... this makes sense, because only one connection to an IMAP folder at a time is permitted by the back end code. Serializing the classifications at the message DB level (rather than allowing the IMAP code to do the serializations later) seems like it might fix this. David, do you think it would? Although I seem to recall someone saying that that would slow down the overall classifications...

this should be serialized, but I don't see how it would happen at the message db level.

bienvenu: well, right now nsMsgDBFolder is passing a null callback to ClassifyMessage. Instead of iterating through all the messages in CallFilterPlugins, it could conceivably start the first message, and have a callback that chains the second message, which would have a callback...etc. Thoughts?

you should ignore the nsMsgDBFolder call. nsImapMailFolder passes in itself as the callback, as does nsMsgLocalMailFolder. What confused me was your reference to the db. This has to do with the folder, not the db. Yes, the folder could chain the uri's. However, we're going to need to change the spam filter plugin anyway to get it to run the url in the right msg window, so that stop will work, etc, so we should do both at once, since I believe the latter is going to require changing the spam filter interface, and the spam filter itself.

To summarize more of this discussion, Dan argues that we should put the chaining in the core libmsg code because we might have multiple spam filter plugins and having the plugins do the chaining will lead to code bloat. My response is that we already have three places in the core libmsg code that invoke the spam filter plugin, so those three places would have to do chaining. If we had four spam filter plugins, then, yes, perhaps that would be a little bit of bloat. An additional response, we would need to add state to nsMsgLocalMailFolder and nsMsgImapMailFolder to handle the chaining (probably a pointer to an array of msg keys to categorize, and the index of the current message), and this state would only be used when categorizing messages, but we would pay the price for this state in every single folder in memory. There are a few other possibilities - add an interface to the msg service that knows how to stream multiple messages to a stream listener, and have the implementations of this interface handling the chaining, and checking for the stop key. Or, similarly, add a utility class that does something similar that we instantiate only when needed. Ultimately, we do need to go through the msg service instead of just fetching the uri directly, so that we can set up the msgWindow on the url, so we'll get progress and status, and enable the stop button handling. We'll need to add a new method to nsIMsgMessageService so we can add "header=filter" to the url. Adding an interface to the msg service that knows how to stream multiple messages to a stream listener would probably still require creating a helper class to do the chaining. Looking at the bayesian spam filter plugin again, I see that it goes through a stream converter before tokenizing. I think this means we'll need to have a special method in the msg service that knows it needs to run through a converter when fetching the message. So either way, we're going to need to add an interface to the msg service. So the question remaining is: do we want to do the chaining in the msg service (or helper class), or in the filter plugin? I still find the possibility of having four filter plugins rather remote, so I'm going to think about this a little more.

I fixed this a while ago, reassinging.

fixed

Using build 20030414 on winxp, macosx and linux it a large folder is being evaluated by Junk Mail Control, I can read a mail message while the JMC is running. I can also stop the JMC using the Stop button. Verified