> I forgot to mention that we are still better than before. > I think with the textile formatter, we would be still at 90Mbyte, wouldn't > we?
If we used the textile formatter now, it would still be cached and only invoked once per message, not once for every logged in user every time a new message is received :) it would still be orders of magnitude more efficient. Still, I will keep the Textile formatter out for now. I find it hard to reuse its combinators, because I want to specify ESME-specific conditions in the evaluation (e.g. strong text, but one which does not include a valid hashtag or username inside). Using the alternative- invoking the Textile parser on each text element parsed already by our MsgParser has its problems, too- everything between tags, urls and usernames is separated in new paragraphs and I have to massage the result additionally to flatten the output. If the details don't make sense to you, the takeaway point is that I've removed the Textile parser because it complicates things, not because of memory consumption.
