5 ms·
> Oh well... ;). Though it might seem like it, it doesn't give me any problems. I usually remember more-less how I named a file with a particular thought I need
by shoover 12y ago
> Oh well... ;). Though it might seem like it, it doesn't give me any problems. I usually remember more-less how I named a file with a particular thought I need, and if I can't, there's always grep to the rescue :).
Makes sense. ido-mode helps. Despite the fact that this is an anti-agenda thread, I should report that I also just discovered how easy it is to search across agenda files by word or regexp. It's like org aware grep. You would need some elisp to load all your .org files into org-agenda-files.
I've always used a handful of very large files, which probably doesn't have any particular advantage other than less buffer switching and facilitating isearch over grep. However, outline levels quickly get deep in large files.
> I'm thinking, maybe I'll eventually get around to spin up a Dropbox daemon on a VPS and make it sync only the wiki/ directory, and then put some simple "save snippet" CGI script to have a browser capture enpoint.
Sounds good.
- TeMPOraL 12y ago> ido-mode helps Especially with flx-ido (fuzzy matching for names) and ido-vertical-mode; now it's just fun to use :). > this is an anti-agenda thread I used to use agenda for doing GTD for over a year sometimes ago; it was awesome for aggregating calendars and TODO lists from multiple files, but I never got around to learning other interesting features. Maybe it's the time to try again :). > However, outline levels quickly get deep in large files. Yup. I still have a lot of deeply-nested long .org files - those are usually project-specific, and contain ideas, design notes, TODO lists, etc. But for personal wiki, it feels more natural for me to keep every topic in a separate file (mind you, some of them are quite long; my notes from "Pragmatic Thinking and Learning" are 750+ lines of 4-level deep nesting, and I'm nowhere near finished :)). I fear even thinking about coalescing them into a single file; when your file hits 10k lines (like one of my project notebooks at work), you find out you need to keep it fully expanded, because navigating with top headers folded gets kind of slow.