Does it matter much? If I hose my repo (which I don't think is that easy, since I've been doing this for years and never had an issue) then I can delete it and clone a new one from my local copy. Especially when it's just me, and I'm only pushing to the repo from one machine at a time.
It can hose your local, too. And it can happen easier than you think--I've seen it happen because a laptop that pushed to Dropbox went to sleep mid-sync and a desktop synced after. Fighting the Dropbox API to unwind it is a huge pain.
git-remote-dropbox works as you would expect a Git remote to work; it's API-driven and actively discourages even syncing the remote repository down to your machine. I would so, so strongly suggest you switch to it if you want to use Dropbox as a store.
Bare-git-repo-on-KBFS is inadvisable for a similar reason, which is why I'm so excited to see what they're doing here.
How would it hose my local? I thought git's design meant that it might possibly pull down new corrupted refs, but whatever I currently had would remain intact, so it's just a matter of reverting. Not so?
I believe it would be Dropbox doing the overwrite. Dropbox will just replace data - it doesn't do anything with respect to the reflog. I suppose it might be safer to work on a local copy and push to a second local copy in dropbox, so your working copy isn't touched by dropbox at all.
Yeah, keeping a local copy outside DB and pushing to a bare repo in DB is what I do. It didn't occur to me that one might work directly in a repository in DB. The hazards there, at least, are quite clear!
OK, so maybe we're using "local" for different things. Are you developing in your local copy of Dropbox, or are you cloning to a local directory using the Dropbox directory as a source (probably bare)? I assumed the former, which is what I meant by "local"; you can end up syncing multiple different instances of the repo and horking the contents of your .git directory (as well as cross-edited files, etc, that bleed changes onto multiple branches).
Both have the possibility of breaking because of concurrent or delayed syncs--like, which is actually HEAD?--but the latter is probably safer than the former. Or you can just use git-remote-dropbox and never have a problem.
If you always, always-always, develop on a single computer, Dropbox-as-normal-file-system can be fine. But if you have a desktop and a laptop, or multiple people partying on it, I get worried. :)
That explains the confusion, I'm talking about keeping a bare repository in Dropbox and cloning it to a non-Dropbox location on each computer where I work. It never occurred to me to keep the working copy itself on DB, that would be silly!
I expect that this could break the bare repository on DB if I ever pushed from two places simultaneously (where "simultaneously" could potentially encompass a period of hours or days if I pushed from an offline computer) but I should be able to repair it by recreating the bare repository.
Using something like git-remote-dropbox seems like a good idea. But at this point, I can just start using Keybase, hooray!
> ...keep the working copy itself on DB, that would be silly!
I don't think it's necessarily silly; it can be very useful in some scenarios.
I keep all my local working copies in a folder synced across several machines. I use Resilio Sync because it is better[1] than Dropbox for this purpose, but it's basically equivalent.
What this lets me do is stop working suddenly, at any moment (baby crying upstairs, or I lost track of time and have to bike to the office for a meeting) get up from my computer and move to another one (in another room in my house, or across town at my employer's office).
The code doesn't have to be in any finished state, needn't compile, I can literally be right in the middle of a line of code. As long as I've saved my work to disk, it will have synced before I reach the next computer, so I can sit down and resume work.
Before I had kids I didn't need this as much, so I just did git push/pull.
But then you have to do the work of pushing your half-finished junk to a different private repo, or rebasing to avoid polluting the git history with a bunch of crap commits just because you had to move, or not do that and just accept having a git history filled with crap.
Frankly I wish more of my work was capable of being distributed like this, but it's really only suitable for collections of plain files, which are amenable to being synced file-by-file. Luckily that includes almost all my programming work, however.
[1]: Resilio Sync is better than Dropbox for this because: it is much faster to sync than Dropbox, it supports symlinks so it doesn't corrupt your data when syncing folders containing them, and it syncs my data only among computers I control, not to any cloud service.
For sure--I'm going to go poke at the Keybase one this afternoon! (Also, to be clear, the Keybase method is essentially the same as git-remote-dropbox. Both set up git remote helpers.)
It took me about two seconds to create a new repository with Keybase and clone it to my computer, so I'm pretty impressed so far.
Thanks for the info about git-remote-dropbox and the potential failure modes of going without, even if they don't all apply to the way I've been doing things. It's still not ideal, so here's hoping Keybase makes it obsolete. If not, I'll keep git-remote-dropbox in mind.
