I have a few old legacy projects on which I work occasionally from time to time, which i would like to integrate in a git workflow - mainly just for me, but it could be nice to have the option to collaborate on some level later.
The first problem is that the only method of accessing the project's code is (s)ftp-only. There is no shell-access given by the provider.
The second problem or contraint is that 3rd parties might change the files on the live ftp server (I know ugly) without using git at all, when i use some post-hook pushing method, this would break. Unfortunately, teaching those external users a git workflow is no option as ignorance is bliss.
What I really would like to do is to somehow keep track of those external changes (2) and benefit for my development workflow by using git.
Imagine I have not touched a project for a while and when i come back i'd like to see what has been done.
But I'm really confused on how and where I should setup repositories.
A super-nice to do Git deployment over FTP is git-ftp. Unfortunately, it's one way, i.e. it assumes the destination files are only ever updated by you (and via git-ftp).
So for your case you'd have to create some sort of synthetic history of remote changes when they occur, and the only way to do that is to get the files and then manually create "the delta" between them. So, for this, I'm with @VonC — you'd have to transfer the files back (rsync does this efficienly, if it's not available, google for an utility which is able to crawl the remote hierarchy using the FTP protocol itself and fetch only the missing/more recent files) and then make the changes be known to Git.
This process may be either done on a separate branch or on a temporary throw-away branch. Basically you fork a branch off a point you think represented the state of remote files last time you looked at them and then run a syncronization program over the checkout. (Note that such a program have to track removals, that is, it has to remove files locally which were removed at the remote side.) Then run git add --all . to make Git stage for commit all updated files, all removals and also add all the presently untracked files. Then record a commit. This commit will actually represent the changes made to the remote. You can then base your further work on this state.