git for pushing local subdir to website

I made a post here about this here: http://stackoverflow.com/questions/600079/is-there-any-way-to-clone-a-git-repositorys-sub-directory-only/1223407#1223407

Basically, I'm working alone on a website (well, it's a facebook app.). What's the best way to sync my local copy with my production? but i want the production to only be linked with the production subdir of my local project.

so, I have a local copy i'm working on that I've organized like so. call the project example. here are the dirs on my local comp. macbook.

example -extra [this holds files i don't want to go to the webserver mainly for security reasons. files like .psd files or .sql files. or readme files. etc.] -example.com [this dir holds all the files i want on my server for production at http://example.com]

i have local git repo here: example/.git

now... how do i just push example.com and not extra and example.com?

do i need to make another git repo in example.com and make example.com a submodule?

i've also thought about using branches... a local branch and a production branch. but i only want one copy of the production files. because i want to be able to check out all the files and zip the dir and send it to someone to see.

that doesn't really make sense conceptually because example is one big project. extra is like extra files and example.com is like what i really want to go on web server...

thanks!

-matt

On Monday 03 August 2009 11:11:13 Matt Di Pasquale wrote:

[...]

There are several ways to do this with git, but since you are not going to make commits on the server itself, have you considered simply pushing the production files you want with e.g. rsync?

yes... i've thought of that. that's a good idea. i just love git and i've never really used rsync. how do i do that? i'll go google it.. i guess it would also be nice to use git though incase i decide i want a collaborator, right? how would u set it up? -matt

On Mon, Aug 3, 2009 at 2:01 PM, Wesley J. Landaker<wjl@icecavern.net> wrote:

On Monday 03 August 2009 12:09:26 Matt Di Pasquale wrote:

Rsync can do a lot of things, but the usual easy way to do it is to do something like this to send files via rsync over an ssh connection:

rsync -av /some_dir/ user@host:/some_dir/

This is essentially the same as using scp, but rsync saves a lot of network bandwidth by only sending the changes between the source and destination.

Setting things up to have a collaboration is kind of orthogonal to installing a production version of your app, but you could use the same server (and possibly the same git repository) for both things if you wanted to.

Here an example of one way to do it:

On the server you could have a normal git repository in, say, /home/me/myapp, and it always has the "production" branch checked out. So now the following directories are interesting:

/home/me/myapp -- repository branch w/ "production" checked out /home/me/myapp/.git -- the actual bare git repository /home/me/myapp/example.com/ -- the files that the web server should see

Now, you can use symlinks (or web-server configuration) to point the webserver to the right locations, for example:

http://example.com/myapp/ -- link to /home/me/myapp/example.com/ http://example.com/myapp.git/ -- link to /home/me/myapp/.git/

Now, collaborators can pull from the http://example.com/myapp.git and get the whole project, but access to http://example.com/myapp/ sees the example.com subdir as it's root, and everything works normally (assuming webserver configuration & permissions are correct).

Now you work on whatever branches you want, and push to the server whenever you want for collaboration using whatever branches you want. As far as the webserver is concerned, nothing changes when you do a push.

Then, when you are ready to change the website, you put your production changes on the "production" branch, push them, and make sure they are checked out on the server (git push doesn't automatically check out the files), e.g.:

client$ # just made changes in production branch and want them "live" client$ git push

server$ cd /home/me/myapp server$ git reset --hard

Personally, I might set something like this if I had a good reason, but otherwise, I'd host my git repository for collaboration separately, and just use rsync to install production files on the server. For one thing, that needs less prerequisites and setup on the server, so it's good if you don't fully control the server.

Either way, a "production" branch is a good idea so you know logically exactly what and when you released things live.

wow. thanks. that gave me some good ideas. i think u were right about rsync. it was pretty easy to set up. assuming the name of my app is myapp, i just put the following alias in my ~/.bash_profile on my local machine.

alias mapush='rsync -e ssh -av --exclude=".DS_Store" ~/Projects/myapp/myapp.com/ matt@mattdipasquale.com:projects/myapp.com/'

btw, is there a way to make aliases that are local to certain directories? like typing mapush works from any directory. what if i just wanted the alias to be push and only work from within ~/Projects/myapp? is that possible?

those aren't the actual paths i used. but u get the point. then i just type mapush and it does it.

and i just use a local git repository in ~/Projects/myapp/.git to track local changes since none are made on the server... so i don't need a git repo there like u said...

it's pretty fast. but will rsync be okay if i decide to move files around a bit? i know git is pretty good about that. and what if i delete certain files? will rsync delete them... like git does? guess i'll figure those questions out in time.

thanks!

-matt

On Mon, Aug 3, 2009 at 2:58 PM, Wesley J. Landaker<wjl@icecavern.net> wrote:

On Monday 03 August 2009 13:13:43 Matt Di Pasquale wrote:

Okay, looks like you've got rsync figured out!

The typical way to do this is to make a "bin" or "scripts" subdirectory of your project, then make a shell script and call it via ./bin/push or ./scripts/push or whatever. You can also do pretty much anything by using shell functions instead of aliases. They work just like shell aliases but are defined differently (man bash, then search for FUNCTIONS).

Since the topic is git, I should also mention that you can also set up repository hooks that make things happen when triggered, e.g. automatically rsync to the server when you commit to "production" for example. Do a man githooks for more info on that.

Rsync will not delete anything unless you add "--delete". If you use that, you probably also want "--delete-after" for a website, which makes sure that deletes are only done after everything else is copied over. The default is the same as "--delete-before", which does deletes before copying things (e.g. to save disk space).

As far as raw network transfer performance, git and rsync are probably pretty comparible, as they both are passing just changes instead of complete files around. Obviously they work very differently internally so e.g. I imagine that git will perform much better in some cases because it tracks content instead of files.