Stop Thinking in Terms of Files

We have heard that smalltalk appears to use model similar to a LISP machine of yore in that the programming environment = the OS = the runtime environment. Once you define a function, it simply exists without being written to a file or compiled into some process that runs it. You can just call it, or undefine it and it ceases to be. From this point of view, it is probably perfectly valid to say that it has no files, because it doesn't need them.

On the other hand, let's assume that your smalltalk image got a little bit corrupted so that some packages/functions/whatever are now missing or not functioning. Or, let's say that you accidentally undefined a function and that was a mistake and now you really want to get it back. How would you do that? The file-based answer is that you hopefully had backups of the files that held definition of that function. What passes for a backup in smalltalk land?

And how do you deal with version control? How do you recover from mistakes? If you wanted to share your crap to someone else collaboratively through e.g. github, how would you do that? You'd have to export your functions into individual files, probably, and packages as directories into git. Someone would check them out, eval them, make changes, and commit updated functions. How does this kind of process look like, in a non-file paradigm, if it is done at all? (Does smalltalk VM even support networking?)

In general how do you even dump the state of the VM in some way that you can show someone what exactly your project is made of, in textual form? It's not very nice to dump an entire image and tell them to just run that. I bet the image is much larger and contains historical stuff that you no longer care about. What if you really just wanted to publish a recipe that can construct something equivalent of that particular image? What does "docker" for smalltalk look like?

If you tell us files suck, tell us also how you solve the same problems that we solve through files. Especially that collaborative programming through github use case interests me a little.